67 files changed, 2514 insertions, 2607 deletions

diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt
index dd4c29ed3..9b931976a 100644
--- a/src/video_core/CMakeLists.txt
+++ b/src/video_core/CMakeLists.txt
@@ -2,10 +2,8 @@ add_subdirectory(host_shaders)
 
 add_library(video_core STATIC
     buffer_cache/buffer_base.h
-    buffer_cache/buffer_block.h
+    buffer_cache/buffer_cache.cpp
     buffer_cache/buffer_cache.h
-    buffer_cache/map_interval.cpp
-    buffer_cache/map_interval.h
     cdma_pusher.cpp
     cdma_pusher.h
     command_classes/codecs/codec.cpp
@@ -152,8 +150,6 @@ add_library(video_core STATIC
     renderer_vulkan/vk_staging_buffer_pool.h
     renderer_vulkan/vk_state_tracker.cpp
     renderer_vulkan/vk_state_tracker.h
-    renderer_vulkan/vk_stream_buffer.cpp
-    renderer_vulkan/vk_stream_buffer.h
     renderer_vulkan/vk_swapchain.cpp
     renderer_vulkan/vk_swapchain.h
     renderer_vulkan/vk_texture_cache.cpp
diff --git a/src/video_core/buffer_cache/buffer_block.h b/src/video_core/buffer_cache/buffer_block.h
deleted file mode 100644
index e9306194a..000000000
--- a/src/video_core/buffer_cache/buffer_block.h
+++ /dev/null
@@ -1,62 +0,0 @@
-// Copyright 2019 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include "common/common_types.h"
-
-namespace VideoCommon {
-
-class BufferBlock {
-public:
-    [[nodiscard]] bool Overlaps(VAddr start, VAddr end) const {
-        return (cpu_addr < end) && (cpu_addr_end > start);
-    }
-
-    [[nodiscard]] bool IsInside(VAddr other_start, VAddr other_end) const {
-        return cpu_addr <= other_start && other_end <= cpu_addr_end;
-    }
-
-    [[nodiscard]] std::size_t Offset(VAddr in_addr) const {
-        return static_cast<std::size_t>(in_addr - cpu_addr);
-    }
-
-    [[nodiscard]] VAddr CpuAddr() const {
-        return cpu_addr;
-    }
-
-    [[nodiscard]] VAddr CpuAddrEnd() const {
-        return cpu_addr_end;
-    }
-
-    void SetCpuAddr(VAddr new_addr) {
-        cpu_addr = new_addr;
-        cpu_addr_end = new_addr + size;
-    }
-
-    [[nodiscard]] std::size_t Size() const {
-        return size;
-    }
-
-    [[nodiscard]] u64 Epoch() const {
-        return epoch;
-    }
-
-    void SetEpoch(u64 new_epoch) {
-        epoch = new_epoch;
-    }
-
-protected:
-    explicit BufferBlock(VAddr cpu_addr_, std::size_t size_) : size{size_} {
-        SetCpuAddr(cpu_addr_);
-    }
-
-private:
-    VAddr cpu_addr{};
-    VAddr cpu_addr_end{};
-    std::size_t size{};
-    u64 epoch{};
-};
-
-} // namespace VideoCommon
diff --git a/src/video_core/buffer_cache/buffer_cache.cpp b/src/video_core/buffer_cache/buffer_cache.cpp
new file mode 100644
index 000000000..ab32294c8
--- /dev/null
+++ b/src/video_core/buffer_cache/buffer_cache.cpp
@@ -0,0 +1,13 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/microprofile.h"
+
+namespace VideoCommon {
+
+MICROPROFILE_DEFINE(GPU_PrepareBuffers, "GPU", "Prepare buffers", MP_RGB(224, 128, 128));
+MICROPROFILE_DEFINE(GPU_BindUploadBuffers, "GPU", "Bind and upload buffers", MP_RGB(224, 128, 128));
+MICROPROFILE_DEFINE(GPU_DownloadMemory, "GPU", "Download buffers", MP_RGB(224, 128, 128));
+
+} // namespace VideoCommon
diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h
index 83b9ee871..e4f3c8e35 100644
--- a/src/video_core/buffer_cache/buffer_cache.h
+++ b/src/video_core/buffer_cache/buffer_cache.h
@@ -4,591 +4,1231 @@
 
 #pragma once
 
-#include <list>
+#include <algorithm>
+#include <array>
+#include <deque>
 #include <memory>
 #include <mutex>
+#include <span>
 #include <unordered_map>
-#include <unordered_set>
-#include <utility>
 #include <vector>
 
 #include <boost/container/small_vector.hpp>
-#include <boost/icl/interval_set.hpp>
-#include <boost/intrusive/set.hpp>
 
-#include "common/alignment.h"
-#include "common/assert.h"
 #include "common/common_types.h"
-#include "common/logging/log.h"
-#include "core/core.h"
+#include "common/div_ceil.h"
+#include "common/microprofile.h"
+#include "common/scope_exit.h"
 #include "core/memory.h"
 #include "core/settings.h"
-#include "video_core/buffer_cache/buffer_block.h"
-#include "video_core/buffer_cache/map_interval.h"
+#include "video_core/buffer_cache/buffer_base.h"
+#include "video_core/delayed_destruction_ring.h"
+#include "video_core/dirty_flags.h"
+#include "video_core/engines/kepler_compute.h"
+#include "video_core/engines/maxwell_3d.h"
 #include "video_core/memory_manager.h"
 #include "video_core/rasterizer_interface.h"
+#include "video_core/texture_cache/slot_vector.h"
+#include "video_core/texture_cache/types.h"
 
 namespace VideoCommon {
 
-template <typename Buffer, typename BufferType, typename StreamBuffer>
+MICROPROFILE_DECLARE(GPU_PrepareBuffers);
+MICROPROFILE_DECLARE(GPU_BindUploadBuffers);
+MICROPROFILE_DECLARE(GPU_DownloadMemory);
+
+using BufferId = SlotId;
+
+constexpr u32 NUM_VERTEX_BUFFERS = 32;
+constexpr u32 NUM_TRANSFORM_FEEDBACK_BUFFERS = 4;
+constexpr u32 NUM_GRAPHICS_UNIFORM_BUFFERS = 18;
+constexpr u32 NUM_COMPUTE_UNIFORM_BUFFERS = 8;
+constexpr u32 NUM_STORAGE_BUFFERS = 16;
+constexpr u32 NUM_STAGES = 5;
+
+template <typename P>
 class BufferCache {
-    using IntervalSet = boost::icl::interval_set<VAddr>;
-    using IntervalType = typename IntervalSet::interval_type;
-    using VectorMapInterval = boost::container::small_vector<MapInterval*, 1>;
+    // Page size for caching purposes.
+    // This is unrelated to the CPU page size and it can be changed as it seems optimal.
+    static constexpr u32 PAGE_BITS = 16;
+    static constexpr u64 PAGE_SIZE = u64{1} << PAGE_BITS;
 
-    static constexpr u64 WRITE_PAGE_BIT = 11;
-    static constexpr u64 BLOCK_PAGE_BITS = 21;
-    static constexpr u64 BLOCK_PAGE_SIZE = 1ULL << BLOCK_PAGE_BITS;
+    static constexpr bool IS_OPENGL = P::IS_OPENGL;
+    static constexpr bool HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS =
+        P::HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS;
+    static constexpr bool HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT =
+        P::HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT;
+    static constexpr bool NEEDS_BIND_UNIFORM_INDEX = P::NEEDS_BIND_UNIFORM_INDEX;
+    static constexpr bool NEEDS_BIND_STORAGE_INDEX = P::NEEDS_BIND_STORAGE_INDEX;
+    static constexpr bool USE_MEMORY_MAPS = P::USE_MEMORY_MAPS;
 
-public:
-    struct BufferInfo {
-        BufferType handle;
-        u64 offset;
-        u64 address;
+    static constexpr BufferId NULL_BUFFER_ID{0};
+
+    using Maxwell = Tegra::Engines::Maxwell3D::Regs;
+
+    using Runtime = typename P::Runtime;
+    using Buffer = typename P::Buffer;
+
+    struct Empty {};
+
+    struct Binding {
+        VAddr cpu_addr{};
+        u32 size{};
+        BufferId buffer_id;
     };
 
-    BufferInfo UploadMemory(GPUVAddr gpu_addr, std::size_t size, std::size_t alignment = 4,
-                            bool is_written = false, bool use_fast_cbuf = false) {
-        std::lock_guard lock{mutex};
+    static constexpr Binding NULL_BINDING{
+        .cpu_addr = 0,
+        .size = 0,
+        .buffer_id = NULL_BUFFER_ID,
+    };
 
-        const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr);
-        if (!cpu_addr) {
-            return GetEmptyBuffer(size);
-        }
+public:
+    static constexpr size_t SKIP_CACHE_SIZE = 4096;
 
-        // Cache management is a big overhead, so only cache entries with a given size.
-        // TODO: Figure out which size is the best for given games.
-        constexpr std::size_t max_stream_size = 0x800;
-        if (use_fast_cbuf || size < max_stream_size) {
-            if (!is_written && !IsRegionWritten(*cpu_addr, *cpu_addr + size - 1)) {
-                const bool is_granular = gpu_memory.IsGranularRange(gpu_addr, size);
-                if (use_fast_cbuf) {
-                    u8* dest;
-                    if (is_granular) {
-                        dest = gpu_memory.GetPointer(gpu_addr);
-                    } else {
-                        staging_buffer.resize(size);
-                        dest = staging_buffer.data();
-                        gpu_memory.ReadBlockUnsafe(gpu_addr, dest, size);
-                    }
-                    return ConstBufferUpload(dest, size);
-                }
-                if (is_granular) {
-                    u8* const host_ptr = gpu_memory.GetPointer(gpu_addr);
-                    return StreamBufferUpload(size, alignment, [host_ptr, size](u8* dest) {
-                        std::memcpy(dest, host_ptr, size);
-                    });
-                } else {
-                    return StreamBufferUpload(size, alignment, [this, gpu_addr, size](u8* dest) {
-                        gpu_memory.ReadBlockUnsafe(gpu_addr, dest, size);
-                    });
-                }
-            }
-        }
+    explicit BufferCache(VideoCore::RasterizerInterface& rasterizer_,
+                         Tegra::Engines::Maxwell3D& maxwell3d_,
+                         Tegra::Engines::KeplerCompute& kepler_compute_,
+                         Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_,
+                         Runtime& runtime_);
 
-        Buffer* const block = GetBlock(*cpu_addr, size);
-        MapInterval* const map = MapAddress(block, gpu_addr, *cpu_addr, size);
-        if (!map) {
-            return GetEmptyBuffer(size);
-        }
-        if (is_written) {
-            map->MarkAsModified(true, GetModifiedTicks());
-            if (Settings::IsGPULevelHigh() &&
-                Settings::values.use_asynchronous_gpu_emulation.GetValue()) {
-                MarkForAsyncFlush(map);
-            }
-            if (!map->is_written) {
-                map->is_written = true;
-                MarkRegionAsWritten(map->start, map->end - 1);
-            }
-        }
+    void TickFrame();
 
-        return BufferInfo{block->Handle(), block->Offset(*cpu_addr), block->Address()};
-    }
+    void WriteMemory(VAddr cpu_addr, u64 size);
 
-    /// Uploads from a host memory. Returns the OpenGL buffer where it's located and its offset.
-    BufferInfo UploadHostMemory(const void* raw_pointer, std::size_t size,
-                                std::size_t alignment = 4) {
-        std::lock_guard lock{mutex};
-        return StreamBufferUpload(size, alignment, [raw_pointer, size](u8* dest) {
-            std::memcpy(dest, raw_pointer, size);
-        });
-    }
+    void CachedWriteMemory(VAddr cpu_addr, u64 size);
 
-    /// Prepares the buffer cache for data uploading
-    /// @param max_size Maximum number of bytes that will be uploaded
-    /// @return True when a stream buffer invalidation was required, false otherwise
-    void Map(std::size_t max_size) {
-        std::lock_guard lock{mutex};
+    void DownloadMemory(VAddr cpu_addr, u64 size);
 
-        std::tie(buffer_ptr, buffer_offset_base) = stream_buffer.Map(max_size, 4);
-        buffer_offset = buffer_offset_base;
-    }
+    void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size);
 
-    /// Finishes the upload stream
-    void Unmap() {
-        std::lock_guard lock{mutex};
-        stream_buffer.Unmap(buffer_offset - buffer_offset_base);
-    }
+    void UpdateGraphicsBuffers(bool is_indexed);
 
-    /// Function called at the end of each frame, inteded for deferred operations
-    void TickFrame() {
-        ++epoch;
+    void UpdateComputeBuffers();
 
-        while (!pending_destruction.empty()) {
-            // Delay at least 4 frames before destruction.
-            // This is due to triple buffering happening on some drivers.
-            static constexpr u64 epochs_to_destroy = 5;
-            if (pending_destruction.front()->Epoch() + epochs_to_destroy > epoch) {
-                break;
-            }
-            pending_destruction.pop();
-        }
-    }
+    void BindHostGeometryBuffers(bool is_indexed);
 
-    /// Write any cached resources overlapping the specified region back to memory
-    void FlushRegion(VAddr addr, std::size_t size) {
-        std::lock_guard lock{mutex};
+    void BindHostStageBuffers(size_t stage);
 
-        VectorMapInterval objects = GetMapsInRange(addr, size);
-        std::sort(objects.begin(), objects.end(),
-                  [](MapInterval* lhs, MapInterval* rhs) { return lhs->ticks < rhs->ticks; });
-        for (MapInterval* object : objects) {
-            if (object->is_modified && object->is_registered) {
-                mutex.unlock();
-                FlushMap(object);
-                mutex.lock();
-            }
-        }
-    }
+    void BindHostComputeBuffers();
 
-    bool MustFlushRegion(VAddr addr, std::size_t size) {
-        std::lock_guard lock{mutex};
+    void SetEnabledUniformBuffers(size_t stage, u32 enabled);
 
-        const VectorMapInterval objects = GetMapsInRange(addr, size);
-        return std::any_of(objects.cbegin(), objects.cend(), [](const MapInterval* map) {
-            return map->is_modified && map->is_registered;
-        });
-    }
+    void SetEnabledComputeUniformBuffers(u32 enabled);
 
-    /// Mark the specified region as being invalidated
-    void InvalidateRegion(VAddr addr, u64 size) {
-        std::lock_guard lock{mutex};
+    void UnbindGraphicsStorageBuffers(size_t stage);
 
-        for (auto& object : GetMapsInRange(addr, size)) {
-            if (object->is_registered) {
-                Unregister(object);
-            }
-        }
-    }
+    void BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset,
+                                   bool is_written);
 
-    void OnCPUWrite(VAddr addr, std::size_t size) {
-        std::lock_guard lock{mutex};
+    void UnbindComputeStorageBuffers();
 
-        for (MapInterval* object : GetMapsInRange(addr, size)) {
-            if (object->is_memory_marked && object->is_registered) {
-                UnmarkMemory(object);
-                object->is_sync_pending = true;
-                marked_for_unregister.emplace_back(object);
-            }
-        }
-    }
+    void BindComputeStorageBuffer(size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset,
+                                  bool is_written);
 
-    void SyncGuestHost() {
-        std::lock_guard lock{mutex};
+    void FlushCachedWrites();
 
-        for (auto& object : marked_for_unregister) {
-            if (object->is_registered) {
-                object->is_sync_pending = false;
-                Unregister(object);
-            }
+    /// Return true when there are uncommitted buffers to be downloaded
+    [[nodiscard]] bool HasUncommittedFlushes() const noexcept;
+
+    /// Return true when the caller should wait for async downloads
+    [[nodiscard]] bool ShouldWaitAsyncFlushes() const noexcept;
+
+    /// Commit asynchronous downloads
+    void CommitAsyncFlushes();
+
+    /// Pop asynchronous downloads
+    void PopAsyncFlushes();
+
+    /// Return true when a CPU region is modified from the GPU
+    [[nodiscard]] bool IsRegionGpuModified(VAddr addr, size_t size);
+
+    std::mutex mutex;
+
+private:
+    template <typename Func>
+    static void ForEachEnabledBit(u32 enabled_mask, Func&& func) {
+        for (u32 index = 0; enabled_mask != 0; ++index, enabled_mask >>= 1) {
+            const int disabled_bits = std::countr_zero(enabled_mask);
+            index += disabled_bits;
+            enabled_mask >>= disabled_bits;
+            func(index);
         }
-        marked_for_unregister.clear();
     }
 
-    void CommitAsyncFlushes() {
-        if (uncommitted_flushes) {
-            auto commit_list = std::make_shared<std::list<MapInterval*>>();
-            for (MapInterval* map : *uncommitted_flushes) {
-                if (map->is_registered && map->is_modified) {
-                    // TODO(Blinkhawk): Implement backend asynchronous flushing
-                    // AsyncFlushMap(map)
-                    commit_list->push_back(map);
-                }
-            }
-            if (!commit_list->empty()) {
-                committed_flushes.push_back(commit_list);
-            } else {
-                committed_flushes.emplace_back();
+    template <typename Func>
+    void ForEachBufferInRange(VAddr cpu_addr, u64 size, Func&& func) {
+        const u64 page_end = Common::DivCeil(cpu_addr + size, PAGE_SIZE);
+        for (u64 page = cpu_addr >> PAGE_BITS; page < page_end;) {
+            const BufferId buffer_id = page_table[page];
+            if (!buffer_id) {
+                ++page;
+                continue;
             }
-        } else {
-            committed_flushes.emplace_back();
+            Buffer& buffer = slot_buffers[buffer_id];
+            func(buffer_id, buffer);
+
+            const VAddr end_addr = buffer.CpuAddr() + buffer.SizeBytes();
+            page = Common::DivCeil(end_addr, PAGE_SIZE);
         }
-        uncommitted_flushes.reset();
     }
 
-    bool ShouldWaitAsyncFlushes() const {
-        return !committed_flushes.empty() && committed_flushes.front() != nullptr;
+    static bool IsRangeGranular(VAddr cpu_addr, size_t size) {
+        return (cpu_addr & ~Core::Memory::PAGE_MASK) ==
+               ((cpu_addr + size) & ~Core::Memory::PAGE_MASK);
     }
 
-    bool HasUncommittedFlushes() const {
-        return uncommitted_flushes != nullptr;
-    }
+    void BindHostIndexBuffer();
 
-    void PopAsyncFlushes() {
-        if (committed_flushes.empty()) {
-            return;
-        }
-        auto& flush_list = committed_flushes.front();
-        if (!flush_list) {
-            committed_flushes.pop_front();
-            return;
-        }
-        for (MapInterval* map : *flush_list) {
-            if (map->is_registered) {
-                // TODO(Blinkhawk): Replace this for reading the asynchronous flush
-                FlushMap(map);
-            }
-        }
-        committed_flushes.pop_front();
-    }
+    void BindHostVertexBuffers();
 
-    virtual BufferInfo GetEmptyBuffer(std::size_t size) = 0;
+    void BindHostGraphicsUniformBuffers(size_t stage);
 
-protected:
-    explicit BufferCache(VideoCore::RasterizerInterface& rasterizer_,
-                         Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_,
-                         StreamBuffer& stream_buffer_)
-        : rasterizer{rasterizer_}, gpu_memory{gpu_memory_}, cpu_memory{cpu_memory_},
-          stream_buffer{stream_buffer_} {}
+    void BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 binding_index, bool needs_bind);
 
-    ~BufferCache() = default;
+    void BindHostGraphicsStorageBuffers(size_t stage);
 
-    virtual std::shared_ptr<Buffer> CreateBlock(VAddr cpu_addr, std::size_t size) = 0;
+    void BindHostTransformFeedbackBuffers();
 
-    virtual BufferInfo ConstBufferUpload(const void* raw_pointer, std::size_t size) {
-        return {};
-    }
+    void BindHostComputeUniformBuffers();
 
-    /// Register an object into the cache
-    MapInterval* Register(MapInterval new_map, bool inherit_written = false) {
-        const VAddr cpu_addr = new_map.start;
-        if (!cpu_addr) {
-            LOG_CRITICAL(HW_GPU, "Failed to register buffer with unmapped gpu_address 0x{:016x}",
-                         new_map.gpu_addr);
-            return nullptr;
-        }
-        const std::size_t size = new_map.end - new_map.start;
-        new_map.is_registered = true;
-        rasterizer.UpdatePagesCachedCount(cpu_addr, size, 1);
-        new_map.is_memory_marked = true;
-        if (inherit_written) {
-            MarkRegionAsWritten(new_map.start, new_map.end - 1);
-            new_map.is_written = true;
-        }
-        MapInterval* const storage = mapped_addresses_allocator.Allocate();
-        *storage = new_map;
-        mapped_addresses.insert(*storage);
-        return storage;
-    }
+    void BindHostComputeStorageBuffers();
 
-    void UnmarkMemory(MapInterval* map) {
-        if (!map->is_memory_marked) {
-            return;
-        }
-        const std::size_t size = map->end - map->start;
-        rasterizer.UpdatePagesCachedCount(map->start, size, -1);
-        map->is_memory_marked = false;
-    }
-
-    /// Unregisters an object from the cache
-    void Unregister(MapInterval* map) {
-        UnmarkMemory(map);
-        map->is_registered = false;
-        if (map->is_sync_pending) {
-            map->is_sync_pending = false;
-            marked_for_unregister.remove(map);
+    void DoUpdateGraphicsBuffers(bool is_indexed);
+
+    void DoUpdateComputeBuffers();
+
+    void UpdateIndexBuffer();
+
+    void UpdateVertexBuffers();
+
+    void UpdateVertexBuffer(u32 index);
+
+    void UpdateUniformBuffers(size_t stage);
+
+    void UpdateStorageBuffers(size_t stage);
+
+    void UpdateTransformFeedbackBuffers();
+
+    void UpdateTransformFeedbackBuffer(u32 index);
+
+    void UpdateComputeUniformBuffers();
+
+    void UpdateComputeStorageBuffers();
+
+    void MarkWrittenBuffer(BufferId buffer_id, VAddr cpu_addr, u32 size);
+
+    [[nodiscard]] BufferId FindBuffer(VAddr cpu_addr, u32 size);
+
+    [[nodiscard]] BufferId CreateBuffer(VAddr cpu_addr, u32 wanted_size);
+
+    void Register(BufferId buffer_id);
+
+    void Unregister(BufferId buffer_id);
+
+    template <bool insert>
+    void ChangeRegister(BufferId buffer_id);
+
+    void SynchronizeBuffer(Buffer& buffer, VAddr cpu_addr, u32 size);
+
+    void SynchronizeBufferImpl(Buffer& buffer, VAddr cpu_addr, u32 size);
+
+    void UploadMemory(Buffer& buffer, u64 total_size_bytes, u64 largest_copy,
+                      std::span<BufferCopy> copies);
+
+    void ImmediateUploadMemory(Buffer& buffer, u64 largest_copy,
+                               std::span<const BufferCopy> copies);
+
+    void MappedUploadMemory(Buffer& buffer, u64 total_size_bytes,
+                            std::span<const BufferCopy> copies);
+
+    void DeleteBuffer(BufferId buffer_id);
+
+    void ReplaceBufferDownloads(BufferId old_buffer_id, BufferId new_buffer_id);
+
+    void NotifyBufferDeletion();
+
+    [[nodiscard]] Binding StorageBufferBinding(GPUVAddr ssbo_addr) const;
+
+    [[nodiscard]] std::span<const u8> ImmediateBufferWithData(VAddr cpu_addr, size_t size);
+
+    [[nodiscard]] std::span<u8> ImmediateBuffer(size_t wanted_capacity);
+
+    [[nodiscard]] bool HasFastUniformBufferBound(size_t stage, u32 binding_index) const noexcept;
+
+    VideoCore::RasterizerInterface& rasterizer;
+    Tegra::Engines::Maxwell3D& maxwell3d;
+    Tegra::Engines::KeplerCompute& kepler_compute;
+    Tegra::MemoryManager& gpu_memory;
+    Core::Memory::Memory& cpu_memory;
+    Runtime& runtime;
+
+    SlotVector<Buffer> slot_buffers;
+    DelayedDestructionRing<Buffer, 8> delayed_destruction_ring;
+
+    u32 last_index_count = 0;
+
+    Binding index_buffer;
+    std::array<Binding, NUM_VERTEX_BUFFERS> vertex_buffers;
+    std::array<std::array<Binding, NUM_GRAPHICS_UNIFORM_BUFFERS>, NUM_STAGES> uniform_buffers;
+    std::array<std::array<Binding, NUM_STORAGE_BUFFERS>, NUM_STAGES> storage_buffers;
+    std::array<Binding, NUM_TRANSFORM_FEEDBACK_BUFFERS> transform_feedback_buffers;
+
+    std::array<Binding, NUM_COMPUTE_UNIFORM_BUFFERS> compute_uniform_buffers;
+    std::array<Binding, NUM_STORAGE_BUFFERS> compute_storage_buffers;
+
+    std::array<u32, NUM_STAGES> enabled_uniform_buffers{};
+    u32 enabled_compute_uniform_buffers = 0;
+
+    std::array<u32, NUM_STAGES> enabled_storage_buffers{};
+    std::array<u32, NUM_STAGES> written_storage_buffers{};
+    u32 enabled_compute_storage_buffers = 0;
+    u32 written_compute_storage_buffers = 0;
+
+    std::array<u32, NUM_STAGES> fast_bound_uniform_buffers{};
+
+    bool has_deleted_buffers = false;
+
+    std::conditional_t<HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS, std::array<u32, NUM_STAGES>, Empty>
+        dirty_uniform_buffers{};
+
+    std::vector<BufferId> cached_write_buffer_ids;
+
+    // TODO: This data structure is not optimal and it should be reworked
+    std::vector<BufferId> uncommitted_downloads;
+    std::deque<std::vector<BufferId>> committed_downloads;
+
+    size_t immediate_buffer_capacity = 0;
+    std::unique_ptr<u8[]> immediate_buffer_alloc;
+
+    std::array<BufferId, ((1ULL << 39) >> PAGE_BITS)> page_table;
+};
+
+template <class P>
+BufferCache<P>::BufferCache(VideoCore::RasterizerInterface& rasterizer_,
+                            Tegra::Engines::Maxwell3D& maxwell3d_,
+                            Tegra::Engines::KeplerCompute& kepler_compute_,
+                            Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_,
+                            Runtime& runtime_)
+    : rasterizer{rasterizer_}, maxwell3d{maxwell3d_}, kepler_compute{kepler_compute_},
+      gpu_memory{gpu_memory_}, cpu_memory{cpu_memory_}, runtime{runtime_} {
+    // Ensure the first slot is used for the null buffer
+    void(slot_buffers.insert(runtime, NullBufferParams{}));
+}
+
+template <class P>
+void BufferCache<P>::TickFrame() {
+    delayed_destruction_ring.Tick();
+}
+
+template <class P>
+void BufferCache<P>::WriteMemory(VAddr cpu_addr, u64 size) {
+    ForEachBufferInRange(cpu_addr, size, [&](BufferId, Buffer& buffer) {
+        buffer.MarkRegionAsCpuModified(cpu_addr, size);
+    });
+}
+
+template <class P>
+void BufferCache<P>::CachedWriteMemory(VAddr cpu_addr, u64 size) {
+    ForEachBufferInRange(cpu_addr, size, [&](BufferId buffer_id, Buffer& buffer) {
+        if (!buffer.HasCachedWrites()) {
+            cached_write_buffer_ids.push_back(buffer_id);
         }
-        if (map->is_written) {
-            UnmarkRegionAsWritten(map->start, map->end - 1);
+        buffer.CachedCpuWrite(cpu_addr, size);
+    });
+}
+
+template <class P>
+void BufferCache<P>::DownloadMemory(VAddr cpu_addr, u64 size) {
+    ForEachBufferInRange(cpu_addr, size, [&](BufferId, Buffer& buffer) {
+        boost::container::small_vector<BufferCopy, 1> copies;
+        u64 total_size_bytes = 0;
+        u64 largest_copy = 0;
+        buffer.ForEachDownloadRange(cpu_addr, size, [&](u64 range_offset, u64 range_size) {
+            copies.push_back(BufferCopy{
+                .src_offset = range_offset,
+                .dst_offset = total_size_bytes,
+                .size = range_size,
+            });
+            total_size_bytes += range_size;
+            largest_copy = std::max(largest_copy, range_size);
+        });
+        if (total_size_bytes == 0) {
+            return;
         }
-        const auto it = mapped_addresses.find(*map);
-        ASSERT(it != mapped_addresses.end());
-        mapped_addresses.erase(it);
-        mapped_addresses_allocator.Release(map);
-    }
-
-private:
-    MapInterval* MapAddress(Buffer* block, GPUVAddr gpu_addr, VAddr cpu_addr, std::size_t size) {
-        const VectorMapInterval overlaps = GetMapsInRange(cpu_addr, size);
-        if (overlaps.empty()) {
-            const VAddr cpu_addr_end = cpu_addr + size;
-            if (gpu_memory.IsGranularRange(gpu_addr, size)) {
-                u8* const host_ptr = gpu_memory.GetPointer(gpu_addr);
-                block->Upload(block->Offset(cpu_addr), size, host_ptr);
-            } else {
-                staging_buffer.resize(size);
-                gpu_memory.ReadBlockUnsafe(gpu_addr, staging_buffer.data(), size);
-                block->Upload(block->Offset(cpu_addr), size, staging_buffer.data());
+        MICROPROFILE_SCOPE(GPU_DownloadMemory);
+
+        if constexpr (USE_MEMORY_MAPS) {
+            auto download_staging = runtime.DownloadStagingBuffer(total_size_bytes);
+            const u8* const mapped_memory = download_staging.mapped_span.data();
+            const std::span<BufferCopy> copies_span(copies.data(), copies.data() + copies.size());
+            runtime.CopyBuffer(download_staging.buffer, buffer, copies_span);
+            runtime.Finish();
+            for (const BufferCopy& copy : copies) {
+                const VAddr copy_cpu_addr = buffer.CpuAddr() + copy.src_offset;
+                const u8* copy_mapped_memory = mapped_memory + copy.dst_offset;
+                cpu_memory.WriteBlockUnsafe(copy_cpu_addr, copy_mapped_memory, copy.size);
             }
-            return Register(MapInterval(cpu_addr, cpu_addr_end, gpu_addr));
-        }
-
-        const VAddr cpu_addr_end = cpu_addr + size;
-        if (overlaps.size() == 1) {
-            MapInterval* const current_map = overlaps[0];
-            if (current_map->IsInside(cpu_addr, cpu_addr_end)) {
-                return current_map;
+        } else {
+            const std::span<u8> immediate_buffer = ImmediateBuffer(largest_copy);
+            for (const BufferCopy& copy : copies) {
+                buffer.ImmediateDownload(copy.src_offset, immediate_buffer.subspan(0, copy.size));
+                const VAddr copy_cpu_addr = buffer.CpuAddr() + copy.src_offset;
+                cpu_memory.WriteBlockUnsafe(copy_cpu_addr, immediate_buffer.data(), copy.size);
             }
         }
-        VAddr new_start = cpu_addr;
-        VAddr new_end = cpu_addr_end;
-        bool write_inheritance = false;
-        bool modified_inheritance = false;
-        // Calculate new buffer parameters
-        for (MapInterval* overlap : overlaps) {
-            new_start = std::min(overlap->start, new_start);
-            new_end = std::max(overlap->end, new_end);
-            write_inheritance |= overlap->is_written;
-            modified_inheritance |= overlap->is_modified;
+    });
+}
+
+template <class P>
+void BufferCache<P>::BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr,
+                                               u32 size) {
+    const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr);
+    if (!cpu_addr) {
+        uniform_buffers[stage][index] = NULL_BINDING;
+        return;
+    }
+    const Binding binding{
+        .cpu_addr = *cpu_addr,
+        .size = size,
+        .buffer_id = BufferId{},
+    };
+    uniform_buffers[stage][index] = binding;
+}
+
+template <class P>
+void BufferCache<P>::UpdateGraphicsBuffers(bool is_indexed) {
+    MICROPROFILE_SCOPE(GPU_PrepareBuffers);
+    do {
+        has_deleted_buffers = false;
+        DoUpdateGraphicsBuffers(is_indexed);
+    } while (has_deleted_buffers);
+}
+
+template <class P>
+void BufferCache<P>::UpdateComputeBuffers() {
+    MICROPROFILE_SCOPE(GPU_PrepareBuffers);
+    do {
+        has_deleted_buffers = false;
+        DoUpdateComputeBuffers();
+    } while (has_deleted_buffers);
+}
+
+template <class P>
+void BufferCache<P>::BindHostGeometryBuffers(bool is_indexed) {
+    MICROPROFILE_SCOPE(GPU_BindUploadBuffers);
+    if (is_indexed) {
+        BindHostIndexBuffer();
+    } else if constexpr (!HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT) {
+        const auto& regs = maxwell3d.regs;
+        if (regs.draw.topology == Maxwell::PrimitiveTopology::Quads) {
+            runtime.BindQuadArrayIndexBuffer(regs.vertex_buffer.first, regs.vertex_buffer.count);
         }
-        GPUVAddr new_gpu_addr = gpu_addr + new_start - cpu_addr;
-        for (auto& overlap : overlaps) {
-            Unregister(overlap);
+    }
+    BindHostVertexBuffers();
+    BindHostTransformFeedbackBuffers();
+}
+
+template <class P>
+void BufferCache<P>::BindHostStageBuffers(size_t stage) {
+    MICROPROFILE_SCOPE(GPU_BindUploadBuffers);
+    BindHostGraphicsUniformBuffers(stage);
+    BindHostGraphicsStorageBuffers(stage);
+}
+
+template <class P>
+void BufferCache<P>::BindHostComputeBuffers() {
+    MICROPROFILE_SCOPE(GPU_BindUploadBuffers);
+    BindHostComputeUniformBuffers();
+    BindHostComputeStorageBuffers();
+}
+
+template <class P>
+void BufferCache<P>::SetEnabledUniformBuffers(size_t stage, u32 enabled) {
+    if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) {
+        if (enabled_uniform_buffers[stage] != enabled) {
+            dirty_uniform_buffers[stage] = ~u32{0};
         }
-        UpdateBlock(block, new_start, new_end, overlaps);
-
-        const MapInterval new_map{new_start, new_end, new_gpu_addr};
-        MapInterval* const map = Register(new_map, write_inheritance);
-        if (!map) {
-            return nullptr;
+    }
+    enabled_uniform_buffers[stage] = enabled;
+}
+
+template <class P>
+void BufferCache<P>::SetEnabledComputeUniformBuffers(u32 enabled) {
+    enabled_compute_uniform_buffers = enabled;
+}
+
+template <class P>
+void BufferCache<P>::UnbindGraphicsStorageBuffers(size_t stage) {
+    enabled_storage_buffers[stage] = 0;
+    written_storage_buffers[stage] = 0;
+}
+
+template <class P>
+void BufferCache<P>::BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, u32 cbuf_index,
+                                               u32 cbuf_offset, bool is_written) {
+    enabled_storage_buffers[stage] |= 1U << ssbo_index;
+    written_storage_buffers[stage] |= (is_written ? 1U : 0U) << ssbo_index;
+
+    const auto& cbufs = maxwell3d.state.shader_stages[stage];
+    const GPUVAddr ssbo_addr = cbufs.const_buffers[cbuf_index].address + cbuf_offset;
+    storage_buffers[stage][ssbo_index] = StorageBufferBinding(ssbo_addr);
+}
+
+template <class P>
+void BufferCache<P>::UnbindComputeStorageBuffers() {
+    enabled_compute_storage_buffers = 0;
+    written_compute_storage_buffers = 0;
+}
+
+template <class P>
+void BufferCache<P>::BindComputeStorageBuffer(size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset,
+                                              bool is_written) {
+    enabled_compute_storage_buffers |= 1U << ssbo_index;
+    written_compute_storage_buffers |= (is_written ? 1U : 0U) << ssbo_index;
+
+    const auto& launch_desc = kepler_compute.launch_description;
+    ASSERT(((launch_desc.const_buffer_enable_mask >> cbuf_index) & 1) != 0);
+
+    const auto& cbufs = launch_desc.const_buffer_config;
+    const GPUVAddr ssbo_addr = cbufs[cbuf_index].Address() + cbuf_offset;
+    compute_storage_buffers[ssbo_index] = StorageBufferBinding(ssbo_addr);
+}
+
+template <class P>
+void BufferCache<P>::FlushCachedWrites() {
+    for (const BufferId buffer_id : cached_write_buffer_ids) {
+        slot_buffers[buffer_id].FlushCachedWrites();
+    }
+    cached_write_buffer_ids.clear();
+}
+
+template <class P>
+bool BufferCache<P>::HasUncommittedFlushes() const noexcept {
+    return !uncommitted_downloads.empty();
+}
+
+template <class P>
+bool BufferCache<P>::ShouldWaitAsyncFlushes() const noexcept {
+    return !committed_downloads.empty() && !committed_downloads.front().empty();
+}
+
+template <class P>
+void BufferCache<P>::CommitAsyncFlushes() {
+    // This is intentionally passing the value by copy
+    committed_downloads.push_front(uncommitted_downloads);
+    uncommitted_downloads.clear();
+}
+
+template <class P>
+void BufferCache<P>::PopAsyncFlushes() {
+    if (committed_downloads.empty()) {
+        return;
+    }
+    auto scope_exit_pop_download = detail::ScopeExit([this] { committed_downloads.pop_back(); });
+    const std::span<const BufferId> download_ids = committed_downloads.back();
+    if (download_ids.empty()) {
+        return;
+    }
+    MICROPROFILE_SCOPE(GPU_DownloadMemory);
+
+    boost::container::small_vector<std::pair<BufferCopy, BufferId>, 1> downloads;
+    u64 total_size_bytes = 0;
+    u64 largest_copy = 0;
+    for (const BufferId buffer_id : download_ids) {
+        slot_buffers[buffer_id].ForEachDownloadRange([&](u64 range_offset, u64 range_size) {
+            downloads.push_back({
+                BufferCopy{
+                    .src_offset = range_offset,
+                    .dst_offset = total_size_bytes,
+                    .size = range_size,
+                },
+                buffer_id,
+            });
+            total_size_bytes += range_size;
+            largest_copy = std::max(largest_copy, range_size);
+        });
+    }
+    if (downloads.empty()) {
+        return;
+    }
+    if constexpr (USE_MEMORY_MAPS) {
+        auto download_staging = runtime.DownloadStagingBuffer(total_size_bytes);
+        for (const auto [copy, buffer_id] : downloads) {
+            const std::array copies{copy};
+            runtime.CopyBuffer(download_staging.buffer, slot_buffers[buffer_id], copies);
         }
-        if (modified_inheritance) {
-            map->MarkAsModified(true, GetModifiedTicks());
-            if (Settings::IsGPULevelHigh() &&
-                Settings::values.use_asynchronous_gpu_emulation.GetValue()) {
-                MarkForAsyncFlush(map);
-            }
+        runtime.Finish();
+        for (const auto [copy, buffer_id] : downloads) {
+            const Buffer& buffer = slot_buffers[buffer_id];
+            const VAddr cpu_addr = buffer.CpuAddr() + copy.src_offset;
+            const u8* read_mapped_memory = download_staging.mapped_span.data() + copy.dst_offset;
+            cpu_memory.WriteBlockUnsafe(cpu_addr, read_mapped_memory, copy.size);
+        }
+    } else {
+        const std::span<u8> immediate_buffer = ImmediateBuffer(largest_copy);
+        for (const auto [copy, buffer_id] : downloads) {
+            Buffer& buffer = slot_buffers[buffer_id];
+            buffer.ImmediateDownload(copy.src_offset, immediate_buffer.subspan(0, copy.size));
+            const VAddr cpu_addr = buffer.CpuAddr() + copy.src_offset;
+            cpu_memory.WriteBlockUnsafe(cpu_addr, immediate_buffer.data(), copy.size);
         }
-        return map;
     }
-
-    void UpdateBlock(Buffer* block, VAddr start, VAddr end, const VectorMapInterval& overlaps) {
-        const IntervalType base_interval{start, end};
-        IntervalSet interval_set{};
-        interval_set.add(base_interval);
-        for (auto& overlap : overlaps) {
-            const IntervalType subtract{overlap->start, overlap->end};
-            interval_set.subtract(subtract);
+}
+
+template <class P>
+bool BufferCache<P>::IsRegionGpuModified(VAddr addr, size_t size) {
+    const u64 page_end = Common::DivCeil(addr + size, PAGE_SIZE);
+    for (u64 page = addr >> PAGE_BITS; page < page_end;) {
+        const BufferId image_id = page_table[page];
+        if (!image_id) {
+            ++page;
+            continue;
         }
-        for (auto& interval : interval_set) {
-            const std::size_t size = interval.upper() - interval.lower();
-            if (size == 0) {
-                continue;
-            }
-            staging_buffer.resize(size);
-            cpu_memory.ReadBlockUnsafe(interval.lower(), staging_buffer.data(), size);
-            block->Upload(block->Offset(interval.lower()), size, staging_buffer.data());
+        Buffer& buffer = slot_buffers[image_id];
+        if (buffer.IsRegionGpuModified(addr, size)) {
+            return true;
         }
+        const VAddr end_addr = buffer.CpuAddr() + buffer.SizeBytes();
+        page = Common::DivCeil(end_addr, PAGE_SIZE);
     }
-
-    VectorMapInterval GetMapsInRange(VAddr addr, std::size_t size) {
-        VectorMapInterval result;
-        if (size == 0) {
-            return result;
+    return false;
+}
+
+template <class P>
+void BufferCache<P>::BindHostIndexBuffer() {
+    Buffer& buffer = slot_buffers[index_buffer.buffer_id];
+    const u32 offset = buffer.Offset(index_buffer.cpu_addr);
+    const u32 size = index_buffer.size;
+    SynchronizeBuffer(buffer, index_buffer.cpu_addr, size);
+    if constexpr (HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT) {
+        runtime.BindIndexBuffer(buffer, offset, size);
+    } else {
+        runtime.BindIndexBuffer(maxwell3d.regs.draw.topology, maxwell3d.regs.index_array.format,
+                                maxwell3d.regs.index_array.first, maxwell3d.regs.index_array.count,
+                                buffer, offset, size);
+    }
+}
+
+template <class P>
+void BufferCache<P>::BindHostVertexBuffers() {
+    auto& flags = maxwell3d.dirty.flags;
+    for (u32 index = 0; index < NUM_VERTEX_BUFFERS; ++index) {
+        const Binding& binding = vertex_buffers[index];
+        Buffer& buffer = slot_buffers[binding.buffer_id];
+        SynchronizeBuffer(buffer, binding.cpu_addr, binding.size);
+        if (!flags[Dirty::VertexBuffer0 + index]) {
+            continue;
         }
+        flags[Dirty::VertexBuffer0 + index] = false;
+
+        const u32 stride = maxwell3d.regs.vertex_array[index].stride;
+        const u32 offset = buffer.Offset(binding.cpu_addr);
+        runtime.BindVertexBuffer(index, buffer, offset, binding.size, stride);
+    }
+}
 
-        const VAddr addr_end = addr + size;
-        auto it = mapped_addresses.lower_bound(addr);
-        if (it != mapped_addresses.begin()) {
-            --it;
+template <class P>
+void BufferCache<P>::BindHostGraphicsUniformBuffers(size_t stage) {
+    u32 dirty = ~0U;
+    if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) {
+        dirty = std::exchange(dirty_uniform_buffers[stage], 0);
+    }
+    u32 binding_index = 0;
+    ForEachEnabledBit(enabled_uniform_buffers[stage], [&](u32 index) {
+        const bool needs_bind = ((dirty >> index) & 1) != 0;
+        BindHostGraphicsUniformBuffer(stage, index, binding_index, needs_bind);
+        if constexpr (NEEDS_BIND_UNIFORM_INDEX) {
+            ++binding_index;
         }
-        while (it != mapped_addresses.end() && it->start < addr_end) {
-            if (it->Overlaps(addr, addr_end)) {
-                result.push_back(&*it);
+    });
+}
+
+template <class P>
+void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 binding_index,
+                                                   bool needs_bind) {
+    const Binding& binding = uniform_buffers[stage][index];
+    const VAddr cpu_addr = binding.cpu_addr;
+    const u32 size = binding.size;
+    Buffer& buffer = slot_buffers[binding.buffer_id];
+    if constexpr (IS_OPENGL) {
+        if (size <= SKIP_CACHE_SIZE && !buffer.IsRegionGpuModified(cpu_addr, size)) {
+            if (runtime.HasFastBufferSubData()) {
+                // Fast path for Nvidia
+                if (!HasFastUniformBufferBound(stage, binding_index)) {
+                    // We only have to bind when the currently bound buffer is not the fast version
+                    fast_bound_uniform_buffers[stage] |= 1U << binding_index;
+                    runtime.BindFastUniformBuffer(stage, binding_index, size);
+                }
+                const auto span = ImmediateBufferWithData(cpu_addr, size);
+                runtime.PushFastUniformBuffer(stage, binding_index, span);
+            } else {
+                // Stream buffer path to avoid stalling on non-Nvidia drivers
+                const auto span = runtime.BindMappedUniformBuffer(stage, binding_index, size);
+                cpu_memory.ReadBlockUnsafe(cpu_addr, span.data(), size);
             }
-            ++it;
+            return;
         }
-        return result;
     }
-
-    /// Returns a ticks counter used for tracking when cached objects were last modified
-    u64 GetModifiedTicks() {
-        return ++modified_ticks;
+    // Classic cached path
+    SynchronizeBuffer(buffer, cpu_addr, size);
+    if (!needs_bind && !HasFastUniformBufferBound(stage, binding_index)) {
+        // Skip binding if it's not needed and if the bound buffer is not the fast version
+        // This exists to avoid instances where the fast buffer is bound and a GPU write happens
+        return;
     }
+    fast_bound_uniform_buffers[stage] &= ~(1U << binding_index);
 
-    void FlushMap(MapInterval* map) {
-        const auto it = blocks.find(map->start >> BLOCK_PAGE_BITS);
-        ASSERT_OR_EXECUTE(it != blocks.end(), return;);
-
-        std::shared_ptr<Buffer> block = it->second;
-
-        const std::size_t size = map->end - map->start;
-        staging_buffer.resize(size);
-        block->Download(block->Offset(map->start), size, staging_buffer.data());
-        cpu_memory.WriteBlockUnsafe(map->start, staging_buffer.data(), size);
-        map->MarkAsModified(false, 0);
+    const u32 offset = buffer.Offset(cpu_addr);
+    if constexpr (NEEDS_BIND_UNIFORM_INDEX) {
+        runtime.BindUniformBuffer(stage, binding_index, buffer, offset, size);
+    } else {
+        runtime.BindUniformBuffer(buffer, offset, size);
     }
+}
+
+template <class P>
+void BufferCache<P>::BindHostGraphicsStorageBuffers(size_t stage) {
+    u32 binding_index = 0;
+    ForEachEnabledBit(enabled_storage_buffers[stage], [&](u32 index) {
+        const Binding& binding = storage_buffers[stage][index];
+        Buffer& buffer = slot_buffers[binding.buffer_id];
+        const u32 size = binding.size;
+        SynchronizeBuffer(buffer, binding.cpu_addr, size);
+
+        const u32 offset = buffer.Offset(binding.cpu_addr);
+        const bool is_written = ((written_storage_buffers[stage] >> index) & 1) != 0;
+        if constexpr (NEEDS_BIND_STORAGE_INDEX) {
+            runtime.BindStorageBuffer(stage, binding_index, buffer, offset, size, is_written);
+            ++binding_index;
+        } else {
+            runtime.BindStorageBuffer(buffer, offset, size, is_written);
+        }
+    });
+}
 
-    template <typename Callable>
-    BufferInfo StreamBufferUpload(std::size_t size, std::size_t alignment, Callable&& callable) {
-        AlignBuffer(alignment);
-        const std::size_t uploaded_offset = buffer_offset;
-        callable(buffer_ptr);
-
-        buffer_ptr += size;
-        buffer_offset += size;
-        return BufferInfo{stream_buffer.Handle(), uploaded_offset, stream_buffer.Address()};
+template <class P>
+void BufferCache<P>::BindHostTransformFeedbackBuffers() {
+    if (maxwell3d.regs.tfb_enabled == 0) {
+        return;
     }
-
-    void AlignBuffer(std::size_t alignment) {
-        // Align the offset, not the mapped pointer
-        const std::size_t offset_aligned = Common::AlignUp(buffer_offset, alignment);
-        buffer_ptr += offset_aligned - buffer_offset;
-        buffer_offset = offset_aligned;
+    for (u32 index = 0; index < NUM_TRANSFORM_FEEDBACK_BUFFERS; ++index) {
+        const Binding& binding = transform_feedback_buffers[index];
+        Buffer& buffer = slot_buffers[binding.buffer_id];
+        const u32 size = binding.size;
+        SynchronizeBuffer(buffer, binding.cpu_addr, size);
+
+        const u32 offset = buffer.Offset(binding.cpu_addr);
+        runtime.BindTransformFeedbackBuffer(index, buffer, offset, size);
     }
+}
 
-    std::shared_ptr<Buffer> EnlargeBlock(std::shared_ptr<Buffer> buffer) {
-        const std::size_t old_size = buffer->Size();
-        const std::size_t new_size = old_size + BLOCK_PAGE_SIZE;
-        const VAddr cpu_addr = buffer->CpuAddr();
-        std::shared_ptr<Buffer> new_buffer = CreateBlock(cpu_addr, new_size);
-        new_buffer->CopyFrom(*buffer, 0, 0, old_size);
-        QueueDestruction(std::move(buffer));
-
-        const VAddr cpu_addr_end = cpu_addr + new_size - 1;
-        const u64 page_end = cpu_addr_end >> BLOCK_PAGE_BITS;
-        for (u64 page_start = cpu_addr >> BLOCK_PAGE_BITS; page_start <= page_end; ++page_start) {
-            blocks.insert_or_assign(page_start, new_buffer);
+template <class P>
+void BufferCache<P>::BindHostComputeUniformBuffers() {
+    if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) {
+        // Mark all uniform buffers as dirty
+        dirty_uniform_buffers.fill(~u32{0});
+    }
+    u32 binding_index = 0;
+    ForEachEnabledBit(enabled_compute_uniform_buffers, [&](u32 index) {
+        const Binding& binding = compute_uniform_buffers[index];
+        Buffer& buffer = slot_buffers[binding.buffer_id];
+        const u32 size = binding.size;
+        SynchronizeBuffer(buffer, binding.cpu_addr, size);
+
+        const u32 offset = buffer.Offset(binding.cpu_addr);
+        if constexpr (NEEDS_BIND_UNIFORM_INDEX) {
+            runtime.BindComputeUniformBuffer(binding_index, buffer, offset, size);
+            ++binding_index;
+        } else {
+            runtime.BindUniformBuffer(buffer, offset, size);
         }
+    });
+}
+
+template <class P>
+void BufferCache<P>::BindHostComputeStorageBuffers() {
+    u32 binding_index = 0;
+    ForEachEnabledBit(enabled_compute_storage_buffers, [&](u32 index) {
+        const Binding& binding = compute_storage_buffers[index];
+        Buffer& buffer = slot_buffers[binding.buffer_id];
+        const u32 size = binding.size;
+        SynchronizeBuffer(buffer, binding.cpu_addr, size);
+
+        const u32 offset = buffer.Offset(binding.cpu_addr);
+        const bool is_written = ((written_compute_storage_buffers >> index) & 1) != 0;
+        if constexpr (NEEDS_BIND_STORAGE_INDEX) {
+            runtime.BindComputeStorageBuffer(binding_index, buffer, offset, size, is_written);
+            ++binding_index;
+        } else {
+            runtime.BindStorageBuffer(buffer, offset, size, is_written);
+        }
+    });
+}
 
-        return new_buffer;
+template <class P>
+void BufferCache<P>::DoUpdateGraphicsBuffers(bool is_indexed) {
+    if (is_indexed) {
+        UpdateIndexBuffer();
     }
+    UpdateVertexBuffers();
+    UpdateTransformFeedbackBuffers();
+    for (size_t stage = 0; stage < NUM_STAGES; ++stage) {
+        UpdateUniformBuffers(stage);
+        UpdateStorageBuffers(stage);
+    }
+}
+
+template <class P>
+void BufferCache<P>::DoUpdateComputeBuffers() {
+    UpdateComputeUniformBuffers();
+    UpdateComputeStorageBuffers();
+}
+
+template <class P>
+void BufferCache<P>::UpdateIndexBuffer() {
+    // We have to check for the dirty flags and index count
+    // The index count is currently changed without updating the dirty flags
+    const auto& index_array = maxwell3d.regs.index_array;
+    auto& flags = maxwell3d.dirty.flags;
+    if (!flags[Dirty::IndexBuffer] && last_index_count == index_array.count) {
+        return;
+    }
+    flags[Dirty::IndexBuffer] = false;
+    last_index_count = index_array.count;
+
+    const GPUVAddr gpu_addr_begin = index_array.StartAddress();
+    const GPUVAddr gpu_addr_end = index_array.EndAddress();
+    const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr_begin);
+    const u32 address_size = static_cast<u32>(gpu_addr_end - gpu_addr_begin);
+    const u32 draw_size = index_array.count * index_array.FormatSizeInBytes();
+    const u32 size = std::min(address_size, draw_size);
+    if (size == 0 || !cpu_addr) {
+        index_buffer = NULL_BINDING;
+        return;
+    }
+    index_buffer = Binding{
+        .cpu_addr = *cpu_addr,
+        .size = size,
+        .buffer_id = FindBuffer(*cpu_addr, size),
+    };
+}
 
-    std::shared_ptr<Buffer> MergeBlocks(std::shared_ptr<Buffer> first,
-                                        std::shared_ptr<Buffer> second) {
-        const std::size_t size_1 = first->Size();
-        const std::size_t size_2 = second->Size();
-        const VAddr first_addr = first->CpuAddr();
-        const VAddr second_addr = second->CpuAddr();
-        const VAddr new_addr = std::min(first_addr, second_addr);
-        const std::size_t new_size = size_1 + size_2;
-
-        std::shared_ptr<Buffer> new_buffer = CreateBlock(new_addr, new_size);
-        new_buffer->CopyFrom(*first, 0, new_buffer->Offset(first_addr), size_1);
-        new_buffer->CopyFrom(*second, 0, new_buffer->Offset(second_addr), size_2);
-        QueueDestruction(std::move(first));
-        QueueDestruction(std::move(second));
+template <class P>
+void BufferCache<P>::UpdateVertexBuffers() {
+    auto& flags = maxwell3d.dirty.flags;
+    if (!maxwell3d.dirty.flags[Dirty::VertexBuffers]) {
+        return;
+    }
+    flags[Dirty::VertexBuffers] = false;
 
-        const VAddr cpu_addr_end = new_addr + new_size - 1;
-        const u64 page_end = cpu_addr_end >> BLOCK_PAGE_BITS;
-        for (u64 page_start = new_addr >> BLOCK_PAGE_BITS; page_start <= page_end; ++page_start) {
-            blocks.insert_or_assign(page_start, new_buffer);
-        }
-        return new_buffer;
+    for (u32 index = 0; index < NUM_VERTEX_BUFFERS; ++index) {
+        UpdateVertexBuffer(index);
     }
+}
 
-    Buffer* GetBlock(VAddr cpu_addr, std::size_t size) {
-        std::shared_ptr<Buffer> found;
+template <class P>
+void BufferCache<P>::UpdateVertexBuffer(u32 index) {
+    if (!maxwell3d.dirty.flags[Dirty::VertexBuffer0 + index]) {
+        return;
+    }
+    const auto& array = maxwell3d.regs.vertex_array[index];
+    const auto& limit = maxwell3d.regs.vertex_array_limit[index];
+    const GPUVAddr gpu_addr_begin = array.StartAddress();
+    const GPUVAddr gpu_addr_end = limit.LimitAddress() + 1;
+    const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr_begin);
+    const u32 address_size = static_cast<u32>(gpu_addr_end - gpu_addr_begin);
+    const u32 size = address_size; // TODO: Analyze stride and number of vertices
+    if (array.enable == 0 || size == 0 || !cpu_addr) {
+        vertex_buffers[index] = NULL_BINDING;
+        return;
+    }
+    vertex_buffers[index] = Binding{
+        .cpu_addr = *cpu_addr,
+        .size = size,
+        .buffer_id = FindBuffer(*cpu_addr, size),
+    };
+}
+
+template <class P>
+void BufferCache<P>::UpdateUniformBuffers(size_t stage) {
+    ForEachEnabledBit(enabled_uniform_buffers[stage], [&](u32 index) {
+        Binding& binding = uniform_buffers[stage][index];
+        if (binding.buffer_id) {
+            // Already updated
+            return;
+        }
+        // Mark as dirty
+        if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) {
+            dirty_uniform_buffers[stage] |= 1U << index;
+        }
+        // Resolve buffer
+        binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size);
+    });
+}
+
+template <class P>
+void BufferCache<P>::UpdateStorageBuffers(size_t stage) {
+    const u32 written_mask = written_storage_buffers[stage];
+    ForEachEnabledBit(enabled_storage_buffers[stage], [&](u32 index) {
+        // Resolve buffer
+        Binding& binding = storage_buffers[stage][index];
+        const BufferId buffer_id = FindBuffer(binding.cpu_addr, binding.size);
+        binding.buffer_id = buffer_id;
+        // Mark buffer as written if needed
+        if (((written_mask >> index) & 1) != 0) {
+            MarkWrittenBuffer(buffer_id, binding.cpu_addr, binding.size);
+        }
+    });
+}
 
-        const VAddr cpu_addr_end = cpu_addr + size - 1;
-        const u64 page_end = cpu_addr_end >> BLOCK_PAGE_BITS;
-        for (u64 page_start = cpu_addr >> BLOCK_PAGE_BITS; page_start <= page_end; ++page_start) {
-            auto it = blocks.find(page_start);
-            if (it == blocks.end()) {
-                if (found) {
-                    found = EnlargeBlock(found);
-                    continue;
-                }
-                const VAddr start_addr = page_start << BLOCK_PAGE_BITS;
-                found = CreateBlock(start_addr, BLOCK_PAGE_SIZE);
-                blocks.insert_or_assign(page_start, found);
-                continue;
-            }
-            if (!found) {
-                found = it->second;
-                continue;
-            }
-            if (found != it->second) {
-                found = MergeBlocks(std::move(found), it->second);
+template <class P>
+void BufferCache<P>::UpdateTransformFeedbackBuffers() {
+    if (maxwell3d.regs.tfb_enabled == 0) {
+        return;
+    }
+    for (u32 index = 0; index < NUM_TRANSFORM_FEEDBACK_BUFFERS; ++index) {
+        UpdateTransformFeedbackBuffer(index);
+    }
+}
+
+template <class P>
+void BufferCache<P>::UpdateTransformFeedbackBuffer(u32 index) {
+    const auto& binding = maxwell3d.regs.tfb_bindings[index];
+    const GPUVAddr gpu_addr = binding.Address() + binding.buffer_offset;
+    const u32 size = binding.buffer_size;
+    const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr);
+    if (binding.buffer_enable == 0 || size == 0 || !cpu_addr) {
+        transform_feedback_buffers[index] = NULL_BINDING;
+        return;
+    }
+    const BufferId buffer_id = FindBuffer(*cpu_addr, size);
+    transform_feedback_buffers[index] = Binding{
+        .cpu_addr = *cpu_addr,
+        .size = size,
+        .buffer_id = buffer_id,
+    };
+    MarkWrittenBuffer(buffer_id, *cpu_addr, size);
+}
+
+template <class P>
+void BufferCache<P>::UpdateComputeUniformBuffers() {
+    ForEachEnabledBit(enabled_compute_uniform_buffers, [&](u32 index) {
+        Binding& binding = compute_uniform_buffers[index];
+        binding = NULL_BINDING;
+        const auto& launch_desc = kepler_compute.launch_description;
+        if (((launch_desc.const_buffer_enable_mask >> index) & 1) != 0) {
+            const auto& cbuf = launch_desc.const_buffer_config[index];
+            const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(cbuf.Address());
+            if (cpu_addr) {
+                binding.cpu_addr = *cpu_addr;
+                binding.size = cbuf.size;
             }
         }
-        return found.get();
+        binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size);
+    });
+}
+
+template <class P>
+void BufferCache<P>::UpdateComputeStorageBuffers() {
+    ForEachEnabledBit(enabled_compute_storage_buffers, [&](u32 index) {
+        // Resolve buffer
+        Binding& binding = compute_storage_buffers[index];
+        const BufferId buffer_id = FindBuffer(binding.cpu_addr, binding.size);
+        binding.buffer_id = buffer_id;
+        // Mark as written if needed
+        if (((written_compute_storage_buffers >> index) & 1) != 0) {
+            MarkWrittenBuffer(buffer_id, binding.cpu_addr, binding.size);
+        }
+    });
+}
+
+template <class P>
+void BufferCache<P>::MarkWrittenBuffer(BufferId buffer_id, VAddr cpu_addr, u32 size) {
+    Buffer& buffer = slot_buffers[buffer_id];
+    buffer.MarkRegionAsGpuModified(cpu_addr, size);
+
+    const bool is_accuracy_high = Settings::IsGPULevelHigh();
+    const bool is_async = Settings::values.use_asynchronous_gpu_emulation.GetValue();
+    if (!is_accuracy_high || !is_async) {
+        return;
     }
+    if (std::ranges::find(uncommitted_downloads, buffer_id) != uncommitted_downloads.end()) {
+        // Already inserted
+        return;
+    }
+    uncommitted_downloads.push_back(buffer_id);
+}
 
-    void MarkRegionAsWritten(VAddr start, VAddr end) {
-        const u64 page_end = end >> WRITE_PAGE_BIT;
-        for (u64 page_start = start >> WRITE_PAGE_BIT; page_start <= page_end; ++page_start) {
-            if (const auto [it, inserted] = written_pages.emplace(page_start, 1); !inserted) {
-                ++it->second;
-            }
+template <class P>
+BufferId BufferCache<P>::FindBuffer(VAddr cpu_addr, u32 size) {
+    if (cpu_addr == 0) {
+        return NULL_BUFFER_ID;
+    }
+    const u64 page = cpu_addr >> PAGE_BITS;
+    const BufferId buffer_id = page_table[page];
+    if (!buffer_id) {
+        return CreateBuffer(cpu_addr, size);
+    }
+    const Buffer& buffer = slot_buffers[buffer_id];
+    if (buffer.IsInBounds(cpu_addr, size)) {
+        return buffer_id;
+    }
+    return CreateBuffer(cpu_addr, size);
+}
+
+template <class P>
+BufferId BufferCache<P>::CreateBuffer(VAddr cpu_addr, u32 wanted_size) {
+    std::vector<BufferId> overlap_ids;
+    VAddr cpu_addr_begin = cpu_addr;
+    VAddr cpu_addr_end = cpu_addr + wanted_size;
+    for (; cpu_addr >> PAGE_BITS < Common::DivCeil(cpu_addr_end, PAGE_SIZE);
+         cpu_addr += PAGE_SIZE) {
+        const BufferId overlap_id = page_table[cpu_addr >> PAGE_BITS];
+        if (!overlap_id) {
+            continue;
+        }
+        Buffer& overlap = slot_buffers[overlap_id];
+        if (overlap.IsPicked()) {
+            continue;
+        }
+        overlap.Pick();
+        overlap_ids.push_back(overlap_id);
+        const VAddr overlap_cpu_addr = overlap.CpuAddr();
+        if (overlap_cpu_addr < cpu_addr_begin) {
+            cpu_addr = cpu_addr_begin = overlap_cpu_addr;
         }
+        cpu_addr_end = std::max(cpu_addr_end, overlap_cpu_addr + overlap.SizeBytes());
     }
-
-    void UnmarkRegionAsWritten(VAddr start, VAddr end) {
-        const u64 page_end = end >> WRITE_PAGE_BIT;
-        for (u64 page_start = start >> WRITE_PAGE_BIT; page_start <= page_end; ++page_start) {
-            auto it = written_pages.find(page_start);
-            if (it != written_pages.end()) {
-                if (it->second > 1) {
-                    --it->second;
-                } else {
-                    written_pages.erase(it);
-                }
-            }
+    const u32 size = static_cast<u32>(cpu_addr_end - cpu_addr_begin);
+    const BufferId new_buffer_id = slot_buffers.insert(runtime, rasterizer, cpu_addr_begin, size);
+    Buffer& new_buffer = slot_buffers[new_buffer_id];
+
+    for (const BufferId overlap_id : overlap_ids) {
+        Buffer& overlap = slot_buffers[overlap_id];
+        overlap.Unpick();
+
+        std::vector<BufferCopy> copies;
+        const size_t dst_base_offset = overlap.CpuAddr() - new_buffer.CpuAddr();
+        overlap.ForEachDownloadRange([&](u64 begin, u64 range_size) {
+            copies.push_back(BufferCopy{
+                .src_offset = begin,
+                .dst_offset = dst_base_offset + begin,
+                .size = range_size,
+            });
+            new_buffer.UnmarkRegionAsCpuModified(begin, range_size);
+            new_buffer.MarkRegionAsGpuModified(begin, range_size);
+        });
+        if (!copies.empty()) {
+            runtime.CopyBuffer(slot_buffers[new_buffer_id], overlap, copies);
+        }
+        ReplaceBufferDownloads(overlap_id, new_buffer_id);
+        DeleteBuffer(overlap_id);
+    }
+    Register(new_buffer_id);
+    return new_buffer_id;
+}
+
+template <class P>
+void BufferCache<P>::Register(BufferId buffer_id) {
+    ChangeRegister<true>(buffer_id);
+}
+
+template <class P>
+void BufferCache<P>::Unregister(BufferId buffer_id) {
+    ChangeRegister<false>(buffer_id);
+}
+
+template <class P>
+template <bool insert>
+void BufferCache<P>::ChangeRegister(BufferId buffer_id) {
+    const Buffer& buffer = slot_buffers[buffer_id];
+    const VAddr cpu_addr_begin = buffer.CpuAddr();
+    const VAddr cpu_addr_end = cpu_addr_begin + buffer.SizeBytes();
+    const u64 page_begin = cpu_addr_begin / PAGE_SIZE;
+    const u64 page_end = Common::DivCeil(cpu_addr_end, PAGE_SIZE);
+    for (u64 page = page_begin; page != page_end; ++page) {
+        if constexpr (insert) {
+            page_table[page] = buffer_id;
+        } else {
+            page_table[page] = BufferId{};
         }
     }
+}
 
-    bool IsRegionWritten(VAddr start, VAddr end) const {
-        const u64 page_end = end >> WRITE_PAGE_BIT;
-        for (u64 page_start = start >> WRITE_PAGE_BIT; page_start <= page_end; ++page_start) {
-            if (written_pages.contains(page_start)) {
-                return true;
+template <class P>
+void BufferCache<P>::SynchronizeBuffer(Buffer& buffer, VAddr cpu_addr, u32 size) {
+    if (buffer.CpuAddr() == 0) {
+        return;
+    }
+    SynchronizeBufferImpl(buffer, cpu_addr, size);
+}
+
+template <class P>
+void BufferCache<P>::SynchronizeBufferImpl(Buffer& buffer, VAddr cpu_addr, u32 size) {
+    boost::container::small_vector<BufferCopy, 4> copies;
+    u64 total_size_bytes = 0;
+    u64 largest_copy = 0;
+    buffer.ForEachUploadRange(cpu_addr, size, [&](u64 range_offset, u64 range_size) {
+        copies.push_back(BufferCopy{
+            .src_offset = total_size_bytes,
+            .dst_offset = range_offset,
+            .size = range_size,
+        });
+        total_size_bytes += range_size;
+        largest_copy = std::max(largest_copy, range_size);
+    });
+    if (total_size_bytes == 0) {
+        return;
+    }
+    const std::span<BufferCopy> copies_span(copies.data(), copies.size());
+    UploadMemory(buffer, total_size_bytes, largest_copy, copies_span);
+}
+
+template <class P>
+void BufferCache<P>::UploadMemory(Buffer& buffer, u64 total_size_bytes, u64 largest_copy,
+                                  std::span<BufferCopy> copies) {
+    if constexpr (USE_MEMORY_MAPS) {
+        MappedUploadMemory(buffer, total_size_bytes, copies);
+    } else {
+        ImmediateUploadMemory(buffer, largest_copy, copies);
+    }
+}
+
+template <class P>
+void BufferCache<P>::ImmediateUploadMemory(Buffer& buffer, u64 largest_copy,
+                                           std::span<const BufferCopy> copies) {
+    std::span<u8> immediate_buffer;
+    for (const BufferCopy& copy : copies) {
+        std::span<const u8> upload_span;
+        const VAddr cpu_addr = buffer.CpuAddr() + copy.dst_offset;
+        if (IsRangeGranular(cpu_addr, copy.size)) {
+            upload_span = std::span(cpu_memory.GetPointer(cpu_addr), copy.size);
+        } else {
+            if (immediate_buffer.empty()) {
+                immediate_buffer = ImmediateBuffer(largest_copy);
             }
+            cpu_memory.ReadBlockUnsafe(cpu_addr, immediate_buffer.data(), copy.size);
+            upload_span = immediate_buffer.subspan(0, copy.size);
         }
-        return false;
+        buffer.ImmediateUpload(copy.dst_offset, upload_span);
     }
-
-    void QueueDestruction(std::shared_ptr<Buffer> buffer) {
-        buffer->SetEpoch(epoch);
-        pending_destruction.push(std::move(buffer));
+}
+
+template <class P>
+void BufferCache<P>::MappedUploadMemory(Buffer& buffer, u64 total_size_bytes,
+                                        std::span<const BufferCopy> copies) {
+    auto upload_staging = runtime.UploadStagingBuffer(total_size_bytes);
+    const std::span<u8> staging_pointer = upload_staging.mapped_span;
+    for (const BufferCopy& copy : copies) {
+        const VAddr cpu_addr = buffer.CpuAddr() + copy.dst_offset;
+        u8* const src_pointer = staging_pointer.data() + copy.src_offset;
+        cpu_memory.ReadBlockUnsafe(cpu_addr, src_pointer, copy.size);
     }
-
-    void MarkForAsyncFlush(MapInterval* map) {
-        if (!uncommitted_flushes) {
-            uncommitted_flushes = std::make_shared<std::unordered_set<MapInterval*>>();
+    runtime.CopyBuffer(buffer, upload_staging.buffer, copies);
+}
+
+template <class P>
+void BufferCache<P>::DeleteBuffer(BufferId buffer_id) {
+    const auto scalar_replace = [buffer_id](Binding& binding) {
+        if (binding.buffer_id == buffer_id) {
+            binding.buffer_id = BufferId{};
+        }
+    };
+    const auto replace = [scalar_replace](std::span<Binding> bindings) {
+        std::ranges::for_each(bindings, scalar_replace);
+    };
+    scalar_replace(index_buffer);
+    replace(vertex_buffers);
+    std::ranges::for_each(uniform_buffers, replace);
+    std::ranges::for_each(storage_buffers, replace);
+    replace(transform_feedback_buffers);
+    replace(compute_uniform_buffers);
+    replace(compute_storage_buffers);
+    std::erase(cached_write_buffer_ids, buffer_id);
+
+    // Mark the whole buffer as CPU written to stop tracking CPU writes
+    Buffer& buffer = slot_buffers[buffer_id];
+    buffer.MarkRegionAsCpuModified(buffer.CpuAddr(), buffer.SizeBytes());
+
+    Unregister(buffer_id);
+    delayed_destruction_ring.Push(std::move(slot_buffers[buffer_id]));
+
+    NotifyBufferDeletion();
+}
+
+template <class P>
+void BufferCache<P>::ReplaceBufferDownloads(BufferId old_buffer_id, BufferId new_buffer_id) {
+    const auto replace = [old_buffer_id, new_buffer_id](std::vector<BufferId>& buffers) {
+        std::ranges::replace(buffers, old_buffer_id, new_buffer_id);
+        if (auto it = std::ranges::find(buffers, new_buffer_id); it != buffers.end()) {
+            buffers.erase(std::remove(it + 1, buffers.end(), new_buffer_id), buffers.end());
         }
-        uncommitted_flushes->insert(map);
+    };
+    replace(uncommitted_downloads);
+    std::ranges::for_each(committed_downloads, replace);
+}
+
+template <class P>
+void BufferCache<P>::NotifyBufferDeletion() {
+    if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) {
+        dirty_uniform_buffers.fill(~u32{0});
     }
+    auto& flags = maxwell3d.dirty.flags;
+    flags[Dirty::IndexBuffer] = true;
+    flags[Dirty::VertexBuffers] = true;
+    for (u32 index = 0; index < NUM_VERTEX_BUFFERS; ++index) {
+        flags[Dirty::VertexBuffer0 + index] = true;
+    }
+    has_deleted_buffers = true;
+}
+
+template <class P>
+typename BufferCache<P>::Binding BufferCache<P>::StorageBufferBinding(GPUVAddr ssbo_addr) const {
+    const GPUVAddr gpu_addr = gpu_memory.Read<u64>(ssbo_addr);
+    const u32 size = gpu_memory.Read<u32>(ssbo_addr + 8);
+    const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr);
+    if (!cpu_addr || size == 0) {
+        return NULL_BINDING;
+    }
+    const Binding binding{
+        .cpu_addr = *cpu_addr,
+        .size = size,
+        .buffer_id = BufferId{},
+    };
+    return binding;
+}
+
+template <class P>
+std::span<const u8> BufferCache<P>::ImmediateBufferWithData(VAddr cpu_addr, size_t size) {
+    u8* const base_pointer = cpu_memory.GetPointer(cpu_addr);
+    if (IsRangeGranular(cpu_addr, size) ||
+        base_pointer + size == cpu_memory.GetPointer(cpu_addr + size)) {
+        return std::span(base_pointer, size);
+    } else {
+        const std::span<u8> span = ImmediateBuffer(size);
+        cpu_memory.ReadBlockUnsafe(cpu_addr, span.data(), size);
+        return span;
+    }
+}
 
-    VideoCore::RasterizerInterface& rasterizer;
-    Tegra::MemoryManager& gpu_memory;
-    Core::Memory::Memory& cpu_memory;
-    StreamBuffer& stream_buffer;
-
-    u8* buffer_ptr = nullptr;
-    u64 buffer_offset = 0;
-    u64 buffer_offset_base = 0;
-
-    MapIntervalAllocator mapped_addresses_allocator;
-    boost::intrusive::set<MapInterval, boost::intrusive::compare<MapIntervalCompare>>
-        mapped_addresses;
-
-    std::unordered_map<u64, u32> written_pages;
-    std::unordered_map<u64, std::shared_ptr<Buffer>> blocks;
-
-    std::queue<std::shared_ptr<Buffer>> pending_destruction;
-    u64 epoch = 0;
-    u64 modified_ticks = 0;
-
-    std::vector<u8> staging_buffer;
-
-    std::list<MapInterval*> marked_for_unregister;
-
-    std::shared_ptr<std::unordered_set<MapInterval*>> uncommitted_flushes;
-    std::list<std::shared_ptr<std::list<MapInterval*>>> committed_flushes;
-
-    std::recursive_mutex mutex;
-};
+template <class P>
+std::span<u8> BufferCache<P>::ImmediateBuffer(size_t wanted_capacity) {
+    if (wanted_capacity > immediate_buffer_capacity) {
+        immediate_buffer_capacity = wanted_capacity;
+        immediate_buffer_alloc = std::make_unique<u8[]>(wanted_capacity);
+    }
+    return std::span<u8>(immediate_buffer_alloc.get(), wanted_capacity);
+}
+
+template <class P>
+bool BufferCache<P>::HasFastUniformBufferBound(size_t stage, u32 binding_index) const noexcept {
+    if constexpr (IS_OPENGL) {
+        return ((fast_bound_uniform_buffers[stage] >> binding_index) & 1) != 0;
+    } else {
+        // Only OpenGL has fast uniform buffers
+        return false;
+    }
+}
 
 } // namespace VideoCommon
diff --git a/src/video_core/buffer_cache/map_interval.cpp b/src/video_core/buffer_cache/map_interval.cpp
deleted file mode 100644
index 62587e18a..000000000
--- a/src/video_core/buffer_cache/map_interval.cpp
+++ /dev/null
@@ -1,33 +0,0 @@
-// Copyright 2020 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include <algorithm>
-#include <array>
-#include <cstddef>
-#include <memory>
-
-#include "video_core/buffer_cache/map_interval.h"
-
-namespace VideoCommon {
-
-MapIntervalAllocator::MapIntervalAllocator() {
-    FillFreeList(first_chunk);
-}
-
-MapIntervalAllocator::~MapIntervalAllocator() = default;
-
-void MapIntervalAllocator::AllocateNewChunk() {
-    *new_chunk = std::make_unique<Chunk>();
-    FillFreeList(**new_chunk);
-    new_chunk = &(*new_chunk)->next;
-}
-
-void MapIntervalAllocator::FillFreeList(Chunk& chunk) {
-    const std::size_t old_size = free_list.size();
-    free_list.resize(old_size + chunk.data.size());
-    std::transform(chunk.data.rbegin(), chunk.data.rend(), free_list.begin() + old_size,
-                   [](MapInterval& interval) { return &interval; });
-}
-
-} // namespace VideoCommon
diff --git a/src/video_core/buffer_cache/map_interval.h b/src/video_core/buffer_cache/map_interval.h
deleted file mode 100644
index ef974b08a..000000000
--- a/src/video_core/buffer_cache/map_interval.h
+++ /dev/null
@@ -1,93 +0,0 @@
-// Copyright 2019 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <array>
-#include <cstddef>
-#include <memory>
-#include <vector>
-
-#include <boost/intrusive/set_hook.hpp>
-
-#include "common/common_types.h"
-#include "video_core/gpu.h"
-
-namespace VideoCommon {
-
-struct MapInterval : public boost::intrusive::set_base_hook<boost::intrusive::optimize_size<true>> {
-    MapInterval() = default;
-
-    /*implicit*/ MapInterval(VAddr start_) noexcept : start{start_} {}
-
-    explicit MapInterval(VAddr start_, VAddr end_, GPUVAddr gpu_addr_) noexcept
-        : start{start_}, end{end_}, gpu_addr{gpu_addr_} {}
-
-    bool IsInside(VAddr other_start, VAddr other_end) const noexcept {
-        return start <= other_start && other_end <= end;
-    }
-
-    bool Overlaps(VAddr other_start, VAddr other_end) const noexcept {
-        return start < other_end && other_start < end;
-    }
-
-    void MarkAsModified(bool is_modified_, u64 ticks_) noexcept {
-        is_modified = is_modified_;
-        ticks = ticks_;
-    }
-
-    boost::intrusive::set_member_hook<> member_hook_;
-    VAddr start = 0;
-    VAddr end = 0;
-    GPUVAddr gpu_addr = 0;
-    u64 ticks = 0;
-    bool is_written = false;
-    bool is_modified = false;
-    bool is_registered = false;
-    bool is_memory_marked = false;
-    bool is_sync_pending = false;
-};
-
-struct MapIntervalCompare {
-    constexpr bool operator()(const MapInterval& lhs, const MapInterval& rhs) const noexcept {
-        return lhs.start < rhs.start;
-    }
-};
-
-class MapIntervalAllocator {
-public:
-    MapIntervalAllocator();
-    ~MapIntervalAllocator();
-
-    MapInterval* Allocate() {
-        if (free_list.empty()) {
-            AllocateNewChunk();
-        }
-        MapInterval* const interval = free_list.back();
-        free_list.pop_back();
-        return interval;
-    }
-
-    void Release(MapInterval* interval) {
-        free_list.push_back(interval);
-    }
-
-private:
-    struct Chunk {
-        std::unique_ptr<Chunk> next;
-        std::array<MapInterval, 0x8000> data;
-    };
-
-    void AllocateNewChunk();
-
-    void FillFreeList(Chunk& chunk);
-
-    std::vector<MapInterval*> free_list;
-
-    Chunk first_chunk;
-
-    std::unique_ptr<Chunk>* new_chunk = &first_chunk.next;
-};
-
-} // namespace VideoCommon
diff --git a/src/video_core/command_classes/vic.cpp b/src/video_core/command_classes/vic.cpp
index 55e632346..2b7569335 100644
--- a/src/video_core/command_classes/vic.cpp
+++ b/src/video_core/command_classes/vic.cpp
@@ -110,12 +110,10 @@ void Vic::Execute() {
                                            converted_frame_buffer.get(), block_height, 0, 0);
 
             gpu.MemoryManager().WriteBlock(output_surface_luma_address, swizzled_data.data(), size);
-            gpu.Maxwell3D().OnMemoryWrite();
         } else {
             // send pitch linear frame
             gpu.MemoryManager().WriteBlock(output_surface_luma_address, converted_frame_buf_addr,
                                            linear_size);
-            gpu.Maxwell3D().OnMemoryWrite();
         }
         break;
     }
@@ -163,7 +161,6 @@ void Vic::Execute() {
         }
         gpu.MemoryManager().WriteBlock(output_surface_chroma_u_address, chroma_buffer.data(),
                                        chroma_buffer.size());
-        gpu.Maxwell3D().OnMemoryWrite();
         break;
     }
     default:
diff --git a/src/video_core/dirty_flags.cpp b/src/video_core/dirty_flags.cpp
index b1eaac00c..7149af290 100644
--- a/src/video_core/dirty_flags.cpp
+++ b/src/video_core/dirty_flags.cpp
@@ -12,13 +12,30 @@
 #define NUM(field_name) (sizeof(::Tegra::Engines::Maxwell3D::Regs::field_name) / (sizeof(u32)))
 
 namespace VideoCommon::Dirty {
-
+namespace {
 using Tegra::Engines::Maxwell3D;
 
-void SetupDirtyRenderTargets(Tegra::Engines::Maxwell3D::DirtyState::Tables& tables) {
+void SetupDirtyVertexBuffers(Maxwell3D::DirtyState::Tables& tables) {
+    static constexpr std::size_t num_array = 3;
+    for (std::size_t i = 0; i < Maxwell3D::Regs::NumVertexArrays; ++i) {
+        const std::size_t array_offset = OFF(vertex_array) + i * NUM(vertex_array[0]);
+        const std::size_t limit_offset = OFF(vertex_array_limit) + i * NUM(vertex_array_limit[0]);
+
+        FillBlock(tables, array_offset, num_array, VertexBuffer0 + i, VertexBuffers);
+        FillBlock(tables, limit_offset, NUM(vertex_array_limit), VertexBuffer0 + i, VertexBuffers);
+    }
+}
+
+void SetupIndexBuffer(Maxwell3D::DirtyState::Tables& tables) {
+    FillBlock(tables[0], OFF(index_array), NUM(index_array), IndexBuffer);
+}
+
+void SetupDirtyDescriptors(Maxwell3D::DirtyState::Tables& tables) {
     FillBlock(tables[0], OFF(tic), NUM(tic), Descriptors);
     FillBlock(tables[0], OFF(tsc), NUM(tsc), Descriptors);
+}
 
+void SetupDirtyRenderTargets(Maxwell3D::DirtyState::Tables& tables) {
     static constexpr std::size_t num_per_rt = NUM(rt[0]);
     static constexpr std::size_t begin = OFF(rt);
     static constexpr std::size_t num = num_per_rt * Maxwell3D::Regs::NumRenderTargets;
@@ -41,5 +58,13 @@ void SetupDirtyRenderTargets(Tegra::Engines::Maxwell3D::DirtyState::Tables& tabl
         FillBlock(table, OFF(zeta), NUM(zeta), flag);
     }
 }
+} // Anonymous namespace
+
+void SetupDirtyFlags(Maxwell3D::DirtyState::Tables& tables) {
+    SetupDirtyVertexBuffers(tables);
+    SetupIndexBuffer(tables);
+    SetupDirtyDescriptors(tables);
+    SetupDirtyRenderTargets(tables);
+}
 
 } // namespace VideoCommon::Dirty
diff --git a/src/video_core/dirty_flags.h b/src/video_core/dirty_flags.h
index 875527ddd..702688ace 100644
--- a/src/video_core/dirty_flags.h
+++ b/src/video_core/dirty_flags.h
@@ -30,6 +30,12 @@ enum : u8 {
     ColorBuffer7,
     ZetaBuffer,
 
+    VertexBuffers,
+    VertexBuffer0,
+    VertexBuffer31 = VertexBuffer0 + 31,
+
+    IndexBuffer,
+
     LastCommonEntry,
 };
 
@@ -47,6 +53,6 @@ void FillBlock(Tegra::Engines::Maxwell3D::DirtyState::Tables& tables, std::size_
     FillBlock(tables[1], begin, num, index_b);
 }
 
-void SetupDirtyRenderTargets(Tegra::Engines::Maxwell3D::DirtyState::Tables& tables);
+void SetupDirtyFlags(Tegra::Engines::Maxwell3D::DirtyState::Tables& tables);
 
 } // namespace VideoCommon::Dirty
diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp
index 2c8b20024..8b33c04ab 100644
--- a/src/video_core/dma_pusher.cpp
+++ b/src/video_core/dma_pusher.cpp
@@ -23,8 +23,6 @@ void DmaPusher::DispatchCalls() {
     MICROPROFILE_SCOPE(DispatchCalls);
 
     gpu.SyncGuestHost();
-    // On entering GPU code, assume all memory may be touched by the ARM core.
-    gpu.Maxwell3D().OnMemoryWrite();
 
     dma_pushbuffer_subindex = 0;
 
diff --git a/src/video_core/engines/kepler_compute.cpp b/src/video_core/engines/kepler_compute.cpp
index ed29fc7ac..a9b75091e 100644
--- a/src/video_core/engines/kepler_compute.cpp
+++ b/src/video_core/engines/kepler_compute.cpp
@@ -39,7 +39,6 @@ void KeplerCompute::CallMethod(u32 method, u32 method_argument, bool is_last_cal
     case KEPLER_COMPUTE_REG_INDEX(data_upload): {
         upload_state.ProcessData(method_argument, is_last_call);
         if (is_last_call) {
-            system.GPU().Maxwell3D().OnMemoryWrite();
         }
         break;
     }
diff --git a/src/video_core/engines/kepler_memory.cpp b/src/video_core/engines/kepler_memory.cpp
index 9911140e9..560551157 100644
--- a/src/video_core/engines/kepler_memory.cpp
+++ b/src/video_core/engines/kepler_memory.cpp
@@ -33,7 +33,6 @@ void KeplerMemory::CallMethod(u32 method, u32 method_argument, bool is_last_call
     case KEPLERMEMORY_REG_INDEX(data): {
         upload_state.ProcessData(method_argument, is_last_call);
         if (is_last_call) {
-            system.GPU().Maxwell3D().OnMemoryWrite();
         }
         break;
     }
diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp
index d6ba9da5c..75517a4f7 100644
--- a/src/video_core/engines/maxwell_3d.cpp
+++ b/src/video_core/engines/maxwell_3d.cpp
@@ -223,7 +223,6 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume
     case MAXWELL3D_REG_INDEX(data_upload):
         upload_state.ProcessData(argument, is_last_call);
         if (is_last_call) {
-            OnMemoryWrite();
         }
         return;
     case MAXWELL3D_REG_INDEX(fragment_barrier):
@@ -570,17 +569,18 @@ std::optional<u64> Maxwell3D::GetQueryResult() {
     }
 }
 
-void Maxwell3D::ProcessCBBind(std::size_t stage_index) {
+void Maxwell3D::ProcessCBBind(size_t stage_index) {
     // Bind the buffer currently in CB_ADDRESS to the specified index in the desired shader stage.
-    auto& shader = state.shader_stages[stage_index];
-    auto& bind_data = regs.cb_bind[stage_index];
-
-    ASSERT(bind_data.index < Regs::MaxConstBuffers);
-    auto& buffer = shader.const_buffers[bind_data.index];
-
+    const auto& bind_data = regs.cb_bind[stage_index];
+    auto& buffer = state.shader_stages[stage_index].const_buffers[bind_data.index];
     buffer.enabled = bind_data.valid.Value() != 0;
     buffer.address = regs.const_buffer.BufferAddress();
     buffer.size = regs.const_buffer.cb_size;
+
+    const bool is_enabled = bind_data.valid.Value() != 0;
+    const GPUVAddr gpu_addr = is_enabled ? regs.const_buffer.BufferAddress() : 0;
+    const u32 size = is_enabled ? regs.const_buffer.cb_size : 0;
+    rasterizer->BindGraphicsUniformBuffer(stage_index, bind_data.index, gpu_addr, size);
 }
 
 void Maxwell3D::ProcessCBData(u32 value) {
@@ -635,7 +635,6 @@ void Maxwell3D::FinishCBData() {
 
     const u32 id = cb_data_state.id;
     memory_manager.WriteBlock(address, cb_data_state.buffer[id].data(), size);
-    OnMemoryWrite();
 
     cb_data_state.id = null_cb_data;
     cb_data_state.current = null_cb_data;
diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h
index cc94d2678..ffed42a29 100644
--- a/src/video_core/engines/maxwell_3d.h
+++ b/src/video_core/engines/maxwell_3d.h
@@ -1314,8 +1314,7 @@ public:
 
                     GPUVAddr LimitAddress() const {
                         return static_cast<GPUVAddr>((static_cast<GPUVAddr>(limit_high) << 32) |
-                                                     limit_low) +
-                               1;
+                                                     limit_low);
                     }
                 } vertex_array_limit[NumVertexArrays];
 
@@ -1403,6 +1402,7 @@ public:
         };
 
         std::array<ShaderStageInfo, Regs::MaxShaderStage> shader_stages;
+
         u32 current_instance = 0; ///< Current instance to be used to simulate instanced rendering.
     };
 
@@ -1452,11 +1452,6 @@ public:
         return *rasterizer;
     }
 
-    /// Notify a memory write has happened.
-    void OnMemoryWrite() {
-        dirty.flags |= dirty.on_write_stores;
-    }
-
     enum class MMEDrawMode : u32 {
         Undefined,
         Array,
@@ -1478,7 +1473,6 @@ public:
         using Tables = std::array<Table, 2>;
 
         Flags flags;
-        Flags on_write_stores;
         Tables tables{};
     } dirty;
 
@@ -1541,7 +1535,7 @@ private:
     void FinishCBData();
 
     /// Handles a write to the CB_BIND register.
-    void ProcessCBBind(std::size_t stage_index);
+    void ProcessCBBind(size_t stage_index);
 
     /// Handles a write to the VERTEX_END_GL register, triggering a draw.
     void DrawArrays();
diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp
index ba750748c..a2f19559f 100644
--- a/src/video_core/engines/maxwell_dma.cpp
+++ b/src/video_core/engines/maxwell_dma.cpp
@@ -60,9 +60,6 @@ void MaxwellDMA::Launch() {
         return;
     }
 
-    // All copies here update the main memory, so mark all rasterizer states as invalid.
-    system.GPU().Maxwell3D().OnMemoryWrite();
-
     if (is_src_pitch && is_dst_pitch) {
         CopyPitchToPitch();
     } else {
diff --git a/src/video_core/fence_manager.h b/src/video_core/fence_manager.h
index 3512283ff..f055b61e9 100644
--- a/src/video_core/fence_manager.h
+++ b/src/video_core/fence_manager.h
@@ -143,22 +143,26 @@ private:
     }
 
     bool ShouldWait() const {
+        std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
         return texture_cache.ShouldWaitAsyncFlushes() || buffer_cache.ShouldWaitAsyncFlushes() ||
                query_cache.ShouldWaitAsyncFlushes();
     }
 
     bool ShouldFlush() const {
+        std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
         return texture_cache.HasUncommittedFlushes() || buffer_cache.HasUncommittedFlushes() ||
                query_cache.HasUncommittedFlushes();
     }
 
     void PopAsyncFlushes() {
+        std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
         texture_cache.PopAsyncFlushes();
         buffer_cache.PopAsyncFlushes();
         query_cache.PopAsyncFlushes();
     }
 
     void CommitAsyncFlushes() {
+        std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
         texture_cache.CommitAsyncFlushes();
         buffer_cache.CommitAsyncFlushes();
         query_cache.CommitAsyncFlushes();
diff --git a/src/video_core/host_shaders/CMakeLists.txt b/src/video_core/host_shaders/CMakeLists.txt
index 28f2b8614..970120acc 100644
--- a/src/video_core/host_shaders/CMakeLists.txt
+++ b/src/video_core/host_shaders/CMakeLists.txt
@@ -12,7 +12,6 @@ set(SHADER_FILES
     vulkan_blit_depth_stencil.frag
     vulkan_present.frag
     vulkan_present.vert
-    vulkan_quad_array.comp
     vulkan_quad_indexed.comp
     vulkan_uint8.comp
 )
diff --git a/src/video_core/host_shaders/vulkan_quad_array.comp b/src/video_core/host_shaders/vulkan_quad_array.comp
deleted file mode 100644
index 212f4e998..000000000
--- a/src/video_core/host_shaders/vulkan_quad_array.comp
+++ /dev/null
@@ -1,28 +0,0 @@
-// Copyright 2019 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#version 460 core
-
-layout (local_size_x = 1024) in;
-
-layout (std430, set = 0, binding = 0) buffer OutputBuffer {
-    uint output_indexes[];
-};
-
-layout (push_constant) uniform PushConstants {
-    uint first;
-};
-
-void main() {
-    uint primitive = gl_GlobalInvocationID.x;
-    if (primitive * 6 >= output_indexes.length()) {
-        return;
-    }
-
-    const uint quad_map[6] = uint[](0, 1, 2, 0, 2, 3);
-    for (uint vertex = 0; vertex < 6; ++vertex) {
-        uint index = first + primitive * 4 + quad_map[vertex];
-        output_indexes[primitive * 6 + vertex] = index;
-    }
-}
diff --git a/src/video_core/host_shaders/vulkan_uint8.comp b/src/video_core/host_shaders/vulkan_uint8.comp
index ad74d7af9..872291670 100644
--- a/src/video_core/host_shaders/vulkan_uint8.comp
+++ b/src/video_core/host_shaders/vulkan_uint8.comp
@@ -16,9 +16,16 @@ layout (std430, set = 0, binding = 1) writeonly buffer OutputBuffer {
     uint16_t output_indexes[];
 };
 
+uint AssembleIndex(uint id) {
+    // Most primitive restart indices are 0xFF
+    // Hardcode this to 0xFF for now
+    uint index = uint(input_indexes[id]);
+    return index == 0xFF ? 0xFFFF : index;
+}
+
 void main() {
     uint id = gl_GlobalInvocationID.x;
     if (id < input_indexes.length()) {
-        output_indexes[id] = uint16_t(input_indexes[id]);
+        output_indexes[id] = uint16_t(AssembleIndex(id));
     }
 }
diff --git a/src/video_core/rasterizer_interface.h b/src/video_core/rasterizer_interface.h
index 0cb0f387d..50491b758 100644
--- a/src/video_core/rasterizer_interface.h
+++ b/src/video_core/rasterizer_interface.h
@@ -7,6 +7,7 @@
 #include <atomic>
 #include <functional>
 #include <optional>
+#include <span>
 #include "common/common_types.h"
 #include "video_core/engines/fermi_2d.h"
 #include "video_core/gpu.h"
@@ -49,6 +50,10 @@ public:
     /// Records a GPU query and caches it
     virtual void Query(GPUVAddr gpu_addr, QueryType type, std::optional<u64> timestamp) = 0;
 
+    /// Signal an uniform buffer binding
+    virtual void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr,
+                                           u32 size) = 0;
+
     /// Signal a GPU based semaphore as a fence
     virtual void SignalSemaphore(GPUVAddr addr, u32 value) = 0;
 
diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.cpp b/src/video_core/renderer_opengl/gl_buffer_cache.cpp
index 5772cad87..889ad6c56 100644
--- a/src/video_core/renderer_opengl/gl_buffer_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_buffer_cache.cpp
@@ -2,98 +2,235 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include <memory>
+#include <span>
 
-#include <glad/glad.h>
-
-#include "common/assert.h"
-#include "common/microprofile.h"
 #include "video_core/buffer_cache/buffer_cache.h"
-#include "video_core/engines/maxwell_3d.h"
-#include "video_core/rasterizer_interface.h"
 #include "video_core/renderer_opengl/gl_buffer_cache.h"
 #include "video_core/renderer_opengl/gl_device.h"
-#include "video_core/renderer_opengl/gl_rasterizer.h"
-#include "video_core/renderer_opengl/gl_resource_manager.h"
+#include "video_core/vulkan_common/vulkan_device.h"
+#include "video_core/vulkan_common/vulkan_instance.h"
+#include "video_core/vulkan_common/vulkan_library.h"
+#include "video_core/vulkan_common/vulkan_memory_allocator.h"
 
 namespace OpenGL {
+namespace {
+struct BindlessSSBO {
+    GLuint64EXT address;
+    GLsizei length;
+    GLsizei padding;
+};
+static_assert(sizeof(BindlessSSBO) == sizeof(GLuint) * 4);
+
+constexpr std::array PROGRAM_LUT{
+    GL_VERTEX_PROGRAM_NV,   GL_TESS_CONTROL_PROGRAM_NV, GL_TESS_EVALUATION_PROGRAM_NV,
+    GL_GEOMETRY_PROGRAM_NV, GL_FRAGMENT_PROGRAM_NV,
+};
+} // Anonymous namespace
+
+Buffer::Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams null_params)
+    : VideoCommon::BufferBase<VideoCore::RasterizerInterface>(null_params) {}
+
+Buffer::Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rasterizer_,
+               VAddr cpu_addr_, u64 size_bytes_)
+    : VideoCommon::BufferBase<VideoCore::RasterizerInterface>(rasterizer_, cpu_addr_, size_bytes_) {
+    buffer.Create();
+    const std::string name = fmt::format("Buffer 0x{:x}", CpuAddr());
+    glObjectLabel(GL_BUFFER, buffer.handle, static_cast<GLsizei>(name.size()), name.data());
+    if (runtime.device.UseAssemblyShaders()) {
+        CreateMemoryObjects(runtime);
+        glNamedBufferStorageMemEXT(buffer.handle, SizeBytes(), memory_commit.ExportOpenGLHandle(),
+                                   memory_commit.Offset());
+    } else {
+        glNamedBufferData(buffer.handle, SizeBytes(), nullptr, GL_DYNAMIC_DRAW);
+    }
+    if (runtime.has_unified_vertex_buffers) {
+        glGetNamedBufferParameterui64vNV(buffer.handle, GL_BUFFER_GPU_ADDRESS_NV, &address);
+    }
+}
 
-using Maxwell = Tegra::Engines::Maxwell3D::Regs;
+void Buffer::ImmediateUpload(size_t offset, std::span<const u8> data) noexcept {
+    glNamedBufferSubData(buffer.handle, static_cast<GLintptr>(offset),
+                         static_cast<GLsizeiptr>(data.size_bytes()), data.data());
+}
 
-MICROPROFILE_DEFINE(OpenGL_Buffer_Download, "OpenGL", "Buffer Download", MP_RGB(192, 192, 128));
+void Buffer::ImmediateDownload(size_t offset, std::span<u8> data) noexcept {
+    glGetNamedBufferSubData(buffer.handle, static_cast<GLintptr>(offset),
+                            static_cast<GLsizeiptr>(data.size_bytes()), data.data());
+}
 
-Buffer::Buffer(const Device& device_, VAddr cpu_addr_, std::size_t size_)
-    : BufferBlock{cpu_addr_, size_} {
-    gl_buffer.Create();
-    glNamedBufferData(gl_buffer.handle, static_cast<GLsizeiptr>(size_), nullptr, GL_DYNAMIC_DRAW);
-    if (device_.UseAssemblyShaders() || device_.HasVertexBufferUnifiedMemory()) {
-        glMakeNamedBufferResidentNV(gl_buffer.handle, GL_READ_WRITE);
-        glGetNamedBufferParameterui64vNV(gl_buffer.handle, GL_BUFFER_GPU_ADDRESS_NV, &gpu_address);
+void Buffer::MakeResident(GLenum access) noexcept {
+    // Abuse GLenum's order to exit early
+    // GL_NONE (default) < GL_READ_ONLY < GL_READ_WRITE
+    if (access <= current_residency_access || buffer.handle == 0) {
+        return;
+    }
+    if (std::exchange(current_residency_access, access) != GL_NONE) {
+        // If the buffer is already resident, remove its residency before promoting it
+        glMakeNamedBufferNonResidentNV(buffer.handle);
     }
+    glMakeNamedBufferResidentNV(buffer.handle, access);
 }
 
-Buffer::~Buffer() = default;
+GLuint Buffer::SubBuffer(u32 offset) {
+    if (offset == 0) {
+        return buffer.handle;
+    }
+    for (const auto& [sub_buffer, sub_offset] : subs) {
+        if (sub_offset == offset) {
+            return sub_buffer.handle;
+        }
+    }
+    OGLBuffer sub_buffer;
+    sub_buffer.Create();
+    glNamedBufferStorageMemEXT(sub_buffer.handle, SizeBytes() - offset,
+                               memory_commit.ExportOpenGLHandle(), memory_commit.Offset() + offset);
+    return subs.emplace_back(std::move(sub_buffer), offset).first.handle;
+}
 
-void Buffer::Upload(std::size_t offset, std::size_t data_size, const u8* data) {
-    glNamedBufferSubData(Handle(), static_cast<GLintptr>(offset),
-                         static_cast<GLsizeiptr>(data_size), data);
+void Buffer::CreateMemoryObjects(BufferCacheRuntime& runtime) {
+    auto& allocator = runtime.vulkan_memory_allocator;
+    auto& device = runtime.vulkan_device->GetLogical();
+    auto vulkan_buffer = device.CreateBuffer(VkBufferCreateInfo{
+        .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .size = SizeBytes(),
+        .usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT |
+                 VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT |
+                 VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT |
+                 VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT |
+                 VK_BUFFER_USAGE_VERTEX_BUFFER_BIT,
+        .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
+        .queueFamilyIndexCount = 0,
+        .pQueueFamilyIndices = nullptr,
+    });
+    const VkMemoryRequirements requirements = device.GetBufferMemoryRequirements(*vulkan_buffer);
+    memory_commit = allocator->Commit(requirements, Vulkan::MemoryUsage::DeviceLocal);
 }
 
-void Buffer::Download(std::size_t offset, std::size_t data_size, u8* data) {
-    MICROPROFILE_SCOPE(OpenGL_Buffer_Download);
-    const GLsizeiptr gl_size = static_cast<GLsizeiptr>(data_size);
-    const GLintptr gl_offset = static_cast<GLintptr>(offset);
-    if (read_buffer.handle == 0) {
-        read_buffer.Create();
-        glNamedBufferData(read_buffer.handle, static_cast<GLsizeiptr>(Size()), nullptr,
-                          GL_STREAM_READ);
+BufferCacheRuntime::BufferCacheRuntime(const Device& device_, const Vulkan::Device* vulkan_device_,
+                                       Vulkan::MemoryAllocator* vulkan_memory_allocator_)
+    : device{device_}, vulkan_device{vulkan_device_},
+      vulkan_memory_allocator{vulkan_memory_allocator_},
+      stream_buffer{device.HasFastBufferSubData() ? std::nullopt
+                                                  : std::make_optional<StreamBuffer>()} {
+    GLint gl_max_attributes;
+    glGetIntegerv(GL_MAX_VERTEX_ATTRIBS, &gl_max_attributes);
+    max_attributes = static_cast<u32>(gl_max_attributes);
+    use_assembly_shaders = device.UseAssemblyShaders();
+    has_unified_vertex_buffers = device.HasVertexBufferUnifiedMemory();
+
+    for (auto& stage_uniforms : fast_uniforms) {
+        for (OGLBuffer& buffer : stage_uniforms) {
+            buffer.Create();
+            glNamedBufferData(buffer.handle, BufferCache::SKIP_CACHE_SIZE, nullptr, GL_STREAM_DRAW);
+        }
     }
-    glMemoryBarrier(GL_BUFFER_UPDATE_BARRIER_BIT);
-    glCopyNamedBufferSubData(gl_buffer.handle, read_buffer.handle, gl_offset, gl_offset, gl_size);
-    glGetNamedBufferSubData(read_buffer.handle, gl_offset, gl_size, data);
 }
 
-void Buffer::CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst_offset,
-                      std::size_t copy_size) {
-    glCopyNamedBufferSubData(src.Handle(), Handle(), static_cast<GLintptr>(src_offset),
-                             static_cast<GLintptr>(dst_offset), static_cast<GLsizeiptr>(copy_size));
+void BufferCacheRuntime::CopyBuffer(Buffer& dst_buffer, Buffer& src_buffer,
+                                    std::span<const VideoCommon::BufferCopy> copies) {
+    for (const VideoCommon::BufferCopy& copy : copies) {
+        glCopyNamedBufferSubData(
+            src_buffer.Handle(), dst_buffer.Handle(), static_cast<GLintptr>(copy.src_offset),
+            static_cast<GLintptr>(copy.dst_offset), static_cast<GLsizeiptr>(copy.size));
+    }
 }
 
-OGLBufferCache::OGLBufferCache(VideoCore::RasterizerInterface& rasterizer_,
-                               Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_,
-                               const Device& device_, OGLStreamBuffer& stream_buffer_,
-                               StateTracker& state_tracker)
-    : GenericBufferCache{rasterizer_, gpu_memory_, cpu_memory_, stream_buffer_}, device{device_} {
-    if (!device.HasFastBufferSubData()) {
-        return;
+void BufferCacheRuntime::BindIndexBuffer(Buffer& buffer, u32 offset, u32 size) {
+    if (has_unified_vertex_buffers) {
+        buffer.MakeResident(GL_READ_ONLY);
+        glBufferAddressRangeNV(GL_ELEMENT_ARRAY_ADDRESS_NV, 0, buffer.HostGpuAddr() + offset,
+                               static_cast<GLsizeiptr>(size));
+    } else {
+        glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, buffer.Handle());
+        index_buffer_offset = offset;
     }
+}
 
-    static constexpr GLsizeiptr size = static_cast<GLsizeiptr>(Maxwell::MaxConstBufferSize);
-    glCreateBuffers(static_cast<GLsizei>(std::size(cbufs)), std::data(cbufs));
-    for (const GLuint cbuf : cbufs) {
-        glNamedBufferData(cbuf, size, nullptr, GL_STREAM_DRAW);
+void BufferCacheRuntime::BindVertexBuffer(u32 index, Buffer& buffer, u32 offset, u32 size,
+                                          u32 stride) {
+    if (index >= max_attributes) {
+        return;
+    }
+    if (has_unified_vertex_buffers) {
+        buffer.MakeResident(GL_READ_ONLY);
+        glBindVertexBuffer(index, 0, 0, static_cast<GLsizei>(stride));
+        glBufferAddressRangeNV(GL_VERTEX_ATTRIB_ARRAY_ADDRESS_NV, index,
+                               buffer.HostGpuAddr() + offset, static_cast<GLsizeiptr>(size));
+    } else {
+        glBindVertexBuffer(index, buffer.Handle(), static_cast<GLintptr>(offset),
+                           static_cast<GLsizei>(stride));
     }
 }
 
-OGLBufferCache::~OGLBufferCache() {
-    glDeleteBuffers(static_cast<GLsizei>(std::size(cbufs)), std::data(cbufs));
+void BufferCacheRuntime::BindUniformBuffer(size_t stage, u32 binding_index, Buffer& buffer,
+                                           u32 offset, u32 size) {
+    if (use_assembly_shaders) {
+        const GLuint sub_buffer = buffer.SubBuffer(offset);
+        glBindBufferRangeNV(PABO_LUT[stage], binding_index, sub_buffer, 0,
+                            static_cast<GLsizeiptr>(size));
+    } else {
+        const GLuint base_binding = device.GetBaseBindings(stage).uniform_buffer;
+        const GLuint binding = base_binding + binding_index;
+        glBindBufferRange(GL_UNIFORM_BUFFER, binding, buffer.Handle(),
+                          static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size));
+    }
 }
 
-std::shared_ptr<Buffer> OGLBufferCache::CreateBlock(VAddr cpu_addr, std::size_t size) {
-    return std::make_shared<Buffer>(device, cpu_addr, size);
+void BufferCacheRuntime::BindComputeUniformBuffer(u32 binding_index, Buffer& buffer, u32 offset,
+                                                  u32 size) {
+    if (use_assembly_shaders) {
+        glBindBufferRangeNV(GL_COMPUTE_PROGRAM_PARAMETER_BUFFER_NV, binding_index,
+                            buffer.SubBuffer(offset), 0, static_cast<GLsizeiptr>(size));
+    } else {
+        glBindBufferRange(GL_UNIFORM_BUFFER, binding_index, buffer.Handle(),
+                          static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size));
+    }
 }
 
-OGLBufferCache::BufferInfo OGLBufferCache::GetEmptyBuffer(std::size_t) {
-    return {0, 0, 0};
+void BufferCacheRuntime::BindStorageBuffer(size_t stage, u32 binding_index, Buffer& buffer,
+                                           u32 offset, u32 size, bool is_written) {
+    if (use_assembly_shaders) {
+        const BindlessSSBO ssbo{
+            .address = buffer.HostGpuAddr() + offset,
+            .length = static_cast<GLsizei>(size),
+            .padding = 0,
+        };
+        buffer.MakeResident(is_written ? GL_READ_WRITE : GL_READ_ONLY);
+        glProgramLocalParametersI4uivNV(PROGRAM_LUT[stage], binding_index, 1,
+                                        reinterpret_cast<const GLuint*>(&ssbo));
+    } else {
+        const GLuint base_binding = device.GetBaseBindings(stage).shader_storage_buffer;
+        const GLuint binding = base_binding + binding_index;
+        glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding, buffer.Handle(),
+                          static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size));
+    }
 }
 
-OGLBufferCache::BufferInfo OGLBufferCache::ConstBufferUpload(const void* raw_pointer,
-                                                             std::size_t size) {
-    DEBUG_ASSERT(cbuf_cursor < std::size(cbufs));
-    const GLuint cbuf = cbufs[cbuf_cursor++];
+void BufferCacheRuntime::BindComputeStorageBuffer(u32 binding_index, Buffer& buffer, u32 offset,
+                                                  u32 size, bool is_written) {
+    if (use_assembly_shaders) {
+        const BindlessSSBO ssbo{
+            .address = buffer.HostGpuAddr() + offset,
+            .length = static_cast<GLsizei>(size),
+            .padding = 0,
+        };
+        buffer.MakeResident(is_written ? GL_READ_WRITE : GL_READ_ONLY);
+        glProgramLocalParametersI4uivNV(GL_COMPUTE_PROGRAM_NV, binding_index, 1,
+                                        reinterpret_cast<const GLuint*>(&ssbo));
+    } else if (size == 0) {
+        glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding_index, 0, 0, 0);
+    } else {
+        glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding_index, buffer.Handle(),
+                          static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size));
+    }
+}
 
-    glNamedBufferSubData(cbuf, 0, static_cast<GLsizeiptr>(size), raw_pointer);
-    return {cbuf, 0, 0};
+void BufferCacheRuntime::BindTransformFeedbackBuffer(u32 index, Buffer& buffer, u32 offset,
+                                                     u32 size) {
+    glBindBufferRange(GL_TRANSFORM_FEEDBACK_BUFFER, index, buffer.Handle(),
+                      static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size));
 }
 
 } // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.h b/src/video_core/renderer_opengl/gl_buffer_cache.h
index 17ee90316..f4d8871a9 100644
--- a/src/video_core/renderer_opengl/gl_buffer_cache.h
+++ b/src/video_core/renderer_opengl/gl_buffer_cache.h
@@ -5,79 +5,167 @@
 #pragma once
 
 #include <array>
-#include <memory>
+#include <span>
 
+#include "common/alignment.h"
 #include "common/common_types.h"
+#include "common/dynamic_library.h"
 #include "video_core/buffer_cache/buffer_cache.h"
-#include "video_core/engines/maxwell_3d.h"
+#include "video_core/rasterizer_interface.h"
+#include "video_core/renderer_opengl/gl_device.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
 #include "video_core/renderer_opengl/gl_stream_buffer.h"
+#include "video_core/vulkan_common/vulkan_device.h"
+#include "video_core/vulkan_common/vulkan_memory_allocator.h"
 
-namespace Core {
-class System;
-}
+namespace Vulkan {
+class Device;
+class MemoryAllocator;
+} // namespace Vulkan
 
 namespace OpenGL {
 
-class Device;
-class OGLStreamBuffer;
-class RasterizerOpenGL;
-class StateTracker;
+class BufferCacheRuntime;
 
-class Buffer : public VideoCommon::BufferBlock {
+class Buffer : public VideoCommon::BufferBase<VideoCore::RasterizerInterface> {
 public:
-    explicit Buffer(const Device& device_, VAddr cpu_addr_, std::size_t size_);
-    ~Buffer();
+    explicit Buffer(BufferCacheRuntime&, VideoCore::RasterizerInterface& rasterizer, VAddr cpu_addr,
+                    u64 size_bytes);
+    explicit Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams);
 
-    void Upload(std::size_t offset, std::size_t data_size, const u8* data);
+    void ImmediateUpload(size_t offset, std::span<const u8> data) noexcept;
 
-    void Download(std::size_t offset, std::size_t data_size, u8* data);
+    void ImmediateDownload(size_t offset, std::span<u8> data) noexcept;
 
-    void CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst_offset,
-                  std::size_t copy_size);
+    void MakeResident(GLenum access) noexcept;
 
-    GLuint Handle() const noexcept {
-        return gl_buffer.handle;
+    [[nodiscard]] GLuint SubBuffer(u32 offset);
+
+    [[nodiscard]] GLuint64EXT HostGpuAddr() const noexcept {
+        return address;
     }
 
-    u64 Address() const noexcept {
-        return gpu_address;
+    [[nodiscard]] GLuint Handle() const noexcept {
+        return buffer.handle;
     }
 
 private:
-    OGLBuffer gl_buffer;
-    OGLBuffer read_buffer;
-    u64 gpu_address = 0;
+    void CreateMemoryObjects(BufferCacheRuntime& runtime);
+
+    GLuint64EXT address = 0;
+    Vulkan::MemoryCommit memory_commit;
+    OGLBuffer buffer;
+    GLenum current_residency_access = GL_NONE;
+    std::vector<std::pair<OGLBuffer, u32>> subs;
 };
 
-using GenericBufferCache = VideoCommon::BufferCache<Buffer, GLuint, OGLStreamBuffer>;
-class OGLBufferCache final : public GenericBufferCache {
+class BufferCacheRuntime {
+    friend Buffer;
+
 public:
-    explicit OGLBufferCache(VideoCore::RasterizerInterface& rasterizer,
-                            Tegra::MemoryManager& gpu_memory, Core::Memory::Memory& cpu_memory,
-                            const Device& device, OGLStreamBuffer& stream_buffer,
-                            StateTracker& state_tracker);
-    ~OGLBufferCache();
+    static constexpr u8 INVALID_BINDING = std::numeric_limits<u8>::max();
+
+    explicit BufferCacheRuntime(const Device& device_, const Vulkan::Device* vulkan_device_,
+                                Vulkan::MemoryAllocator* vulkan_memory_allocator_);
+
+    void CopyBuffer(Buffer& dst_buffer, Buffer& src_buffer,
+                    std::span<const VideoCommon::BufferCopy> copies);
+
+    void BindIndexBuffer(Buffer& buffer, u32 offset, u32 size);
+
+    void BindVertexBuffer(u32 index, Buffer& buffer, u32 offset, u32 size, u32 stride);
+
+    void BindUniformBuffer(size_t stage, u32 binding_index, Buffer& buffer, u32 offset, u32 size);
+
+    void BindComputeUniformBuffer(u32 binding_index, Buffer& buffer, u32 offset, u32 size);
+
+    void BindStorageBuffer(size_t stage, u32 binding_index, Buffer& buffer, u32 offset, u32 size,
+                           bool is_written);
+
+    void BindComputeStorageBuffer(u32 binding_index, Buffer& buffer, u32 offset, u32 size,
+                                  bool is_written);
 
-    BufferInfo GetEmptyBuffer(std::size_t) override;
+    void BindTransformFeedbackBuffer(u32 index, Buffer& buffer, u32 offset, u32 size);
 
-    void Acquire() noexcept {
-        cbuf_cursor = 0;
+    void BindFastUniformBuffer(size_t stage, u32 binding_index, u32 size) {
+        if (use_assembly_shaders) {
+            const GLuint handle = fast_uniforms[stage][binding_index].handle;
+            const GLsizeiptr gl_size = static_cast<GLsizeiptr>(size);
+            glBindBufferRangeNV(PABO_LUT[stage], binding_index, handle, 0, gl_size);
+        } else {
+            const GLuint base_binding = device.GetBaseBindings(stage).uniform_buffer;
+            const GLuint binding = base_binding + binding_index;
+            glBindBufferRange(GL_UNIFORM_BUFFER, binding,
+                              fast_uniforms[stage][binding_index].handle, 0,
+                              static_cast<GLsizeiptr>(size));
+        }
     }
 
-protected:
-    std::shared_ptr<Buffer> CreateBlock(VAddr cpu_addr, std::size_t size) override;
+    void PushFastUniformBuffer(size_t stage, u32 binding_index, std::span<const u8> data) {
+        if (use_assembly_shaders) {
+            glProgramBufferParametersIuivNV(
+                PABO_LUT[stage], binding_index, 0,
+                static_cast<GLsizei>(data.size_bytes() / sizeof(GLuint)),
+                reinterpret_cast<const GLuint*>(data.data()));
+        } else {
+            glNamedBufferSubData(fast_uniforms[stage][binding_index].handle, 0,
+                                 static_cast<GLsizeiptr>(data.size_bytes()), data.data());
+        }
+    }
+
+    std::span<u8> BindMappedUniformBuffer(size_t stage, u32 binding_index, u32 size) noexcept {
+        const auto [mapped_span, offset] = stream_buffer->Request(static_cast<size_t>(size));
+        const GLuint base_binding = device.GetBaseBindings(stage).uniform_buffer;
+        const GLuint binding = base_binding + binding_index;
+        glBindBufferRange(GL_UNIFORM_BUFFER, binding, stream_buffer->Handle(),
+                          static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size));
+        return mapped_span;
+    }
 
-    BufferInfo ConstBufferUpload(const void* raw_pointer, std::size_t size) override;
+    [[nodiscard]] const GLvoid* IndexOffset() const noexcept {
+        return reinterpret_cast<const GLvoid*>(static_cast<uintptr_t>(index_buffer_offset));
+    }
+
+    [[nodiscard]] bool HasFastBufferSubData() const noexcept {
+        return device.HasFastBufferSubData();
+    }
 
 private:
-    static constexpr std::size_t NUM_CBUFS = Tegra::Engines::Maxwell3D::Regs::MaxConstBuffers *
-                                             Tegra::Engines::Maxwell3D::Regs::MaxShaderProgram;
+    static constexpr std::array PABO_LUT{
+        GL_VERTEX_PROGRAM_PARAMETER_BUFFER_NV,          GL_TESS_CONTROL_PROGRAM_PARAMETER_BUFFER_NV,
+        GL_TESS_EVALUATION_PROGRAM_PARAMETER_BUFFER_NV, GL_GEOMETRY_PROGRAM_PARAMETER_BUFFER_NV,
+        GL_FRAGMENT_PROGRAM_PARAMETER_BUFFER_NV,
+    };
 
     const Device& device;
+    const Vulkan::Device* vulkan_device;
+    Vulkan::MemoryAllocator* vulkan_memory_allocator;
+    std::optional<StreamBuffer> stream_buffer;
+
+    u32 max_attributes = 0;
 
-    std::size_t cbuf_cursor = 0;
-    std::array<GLuint, NUM_CBUFS> cbufs{};
+    bool use_assembly_shaders = false;
+    bool has_unified_vertex_buffers = false;
+
+    std::array<std::array<OGLBuffer, VideoCommon::NUM_GRAPHICS_UNIFORM_BUFFERS>,
+               VideoCommon::NUM_STAGES>
+        fast_uniforms;
+
+    u32 index_buffer_offset = 0;
+};
+
+struct BufferCacheParams {
+    using Runtime = OpenGL::BufferCacheRuntime;
+    using Buffer = OpenGL::Buffer;
+
+    static constexpr bool IS_OPENGL = true;
+    static constexpr bool HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS = true;
+    static constexpr bool HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT = true;
+    static constexpr bool NEEDS_BIND_UNIFORM_INDEX = true;
+    static constexpr bool NEEDS_BIND_STORAGE_INDEX = true;
+    static constexpr bool USE_MEMORY_MAPS = false;
 };
 
+using BufferCache = VideoCommon::BufferCache<BufferCacheParams>;
+
 } // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_device.cpp b/src/video_core/renderer_opengl/gl_device.cpp
index 04c267ee4..0f492f006 100644
--- a/src/video_core/renderer_opengl/gl_device.cpp
+++ b/src/video_core/renderer_opengl/gl_device.cpp
@@ -21,9 +21,7 @@
 #include "video_core/renderer_opengl/gl_resource_manager.h"
 
 namespace OpenGL {
-
 namespace {
-
 // One uniform block is reserved for emulation purposes
 constexpr u32 ReservedUniformBlocks = 1;
 
@@ -197,11 +195,13 @@ bool IsASTCSupported() {
     const bool nsight = std::getenv("NVTX_INJECTION64_PATH") || std::getenv("NSIGHT_LAUNCHED");
     return nsight || HasExtension(extensions, "GL_EXT_debug_tool");
 }
-
 } // Anonymous namespace
 
-Device::Device()
-    : max_uniform_buffers{BuildMaxUniformBuffers()}, base_bindings{BuildBaseBindings()} {
+Device::Device(bool has_vulkan_instance) {
+    if (!GLAD_GL_VERSION_4_6) {
+        LOG_ERROR(Render_OpenGL, "OpenGL 4.6 is not available");
+        throw std::runtime_error{"Insufficient version"};
+    }
     const std::string_view vendor = reinterpret_cast<const char*>(glGetString(GL_VENDOR));
     const std::string_view version = reinterpret_cast<const char*>(glGetString(GL_VERSION));
     const std::vector extensions = GetExtensions();
@@ -217,6 +217,9 @@ Device::Device()
             "Beta driver 443.24 is known to have issues. There might be performance issues.");
         disable_fast_buffer_sub_data = true;
     }
+
+    max_uniform_buffers = BuildMaxUniformBuffers();
+    base_bindings = BuildBaseBindings();
     uniform_buffer_alignment = GetInteger<size_t>(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT);
     shader_storage_alignment = GetInteger<size_t>(GL_SHADER_STORAGE_BUFFER_OFFSET_ALIGNMENT);
     max_vertex_attributes = GetInteger<u32>(GL_MAX_VERTEX_ATTRIBS);
@@ -243,7 +246,8 @@ Device::Device()
 
     use_assembly_shaders = Settings::values.use_assembly_shaders.GetValue() &&
                            GLAD_GL_NV_gpu_program5 && GLAD_GL_NV_compute_program5 &&
-                           GLAD_GL_NV_transform_feedback && GLAD_GL_NV_transform_feedback2;
+                           GLAD_GL_NV_transform_feedback && GLAD_GL_NV_transform_feedback2 &&
+                           has_vulkan_instance;
 
     use_asynchronous_shaders = Settings::values.use_asynchronous_shaders.GetValue();
     use_driver_cache = is_nvidia;
diff --git a/src/video_core/renderer_opengl/gl_device.h b/src/video_core/renderer_opengl/gl_device.h
index 9141de635..eb62ae52d 100644
--- a/src/video_core/renderer_opengl/gl_device.h
+++ b/src/video_core/renderer_opengl/gl_device.h
@@ -10,18 +10,16 @@
 
 namespace OpenGL {
 
-static constexpr u32 EmulationUniformBlockBinding = 0;
-
-class Device final {
+class Device {
 public:
-    struct BaseBindings final {
+    struct BaseBindings {
         u32 uniform_buffer{};
         u32 shader_storage_buffer{};
         u32 sampler{};
         u32 image{};
     };
 
-    explicit Device();
+    explicit Device(bool has_vulkan_instance);
     explicit Device(std::nullptr_t);
 
     u32 GetMaxUniformBuffers(Tegra::Engines::ShaderType shader_type) const noexcept {
diff --git a/src/video_core/renderer_opengl/gl_fence_manager.cpp b/src/video_core/renderer_opengl/gl_fence_manager.cpp
index 3e9c922f5..151290101 100644
--- a/src/video_core/renderer_opengl/gl_fence_manager.cpp
+++ b/src/video_core/renderer_opengl/gl_fence_manager.cpp
@@ -47,7 +47,7 @@ void GLInnerFence::Wait() {
 
 FenceManagerOpenGL::FenceManagerOpenGL(VideoCore::RasterizerInterface& rasterizer_,
                                        Tegra::GPU& gpu_, TextureCache& texture_cache_,
-                                       OGLBufferCache& buffer_cache_, QueryCache& query_cache_)
+                                       BufferCache& buffer_cache_, QueryCache& query_cache_)
     : GenericFenceManager{rasterizer_, gpu_, texture_cache_, buffer_cache_, query_cache_} {}
 
 Fence FenceManagerOpenGL::CreateFence(u32 value, bool is_stubbed) {
diff --git a/src/video_core/renderer_opengl/gl_fence_manager.h b/src/video_core/renderer_opengl/gl_fence_manager.h
index 30dbee613..e714aa115 100644
--- a/src/video_core/renderer_opengl/gl_fence_manager.h
+++ b/src/video_core/renderer_opengl/gl_fence_manager.h
@@ -32,14 +32,13 @@ private:
 };
 
 using Fence = std::shared_ptr<GLInnerFence>;
-using GenericFenceManager =
-    VideoCommon::FenceManager<Fence, TextureCache, OGLBufferCache, QueryCache>;
+using GenericFenceManager = VideoCommon::FenceManager<Fence, TextureCache, BufferCache, QueryCache>;
 
 class FenceManagerOpenGL final : public GenericFenceManager {
 public:
-    explicit FenceManagerOpenGL(VideoCore::RasterizerInterface& rasterizer_, Tegra::GPU& gpu_,
-                                TextureCache& texture_cache_, OGLBufferCache& buffer_cache_,
-                                QueryCache& query_cache_);
+    explicit FenceManagerOpenGL(VideoCore::RasterizerInterface& rasterizer, Tegra::GPU& gpu,
+                                TextureCache& texture_cache, BufferCache& buffer_cache,
+                                QueryCache& query_cache);
 
 protected:
     Fence CreateFence(u32 value, bool is_stubbed) override;
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index ea4ca9a82..52499ee4c 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -44,17 +44,10 @@ using VideoCore::Surface::PixelFormat;
 using VideoCore::Surface::SurfaceTarget;
 using VideoCore::Surface::SurfaceType;
 
-MICROPROFILE_DEFINE(OpenGL_VAO, "OpenGL", "Vertex Format Setup", MP_RGB(128, 128, 192));
-MICROPROFILE_DEFINE(OpenGL_VB, "OpenGL", "Vertex Buffer Setup", MP_RGB(128, 128, 192));
-MICROPROFILE_DEFINE(OpenGL_Shader, "OpenGL", "Shader Setup", MP_RGB(128, 128, 192));
-MICROPROFILE_DEFINE(OpenGL_UBO, "OpenGL", "Const Buffer Setup", MP_RGB(128, 128, 192));
-MICROPROFILE_DEFINE(OpenGL_Index, "OpenGL", "Index Buffer Setup", MP_RGB(128, 128, 192));
-MICROPROFILE_DEFINE(OpenGL_Texture, "OpenGL", "Texture Setup", MP_RGB(128, 128, 192));
-MICROPROFILE_DEFINE(OpenGL_Framebuffer, "OpenGL", "Framebuffer Setup", MP_RGB(128, 128, 192));
 MICROPROFILE_DEFINE(OpenGL_Drawing, "OpenGL", "Drawing", MP_RGB(128, 128, 192));
+MICROPROFILE_DEFINE(OpenGL_Clears, "OpenGL", "Clears", MP_RGB(128, 128, 192));
 MICROPROFILE_DEFINE(OpenGL_Blits, "OpenGL", "Blits", MP_RGB(128, 128, 192));
-MICROPROFILE_DEFINE(OpenGL_CacheManagement, "OpenGL", "Cache Mgmt", MP_RGB(100, 255, 100));
-MICROPROFILE_DEFINE(OpenGL_PrimitiveAssembly, "OpenGL", "Prim Asmbl", MP_RGB(255, 100, 100));
+MICROPROFILE_DEFINE(OpenGL_CacheManagement, "OpenGL", "Cache Management", MP_RGB(100, 255, 100));
 
 namespace {
 
@@ -101,20 +94,6 @@ TextureHandle GetTextureInfo(const Engine& engine, bool via_header_index, const
     return TextureHandle(engine.AccessConstBuffer32(shader_type, buffer, offset), via_header_index);
 }
 
-std::size_t GetConstBufferSize(const Tegra::Engines::ConstBufferInfo& buffer,
-                               const ConstBufferEntry& entry) {
-    if (!entry.IsIndirect()) {
-        return entry.GetSize();
-    }
-    if (buffer.size > Maxwell::MaxConstBufferSize) {
-        LOG_WARNING(Render_OpenGL, "Indirect constbuffer size {} exceeds maximum {}", buffer.size,
-                    Maxwell::MaxConstBufferSize);
-        return Maxwell::MaxConstBufferSize;
-    }
-
-    return buffer.size;
-}
-
 /// Translates hardware transform feedback indices
 /// @param location Hardware location
 /// @return Pair of ARB_transform_feedback3 token stream first and third arguments
@@ -147,14 +126,6 @@ void oglEnable(GLenum cap, bool state) {
     (state ? glEnable : glDisable)(cap);
 }
 
-void UpdateBindlessSSBOs(GLenum target, const BindlessSSBO* ssbos, size_t num_ssbos) {
-    if (num_ssbos == 0) {
-        return;
-    }
-    glProgramLocalParametersI4uivNV(target, 0, static_cast<GLsizei>(num_ssbos),
-                                    reinterpret_cast<const GLuint*>(ssbos));
-}
-
 ImageViewType ImageViewTypeFromEntry(const SamplerEntry& entry) {
     if (entry.is_buffer) {
         return ImageViewType::Buffer;
@@ -196,49 +167,35 @@ ImageViewType ImageViewTypeFromEntry(const ImageEntry& entry) {
 
 RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_,
                                    Core::Memory::Memory& cpu_memory_, const Device& device_,
+                                   const Vulkan::Device* vulkan_device,
+                                   Vulkan::MemoryAllocator* vulkan_memory_allocator,
                                    ScreenInfo& screen_info_, ProgramManager& program_manager_,
                                    StateTracker& state_tracker_)
     : RasterizerAccelerated(cpu_memory_), gpu(gpu_), maxwell3d(gpu.Maxwell3D()),
       kepler_compute(gpu.KeplerCompute()), gpu_memory(gpu.MemoryManager()), device(device_),
       screen_info(screen_info_), program_manager(program_manager_), state_tracker(state_tracker_),
-      stream_buffer(device, state_tracker),
       texture_cache_runtime(device, program_manager, state_tracker),
       texture_cache(texture_cache_runtime, *this, maxwell3d, kepler_compute, gpu_memory),
+      buffer_cache_runtime(device, vulkan_device, vulkan_memory_allocator),
+      buffer_cache(*this, maxwell3d, kepler_compute, gpu_memory, cpu_memory_, buffer_cache_runtime),
       shader_cache(*this, emu_window_, gpu, maxwell3d, kepler_compute, gpu_memory, device),
       query_cache(*this, maxwell3d, gpu_memory),
-      buffer_cache(*this, gpu_memory, cpu_memory_, device, stream_buffer, state_tracker),
       fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache),
       async_shaders(emu_window_) {
-    unified_uniform_buffer.Create();
-    glNamedBufferStorage(unified_uniform_buffer.handle, TOTAL_CONST_BUFFER_BYTES, nullptr, 0);
-
-    if (device.UseAssemblyShaders()) {
-        glCreateBuffers(static_cast<GLsizei>(staging_cbufs.size()), staging_cbufs.data());
-        for (const GLuint cbuf : staging_cbufs) {
-            glNamedBufferStorage(cbuf, static_cast<GLsizeiptr>(Maxwell::MaxConstBufferSize),
-                                 nullptr, 0);
-        }
-    }
     if (device.UseAsynchronousShaders()) {
         async_shaders.AllocateWorkers();
     }
 }
 
-RasterizerOpenGL::~RasterizerOpenGL() {
-    if (device.UseAssemblyShaders()) {
-        glDeleteBuffers(static_cast<GLsizei>(staging_cbufs.size()), staging_cbufs.data());
-    }
-}
+RasterizerOpenGL::~RasterizerOpenGL() = default;
 
-void RasterizerOpenGL::SetupVertexFormat() {
+void RasterizerOpenGL::SyncVertexFormats() {
     auto& flags = maxwell3d.dirty.flags;
     if (!flags[Dirty::VertexFormats]) {
         return;
     }
     flags[Dirty::VertexFormats] = false;
 
-    MICROPROFILE_SCOPE(OpenGL_VAO);
-
     // Use the vertex array as-is, assumes that the data is formatted correctly for OpenGL. Enables
     // the first 16 vertex attributes always, as we don't know which ones are actually used until
     // shader time. Note, Tegra technically supports 32, but we're capping this to 16 for now to
@@ -274,55 +231,7 @@ void RasterizerOpenGL::SetupVertexFormat() {
     }
 }
 
-void RasterizerOpenGL::SetupVertexBuffer() {
-    auto& flags = maxwell3d.dirty.flags;
-    if (!flags[Dirty::VertexBuffers]) {
-        return;
-    }
-    flags[Dirty::VertexBuffers] = false;
-
-    MICROPROFILE_SCOPE(OpenGL_VB);
-
-    const bool use_unified_memory = device.HasVertexBufferUnifiedMemory();
-
-    // Upload all guest vertex arrays sequentially to our buffer
-    const auto& regs = maxwell3d.regs;
-    for (std::size_t index = 0; index < NUM_SUPPORTED_VERTEX_BINDINGS; ++index) {
-        if (!flags[Dirty::VertexBuffer0 + index]) {
-            continue;
-        }
-        flags[Dirty::VertexBuffer0 + index] = false;
-
-        const auto& vertex_array = regs.vertex_array[index];
-        if (!vertex_array.IsEnabled()) {
-            continue;
-        }
-
-        const GPUVAddr start = vertex_array.StartAddress();
-        const GPUVAddr end = regs.vertex_array_limit[index].LimitAddress();
-        ASSERT(end >= start);
-
-        const GLuint gl_index = static_cast<GLuint>(index);
-        const u64 size = end - start;
-        if (size == 0) {
-            glBindVertexBuffer(gl_index, 0, 0, vertex_array.stride);
-            if (use_unified_memory) {
-                glBufferAddressRangeNV(GL_VERTEX_ATTRIB_ARRAY_ADDRESS_NV, gl_index, 0, 0);
-            }
-            continue;
-        }
-        const auto info = buffer_cache.UploadMemory(start, size);
-        if (use_unified_memory) {
-            glBindVertexBuffer(gl_index, 0, 0, vertex_array.stride);
-            glBufferAddressRangeNV(GL_VERTEX_ATTRIB_ARRAY_ADDRESS_NV, gl_index,
-                                   info.address + info.offset, size);
-        } else {
-            glBindVertexBuffer(gl_index, info.handle, info.offset, vertex_array.stride);
-        }
-    }
-}
-
-void RasterizerOpenGL::SetupVertexInstances() {
+void RasterizerOpenGL::SyncVertexInstances() {
     auto& flags = maxwell3d.dirty.flags;
     if (!flags[Dirty::VertexInstances]) {
         return;
@@ -343,17 +252,7 @@ void RasterizerOpenGL::SetupVertexInstances() {
     }
 }
 
-GLintptr RasterizerOpenGL::SetupIndexBuffer() {
-    MICROPROFILE_SCOPE(OpenGL_Index);
-    const auto& regs = maxwell3d.regs;
-    const std::size_t size = CalculateIndexBufferSize();
-    const auto info = buffer_cache.UploadMemory(regs.index_array.IndexStart(), size);
-    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, info.handle);
-    return info.offset;
-}
-
-void RasterizerOpenGL::SetupShaders() {
-    MICROPROFILE_SCOPE(OpenGL_Shader);
+void RasterizerOpenGL::SetupShaders(bool is_indexed) {
     u32 clip_distances = 0;
 
     std::array<Shader*, Maxwell::MaxShaderStage> shaders{};
@@ -410,11 +309,19 @@ void RasterizerOpenGL::SetupShaders() {
         const size_t stage = index == 0 ? 0 : index - 1;
         shaders[stage] = shader;
 
-        SetupDrawConstBuffers(stage, shader);
-        SetupDrawGlobalMemory(stage, shader);
         SetupDrawTextures(shader, stage);
         SetupDrawImages(shader, stage);
 
+        buffer_cache.SetEnabledUniformBuffers(stage, shader->GetEntries().enabled_uniform_buffers);
+
+        buffer_cache.UnbindGraphicsStorageBuffers(stage);
+        u32 ssbo_index = 0;
+        for (const auto& buffer : shader->GetEntries().global_memory_entries) {
+            buffer_cache.BindGraphicsStorageBuffer(stage, ssbo_index, buffer.cbuf_index,
+                                                   buffer.cbuf_offset, buffer.is_written);
+            ++ssbo_index;
+        }
+
         // Workaround for Intel drivers.
         // When a clip distance is enabled but not set in the shader it crops parts of the screen
         // (sometimes it's half the screen, sometimes three quarters). To avoid this, enable the
@@ -430,43 +337,26 @@ void RasterizerOpenGL::SetupShaders() {
     SyncClipEnabled(clip_distances);
     maxwell3d.dirty.flags[Dirty::Shaders] = false;
 
+    buffer_cache.UpdateGraphicsBuffers(is_indexed);
+
     const std::span indices_span(image_view_indices.data(), image_view_indices.size());
     texture_cache.FillGraphicsImageViews(indices_span, image_view_ids);
 
+    buffer_cache.BindHostGeometryBuffers(is_indexed);
+
     size_t image_view_index = 0;
     size_t texture_index = 0;
     size_t image_index = 0;
     for (size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) {
         const Shader* const shader = shaders[stage];
-        if (shader) {
-            const auto base = device.GetBaseBindings(stage);
-            BindTextures(shader->GetEntries(), base.sampler, base.image, image_view_index,
-                         texture_index, image_index);
-        }
-    }
-}
-
-std::size_t RasterizerOpenGL::CalculateVertexArraysSize() const {
-    const auto& regs = maxwell3d.regs;
-
-    std::size_t size = 0;
-    for (u32 index = 0; index < Maxwell::NumVertexArrays; ++index) {
-        if (!regs.vertex_array[index].IsEnabled())
+        if (!shader) {
             continue;
-
-        const GPUVAddr start = regs.vertex_array[index].StartAddress();
-        const GPUVAddr end = regs.vertex_array_limit[index].LimitAddress();
-
-        size += end - start;
-        ASSERT(end >= start);
+        }
+        buffer_cache.BindHostStageBuffers(stage);
+        const auto& base = device.GetBaseBindings(stage);
+        BindTextures(shader->GetEntries(), base.sampler, base.image, image_view_index,
+                     texture_index, image_index);
     }
-
-    return size;
-}
-
-std::size_t RasterizerOpenGL::CalculateIndexBufferSize() const {
-    return static_cast<std::size_t>(maxwell3d.regs.index_array.count) *
-           static_cast<std::size_t>(maxwell3d.regs.index_array.FormatSizeInBytes());
 }
 
 void RasterizerOpenGL::LoadDiskResources(u64 title_id, const std::atomic_bool& stop_loading,
@@ -475,6 +365,7 @@ void RasterizerOpenGL::LoadDiskResources(u64 title_id, const std::atomic_bool& s
 }
 
 void RasterizerOpenGL::Clear() {
+    MICROPROFILE_SCOPE(OpenGL_Clears);
     if (!maxwell3d.ShouldExecute()) {
         return;
     }
@@ -525,11 +416,9 @@ void RasterizerOpenGL::Clear() {
     }
     UNIMPLEMENTED_IF(regs.clear_flags.viewport);
 
-    {
-        auto lock = texture_cache.AcquireLock();
-        texture_cache.UpdateRenderTargets(true);
-        state_tracker.BindFramebuffer(texture_cache.GetFramebuffer()->Handle());
-    }
+    std::scoped_lock lock{texture_cache.mutex};
+    texture_cache.UpdateRenderTargets(true);
+    state_tracker.BindFramebuffer(texture_cache.GetFramebuffer()->Handle());
 
     if (use_color) {
         glClearBufferfv(GL_COLOR, regs.clear_buffers.RT, regs.clear_color);
@@ -541,7 +430,6 @@ void RasterizerOpenGL::Clear() {
     } else if (use_stencil) {
         glClearBufferiv(GL_STENCIL, 0, &regs.clear_stencil);
     }
-
     ++num_queued_commands;
 }
 
@@ -550,75 +438,12 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
 
     query_cache.UpdateCounters();
 
-    SyncViewport();
-    SyncRasterizeEnable();
-    SyncPolygonModes();
-    SyncColorMask();
-    SyncFragmentColorClampState();
-    SyncMultiSampleState();
-    SyncDepthTestState();
-    SyncDepthClamp();
-    SyncStencilTestState();
-    SyncBlendState();
-    SyncLogicOpState();
-    SyncCullMode();
-    SyncPrimitiveRestart();
-    SyncScissorTest();
-    SyncPointState();
-    SyncLineState();
-    SyncPolygonOffset();
-    SyncAlphaTest();
-    SyncFramebufferSRGB();
-
-    buffer_cache.Acquire();
-    current_cbuf = 0;
-
-    std::size_t buffer_size = CalculateVertexArraysSize();
-
-    // Add space for index buffer
-    if (is_indexed) {
-        buffer_size = Common::AlignUp(buffer_size, 4) + CalculateIndexBufferSize();
-    }
-
-    // Uniform space for the 5 shader stages
-    buffer_size =
-        Common::AlignUp<std::size_t>(buffer_size, 4) +
-        (sizeof(MaxwellUniformData) + device.GetUniformBufferAlignment()) * Maxwell::MaxShaderStage;
-
-    // Add space for at least 18 constant buffers
-    buffer_size += Maxwell::MaxConstBuffers *
-                   (Maxwell::MaxConstBufferSize + device.GetUniformBufferAlignment());
-
-    // Prepare the vertex array.
-    buffer_cache.Map(buffer_size);
-
-    // Prepare vertex array format.
-    SetupVertexFormat();
-
-    // Upload vertex and index data.
-    SetupVertexBuffer();
-    SetupVertexInstances();
-    GLintptr index_buffer_offset = 0;
-    if (is_indexed) {
-        index_buffer_offset = SetupIndexBuffer();
-    }
-
-    // Setup emulation uniform buffer.
-    if (!device.UseAssemblyShaders()) {
-        MaxwellUniformData ubo;
-        ubo.SetFromRegs(maxwell3d);
-        const auto info =
-            buffer_cache.UploadHostMemory(&ubo, sizeof(ubo), device.GetUniformBufferAlignment());
-        glBindBufferRange(GL_UNIFORM_BUFFER, EmulationUniformBlockBinding, info.handle, info.offset,
-                          static_cast<GLsizeiptr>(sizeof(ubo)));
-    }
+    SyncState();
 
     // Setup shaders and their used resources.
-    auto lock = texture_cache.AcquireLock();
-    SetupShaders();
+    std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
+    SetupShaders(is_indexed);
 
-    // Signal the buffer cache that we are not going to upload more things.
-    buffer_cache.Unmap();
     texture_cache.UpdateRenderTargets(false);
     state_tracker.BindFramebuffer(texture_cache.GetFramebuffer()->Handle());
     program_manager.BindGraphicsPipeline();
@@ -632,7 +457,7 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
     if (is_indexed) {
         const GLint base_vertex = static_cast<GLint>(maxwell3d.regs.vb_element_base);
         const GLsizei num_vertices = static_cast<GLsizei>(maxwell3d.regs.index_array.count);
-        const GLvoid* offset = reinterpret_cast<const GLvoid*>(index_buffer_offset);
+        const GLvoid* const offset = buffer_cache_runtime.IndexOffset();
         const GLenum format = MaxwellToGL::IndexFormat(maxwell3d.regs.index_array.format);
         if (num_instances == 1 && base_instance == 0 && base_vertex == 0) {
             glDrawElements(primitive_mode, num_vertices, format, offset);
@@ -672,22 +497,22 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
 }
 
 void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) {
-    buffer_cache.Acquire();
-    current_cbuf = 0;
-
     Shader* const kernel = shader_cache.GetComputeKernel(code_addr);
 
-    auto lock = texture_cache.AcquireLock();
+    std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
     BindComputeTextures(kernel);
 
-    const size_t buffer_size = Tegra::Engines::KeplerCompute::NumConstBuffers *
-                               (Maxwell::MaxConstBufferSize + device.GetUniformBufferAlignment());
-    buffer_cache.Map(buffer_size);
-
-    SetupComputeConstBuffers(kernel);
-    SetupComputeGlobalMemory(kernel);
-
-    buffer_cache.Unmap();
+    const auto& entries = kernel->GetEntries();
+    buffer_cache.SetEnabledComputeUniformBuffers(entries.enabled_uniform_buffers);
+    buffer_cache.UnbindComputeStorageBuffers();
+    u32 ssbo_index = 0;
+    for (const auto& buffer : entries.global_memory_entries) {
+        buffer_cache.BindComputeStorageBuffer(ssbo_index, buffer.cbuf_index, buffer.cbuf_offset,
+                                              buffer.is_written);
+        ++ssbo_index;
+    }
+    buffer_cache.UpdateComputeBuffers();
+    buffer_cache.BindHostComputeBuffers();
 
     const auto& launch_desc = kepler_compute.launch_description;
     glDispatchCompute(launch_desc.grid_dim_x, launch_desc.grid_dim_y, launch_desc.grid_dim_z);
@@ -703,6 +528,12 @@ void RasterizerOpenGL::Query(GPUVAddr gpu_addr, VideoCore::QueryType type,
     query_cache.Query(gpu_addr, type, timestamp);
 }
 
+void RasterizerOpenGL::BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr,
+                                                 u32 size) {
+    std::scoped_lock lock{buffer_cache.mutex};
+    buffer_cache.BindGraphicsUniformBuffer(stage, index, gpu_addr, size);
+}
+
 void RasterizerOpenGL::FlushAll() {}
 
 void RasterizerOpenGL::FlushRegion(VAddr addr, u64 size) {
@@ -711,19 +542,23 @@ void RasterizerOpenGL::FlushRegion(VAddr addr, u64 size) {
         return;
     }
     {
-        auto lock = texture_cache.AcquireLock();
+        std::scoped_lock lock{texture_cache.mutex};
         texture_cache.DownloadMemory(addr, size);
     }
-    buffer_cache.FlushRegion(addr, size);
+    {
+        std::scoped_lock lock{buffer_cache.mutex};
+        buffer_cache.DownloadMemory(addr, size);
+    }
     query_cache.FlushRegion(addr, size);
 }
 
 bool RasterizerOpenGL::MustFlushRegion(VAddr addr, u64 size) {
+    std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
     if (!Settings::IsGPULevelHigh()) {
-        return buffer_cache.MustFlushRegion(addr, size);
+        return buffer_cache.IsRegionGpuModified(addr, size);
     }
     return texture_cache.IsRegionGpuModified(addr, size) ||
-           buffer_cache.MustFlushRegion(addr, size);
+           buffer_cache.IsRegionGpuModified(addr, size);
 }
 
 void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size) {
@@ -732,11 +567,14 @@ void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size) {
         return;
     }
     {
-        auto lock = texture_cache.AcquireLock();
+        std::scoped_lock lock{texture_cache.mutex};
         texture_cache.WriteMemory(addr, size);
     }
+    {
+        std::scoped_lock lock{buffer_cache.mutex};
+        buffer_cache.WriteMemory(addr, size);
+    }
     shader_cache.InvalidateRegion(addr, size);
-    buffer_cache.InvalidateRegion(addr, size);
     query_cache.InvalidateRegion(addr, size);
 }
 
@@ -745,26 +583,35 @@ void RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) {
     if (addr == 0 || size == 0) {
         return;
     }
+    shader_cache.OnCPUWrite(addr, size);
     {
-        auto lock = texture_cache.AcquireLock();
+        std::scoped_lock lock{texture_cache.mutex};
         texture_cache.WriteMemory(addr, size);
     }
-    shader_cache.OnCPUWrite(addr, size);
-    buffer_cache.OnCPUWrite(addr, size);
+    {
+        std::scoped_lock lock{buffer_cache.mutex};
+        buffer_cache.CachedWriteMemory(addr, size);
+    }
 }
 
 void RasterizerOpenGL::SyncGuestHost() {
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
-    buffer_cache.SyncGuestHost();
     shader_cache.SyncGuestHost();
+    {
+        std::scoped_lock lock{buffer_cache.mutex};
+        buffer_cache.FlushCachedWrites();
+    }
 }
 
 void RasterizerOpenGL::UnmapMemory(VAddr addr, u64 size) {
     {
-        auto lock = texture_cache.AcquireLock();
+        std::scoped_lock lock{texture_cache.mutex};
         texture_cache.UnmapMemory(addr, size);
     }
-    buffer_cache.OnCPUWrite(addr, size);
+    {
+        std::scoped_lock lock{buffer_cache.mutex};
+        buffer_cache.WriteMemory(addr, size);
+    }
     shader_cache.OnCPUWrite(addr, size);
 }
 
@@ -799,14 +646,7 @@ void RasterizerOpenGL::FlushAndInvalidateRegion(VAddr addr, u64 size) {
 }
 
 void RasterizerOpenGL::WaitForIdle() {
-    // Place a barrier on everything that is not framebuffer related.
-    // This is related to another flag that is not currently implemented.
-    glMemoryBarrier(GL_VERTEX_ATTRIB_ARRAY_BARRIER_BIT | GL_ELEMENT_ARRAY_BARRIER_BIT |
-                    GL_UNIFORM_BARRIER_BIT | GL_TEXTURE_FETCH_BARRIER_BIT |
-                    GL_SHADER_IMAGE_ACCESS_BARRIER_BIT | GL_COMMAND_BARRIER_BIT |
-                    GL_PIXEL_BUFFER_BARRIER_BIT | GL_TEXTURE_UPDATE_BARRIER_BIT |
-                    GL_BUFFER_UPDATE_BARRIER_BIT | GL_TRANSFORM_FEEDBACK_BARRIER_BIT |
-                    GL_SHADER_STORAGE_BARRIER_BIT | GL_QUERY_BUFFER_BARRIER_BIT);
+    glMemoryBarrier(GL_ALL_BARRIER_BITS);
 }
 
 void RasterizerOpenGL::FragmentBarrier() {
@@ -831,18 +671,21 @@ void RasterizerOpenGL::TickFrame() {
     num_queued_commands = 0;
 
     fence_manager.TickFrame();
-    buffer_cache.TickFrame();
     {
-        auto lock = texture_cache.AcquireLock();
+        std::scoped_lock lock{texture_cache.mutex};
         texture_cache.TickFrame();
     }
+    {
+        std::scoped_lock lock{buffer_cache.mutex};
+        buffer_cache.TickFrame();
+    }
 }
 
 bool RasterizerOpenGL::AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Surface& src,
                                              const Tegra::Engines::Fermi2D::Surface& dst,
                                              const Tegra::Engines::Fermi2D::Config& copy_config) {
     MICROPROFILE_SCOPE(OpenGL_Blits);
-    auto lock = texture_cache.AcquireLock();
+    std::scoped_lock lock{texture_cache.mutex};
     texture_cache.BlitImage(dst, src, copy_config);
     return true;
 }
@@ -854,7 +697,7 @@ bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config,
     }
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
 
-    auto lock = texture_cache.AcquireLock();
+    std::scoped_lock lock{texture_cache.mutex};
     ImageView* const image_view{texture_cache.TryFindFramebufferImageView(framebuffer_addr)};
     if (!image_view) {
         return false;
@@ -921,166 +764,6 @@ void RasterizerOpenGL::BindTextures(const ShaderEntries& entries, GLuint base_te
     }
 }
 
-void RasterizerOpenGL::SetupDrawConstBuffers(std::size_t stage_index, Shader* shader) {
-    static constexpr std::array PARAMETER_LUT{
-        GL_VERTEX_PROGRAM_PARAMETER_BUFFER_NV,          GL_TESS_CONTROL_PROGRAM_PARAMETER_BUFFER_NV,
-        GL_TESS_EVALUATION_PROGRAM_PARAMETER_BUFFER_NV, GL_GEOMETRY_PROGRAM_PARAMETER_BUFFER_NV,
-        GL_FRAGMENT_PROGRAM_PARAMETER_BUFFER_NV,
-    };
-    MICROPROFILE_SCOPE(OpenGL_UBO);
-    const auto& stages = maxwell3d.state.shader_stages;
-    const auto& shader_stage = stages[stage_index];
-    const auto& entries = shader->GetEntries();
-    const bool use_unified = entries.use_unified_uniforms;
-    const std::size_t base_unified_offset = stage_index * NUM_CONST_BUFFERS_BYTES_PER_STAGE;
-
-    const auto base_bindings = device.GetBaseBindings(stage_index);
-    u32 binding = device.UseAssemblyShaders() ? 0 : base_bindings.uniform_buffer;
-    for (const auto& entry : entries.const_buffers) {
-        const u32 index = entry.GetIndex();
-        const auto& buffer = shader_stage.const_buffers[index];
-        SetupConstBuffer(PARAMETER_LUT[stage_index], binding, buffer, entry, use_unified,
-                         base_unified_offset + index * Maxwell::MaxConstBufferSize);
-        ++binding;
-    }
-    if (use_unified) {
-        const u32 index = static_cast<u32>(base_bindings.shader_storage_buffer +
-                                           entries.global_memory_entries.size());
-        glBindBufferRange(GL_SHADER_STORAGE_BUFFER, index, unified_uniform_buffer.handle,
-                          base_unified_offset, NUM_CONST_BUFFERS_BYTES_PER_STAGE);
-    }
-}
-
-void RasterizerOpenGL::SetupComputeConstBuffers(Shader* kernel) {
-    MICROPROFILE_SCOPE(OpenGL_UBO);
-    const auto& launch_desc = kepler_compute.launch_description;
-    const auto& entries = kernel->GetEntries();
-    const bool use_unified = entries.use_unified_uniforms;
-
-    u32 binding = 0;
-    for (const auto& entry : entries.const_buffers) {
-        const auto& config = launch_desc.const_buffer_config[entry.GetIndex()];
-        const std::bitset<8> mask = launch_desc.const_buffer_enable_mask.Value();
-        Tegra::Engines::ConstBufferInfo buffer;
-        buffer.address = config.Address();
-        buffer.size = config.size;
-        buffer.enabled = mask[entry.GetIndex()];
-        SetupConstBuffer(GL_COMPUTE_PROGRAM_PARAMETER_BUFFER_NV, binding, buffer, entry,
-                         use_unified, entry.GetIndex() * Maxwell::MaxConstBufferSize);
-        ++binding;
-    }
-    if (use_unified) {
-        const GLuint index = static_cast<GLuint>(entries.global_memory_entries.size());
-        glBindBufferRange(GL_SHADER_STORAGE_BUFFER, index, unified_uniform_buffer.handle, 0,
-                          NUM_CONST_BUFFERS_BYTES_PER_STAGE);
-    }
-}
-
-void RasterizerOpenGL::SetupConstBuffer(GLenum stage, u32 binding,
-                                        const Tegra::Engines::ConstBufferInfo& buffer,
-                                        const ConstBufferEntry& entry, bool use_unified,
-                                        std::size_t unified_offset) {
-    if (!buffer.enabled) {
-        // Set values to zero to unbind buffers
-        if (device.UseAssemblyShaders()) {
-            glBindBufferRangeNV(stage, entry.GetIndex(), 0, 0, 0);
-        } else {
-            glBindBufferRange(GL_UNIFORM_BUFFER, binding, 0, 0, sizeof(float));
-        }
-        return;
-    }
-
-    // Align the actual size so it ends up being a multiple of vec4 to meet the OpenGL std140
-    // UBO alignment requirements.
-    const std::size_t size = Common::AlignUp(GetConstBufferSize(buffer, entry), sizeof(GLvec4));
-
-    const bool fast_upload = !use_unified && device.HasFastBufferSubData();
-
-    const std::size_t alignment = use_unified ? 4 : device.GetUniformBufferAlignment();
-    const GPUVAddr gpu_addr = buffer.address;
-    auto info = buffer_cache.UploadMemory(gpu_addr, size, alignment, false, fast_upload);
-
-    if (device.UseAssemblyShaders()) {
-        UNIMPLEMENTED_IF(use_unified);
-        if (info.offset != 0) {
-            const GLuint staging_cbuf = staging_cbufs[current_cbuf++];
-            glCopyNamedBufferSubData(info.handle, staging_cbuf, info.offset, 0, size);
-            info.handle = staging_cbuf;
-            info.offset = 0;
-        }
-        glBindBufferRangeNV(stage, binding, info.handle, info.offset, size);
-        return;
-    }
-
-    if (use_unified) {
-        glCopyNamedBufferSubData(info.handle, unified_uniform_buffer.handle, info.offset,
-                                 unified_offset, size);
-    } else {
-        glBindBufferRange(GL_UNIFORM_BUFFER, binding, info.handle, info.offset, size);
-    }
-}
-
-void RasterizerOpenGL::SetupDrawGlobalMemory(std::size_t stage_index, Shader* shader) {
-    static constexpr std::array TARGET_LUT = {
-        GL_VERTEX_PROGRAM_NV,   GL_TESS_CONTROL_PROGRAM_NV, GL_TESS_EVALUATION_PROGRAM_NV,
-        GL_GEOMETRY_PROGRAM_NV, GL_FRAGMENT_PROGRAM_NV,
-    };
-    const auto& cbufs{maxwell3d.state.shader_stages[stage_index]};
-    const auto& entries{shader->GetEntries().global_memory_entries};
-
-    std::array<BindlessSSBO, 32> ssbos;
-    ASSERT(entries.size() < ssbos.size());
-
-    const bool assembly_shaders = device.UseAssemblyShaders();
-    u32 binding = assembly_shaders ? 0 : device.GetBaseBindings(stage_index).shader_storage_buffer;
-    for (const auto& entry : entries) {
-        const GPUVAddr addr{cbufs.const_buffers[entry.cbuf_index].address + entry.cbuf_offset};
-        const GPUVAddr gpu_addr{gpu_memory.Read<u64>(addr)};
-        const u32 size{gpu_memory.Read<u32>(addr + 8)};
-        SetupGlobalMemory(binding, entry, gpu_addr, size, &ssbos[binding]);
-        ++binding;
-    }
-    if (assembly_shaders) {
-        UpdateBindlessSSBOs(TARGET_LUT[stage_index], ssbos.data(), entries.size());
-    }
-}
-
-void RasterizerOpenGL::SetupComputeGlobalMemory(Shader* kernel) {
-    const auto& cbufs{kepler_compute.launch_description.const_buffer_config};
-    const auto& entries{kernel->GetEntries().global_memory_entries};
-
-    std::array<BindlessSSBO, 32> ssbos;
-    ASSERT(entries.size() < ssbos.size());
-
-    u32 binding = 0;
-    for (const auto& entry : entries) {
-        const GPUVAddr addr{cbufs[entry.cbuf_index].Address() + entry.cbuf_offset};
-        const GPUVAddr gpu_addr{gpu_memory.Read<u64>(addr)};
-        const u32 size{gpu_memory.Read<u32>(addr + 8)};
-        SetupGlobalMemory(binding, entry, gpu_addr, size, &ssbos[binding]);
-        ++binding;
-    }
-    if (device.UseAssemblyShaders()) {
-        UpdateBindlessSSBOs(GL_COMPUTE_PROGRAM_NV, ssbos.data(), ssbos.size());
-    }
-}
-
-void RasterizerOpenGL::SetupGlobalMemory(u32 binding, const GlobalMemoryEntry& entry,
-                                         GPUVAddr gpu_addr, size_t size, BindlessSSBO* ssbo) {
-    const size_t alignment{device.GetShaderStorageBufferAlignment()};
-    const auto info = buffer_cache.UploadMemory(gpu_addr, size, alignment, entry.is_written);
-    if (device.UseAssemblyShaders()) {
-        *ssbo = BindlessSSBO{
-            .address = static_cast<GLuint64EXT>(info.address + info.offset),
-            .length = static_cast<GLsizei>(size),
-            .padding = 0,
-        };
-    } else {
-        glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding, info.handle, info.offset,
-                          static_cast<GLsizeiptr>(size));
-    }
-}
-
 void RasterizerOpenGL::SetupDrawTextures(const Shader* shader, size_t stage_index) {
     const bool via_header_index =
         maxwell3d.regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex;
@@ -1128,6 +811,30 @@ void RasterizerOpenGL::SetupComputeImages(const Shader* shader) {
     }
 }
 
+void RasterizerOpenGL::SyncState() {
+    SyncViewport();
+    SyncRasterizeEnable();
+    SyncPolygonModes();
+    SyncColorMask();
+    SyncFragmentColorClampState();
+    SyncMultiSampleState();
+    SyncDepthTestState();
+    SyncDepthClamp();
+    SyncStencilTestState();
+    SyncBlendState();
+    SyncLogicOpState();
+    SyncCullMode();
+    SyncPrimitiveRestart();
+    SyncScissorTest();
+    SyncPointState();
+    SyncLineState();
+    SyncPolygonOffset();
+    SyncAlphaTest();
+    SyncFramebufferSRGB();
+    SyncVertexFormats();
+    SyncVertexInstances();
+}
+
 void RasterizerOpenGL::SyncViewport() {
     auto& flags = maxwell3d.dirty.flags;
     const auto& regs = maxwell3d.regs;
@@ -1163,9 +870,11 @@ void RasterizerOpenGL::SyncViewport() {
         if (regs.screen_y_control.y_negate != 0) {
             flip_y = !flip_y;
         }
-        glClipControl(flip_y ? GL_UPPER_LEFT : GL_LOWER_LEFT,
-                      regs.depth_mode == Maxwell::DepthMode::ZeroToOne ? GL_ZERO_TO_ONE
-                                                                       : GL_NEGATIVE_ONE_TO_ONE);
+        const bool is_zero_to_one = regs.depth_mode == Maxwell::DepthMode::ZeroToOne;
+        const GLenum origin = flip_y ? GL_UPPER_LEFT : GL_LOWER_LEFT;
+        const GLenum depth = is_zero_to_one ? GL_ZERO_TO_ONE : GL_NEGATIVE_ONE_TO_ONE;
+        state_tracker.ClipControl(origin, depth);
+        state_tracker.SetYNegate(regs.screen_y_control.y_negate != 0);
     }
 
     if (dirty_viewport) {
@@ -1649,36 +1358,13 @@ void RasterizerOpenGL::BeginTransformFeedback(GLenum primitive_mode) {
     if (regs.tfb_enabled == 0) {
         return;
     }
-
     if (device.UseAssemblyShaders()) {
         SyncTransformFeedback();
     }
-
     UNIMPLEMENTED_IF(regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationControl) ||
                      regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationEval) ||
                      regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::Geometry));
-
-    for (std::size_t index = 0; index < Maxwell::NumTransformFeedbackBuffers; ++index) {
-        const auto& binding = regs.tfb_bindings[index];
-        if (!binding.buffer_enable) {
-            if (enabled_transform_feedback_buffers[index]) {
-                glBindBufferRange(GL_TRANSFORM_FEEDBACK_BUFFER, static_cast<GLuint>(index), 0, 0,
-                                  0);
-            }
-            enabled_transform_feedback_buffers[index] = false;
-            continue;
-        }
-        enabled_transform_feedback_buffers[index] = true;
-
-        auto& tfb_buffer = transform_feedback_buffers[index];
-        tfb_buffer.Create();
-
-        const GLuint handle = tfb_buffer.handle;
-        const std::size_t size = binding.buffer_size;
-        glNamedBufferData(handle, static_cast<GLsizeiptr>(size), nullptr, GL_STREAM_COPY);
-        glBindBufferRange(GL_TRANSFORM_FEEDBACK_BUFFER, static_cast<GLuint>(index), handle, 0,
-                          static_cast<GLsizeiptr>(size));
-    }
+    UNIMPLEMENTED_IF(primitive_mode != GL_POINTS);
 
     // We may have to call BeginTransformFeedbackNV here since they seem to call different
     // implementations on Nvidia's driver (the pointer is different) but we are using
@@ -1692,23 +1378,7 @@ void RasterizerOpenGL::EndTransformFeedback() {
     if (regs.tfb_enabled == 0) {
         return;
     }
-
     glEndTransformFeedback();
-
-    for (std::size_t index = 0; index < Maxwell::NumTransformFeedbackBuffers; ++index) {
-        const auto& binding = regs.tfb_bindings[index];
-        if (!binding.buffer_enable) {
-            continue;
-        }
-        UNIMPLEMENTED_IF(binding.buffer_offset != 0);
-
-        const GLuint handle = transform_feedback_buffers[index].handle;
-        const GPUVAddr gpu_addr = binding.Address();
-        const std::size_t size = binding.buffer_size;
-        const auto info = buffer_cache.UploadMemory(gpu_addr, size, 4, true);
-        glCopyNamedBufferSubData(handle, info.handle, 0, info.offset,
-                                 static_cast<GLsizeiptr>(size));
-    }
 }
 
 } // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h
index 82e03e677..31d69a94c 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer.h
@@ -30,7 +30,6 @@
 #include "video_core/renderer_opengl/gl_shader_decompiler.h"
 #include "video_core/renderer_opengl/gl_shader_manager.h"
 #include "video_core/renderer_opengl/gl_state_tracker.h"
-#include "video_core/renderer_opengl/gl_stream_buffer.h"
 #include "video_core/renderer_opengl/gl_texture_cache.h"
 #include "video_core/shader/async_shaders.h"
 #include "video_core/textures/texture.h"
@@ -47,6 +46,11 @@ namespace Tegra {
 class MemoryManager;
 }
 
+namespace Vulkan {
+class Device;
+class MemoryAllocator;
+} // namespace Vulkan
+
 namespace OpenGL {
 
 struct ScreenInfo;
@@ -63,6 +67,8 @@ class RasterizerOpenGL : public VideoCore::RasterizerAccelerated {
 public:
     explicit RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_,
                               Core::Memory::Memory& cpu_memory_, const Device& device_,
+                              const Vulkan::Device* vulkan_device,
+                              Vulkan::MemoryAllocator* vulkan_memory_allocator,
                               ScreenInfo& screen_info_, ProgramManager& program_manager_,
                               StateTracker& state_tracker_);
     ~RasterizerOpenGL() override;
@@ -72,6 +78,7 @@ public:
     void DispatchCompute(GPUVAddr code_addr) override;
     void ResetCounter(VideoCore::QueryType type) override;
     void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) override;
+    void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size) override;
     void FlushAll() override;
     void FlushRegion(VAddr addr, u64 size) override;
     bool MustFlushRegion(VAddr addr, u64 size) override;
@@ -119,27 +126,6 @@ private:
     void BindTextures(const ShaderEntries& entries, GLuint base_texture, GLuint base_image,
                       size_t& image_view_index, size_t& texture_index, size_t& image_index);
 
-    /// Configures the current constbuffers to use for the draw command.
-    void SetupDrawConstBuffers(std::size_t stage_index, Shader* shader);
-
-    /// Configures the current constbuffers to use for the kernel invocation.
-    void SetupComputeConstBuffers(Shader* kernel);
-
-    /// Configures a constant buffer.
-    void SetupConstBuffer(GLenum stage, u32 binding, const Tegra::Engines::ConstBufferInfo& buffer,
-                          const ConstBufferEntry& entry, bool use_unified,
-                          std::size_t unified_offset);
-
-    /// Configures the current global memory entries to use for the draw command.
-    void SetupDrawGlobalMemory(std::size_t stage_index, Shader* shader);
-
-    /// Configures the current global memory entries to use for the kernel invocation.
-    void SetupComputeGlobalMemory(Shader* kernel);
-
-    /// Configures a global memory buffer.
-    void SetupGlobalMemory(u32 binding, const GlobalMemoryEntry& entry, GPUVAddr gpu_addr,
-                           size_t size, BindlessSSBO* ssbo);
-
     /// Configures the current textures to use for the draw command.
     void SetupDrawTextures(const Shader* shader, size_t stage_index);
 
@@ -152,6 +138,9 @@ private:
     /// Configures images in a compute shader.
     void SetupComputeImages(const Shader* shader);
 
+    /// Syncs state to match guest's
+    void SyncState();
+
     /// Syncs the viewport and depth range to match the guest state
     void SyncViewport();
 
@@ -215,6 +204,12 @@ private:
     /// Syncs the framebuffer sRGB state to match the guest state
     void SyncFramebufferSRGB();
 
+    /// Syncs vertex formats to match the guest state
+    void SyncVertexFormats();
+
+    /// Syncs vertex instances to match the guest state
+    void SyncVertexInstances();
+
     /// Syncs transform feedback state to match guest state
     /// @note Only valid on assembly shaders
     void SyncTransformFeedback();
@@ -225,19 +220,7 @@ private:
     /// End a transform feedback
     void EndTransformFeedback();
 
-    std::size_t CalculateVertexArraysSize() const;
-
-    std::size_t CalculateIndexBufferSize() const;
-
-    /// Updates the current vertex format
-    void SetupVertexFormat();
-
-    void SetupVertexBuffer();
-    void SetupVertexInstances();
-
-    GLintptr SetupIndexBuffer();
-
-    void SetupShaders();
+    void SetupShaders(bool is_indexed);
 
     Tegra::GPU& gpu;
     Tegra::Engines::Maxwell3D& maxwell3d;
@@ -249,12 +232,12 @@ private:
     ProgramManager& program_manager;
     StateTracker& state_tracker;
 
-    OGLStreamBuffer stream_buffer;
     TextureCacheRuntime texture_cache_runtime;
     TextureCache texture_cache;
+    BufferCacheRuntime buffer_cache_runtime;
+    BufferCache buffer_cache;
     ShaderCacheOpenGL shader_cache;
     QueryCache query_cache;
-    OGLBufferCache buffer_cache;
     FenceManagerOpenGL fence_manager;
 
     VideoCommon::Shader::AsyncShaders async_shaders;
@@ -262,20 +245,8 @@ private:
     boost::container::static_vector<u32, MAX_IMAGE_VIEWS> image_view_indices;
     std::array<ImageViewId, MAX_IMAGE_VIEWS> image_view_ids;
     boost::container::static_vector<GLuint, MAX_TEXTURES> sampler_handles;
-    std::array<GLuint, MAX_TEXTURES> texture_handles;
-    std::array<GLuint, MAX_IMAGES> image_handles;
-
-    std::array<OGLBuffer, Tegra::Engines::Maxwell3D::Regs::NumTransformFeedbackBuffers>
-        transform_feedback_buffers;
-    std::bitset<Tegra::Engines::Maxwell3D::Regs::NumTransformFeedbackBuffers>
-        enabled_transform_feedback_buffers;
-
-    static constexpr std::size_t NUM_CONSTANT_BUFFERS =
-        Tegra::Engines::Maxwell3D::Regs::MaxConstBuffers *
-        Tegra::Engines::Maxwell3D::Regs::MaxShaderProgram;
-    std::array<GLuint, NUM_CONSTANT_BUFFERS> staging_cbufs{};
-    std::size_t current_cbuf = 0;
-    OGLBuffer unified_uniform_buffer;
+    std::array<GLuint, MAX_TEXTURES> texture_handles{};
+    std::array<GLuint, MAX_IMAGES> image_handles{};
 
     /// Number of commands queued to the OpenGL driver. Resetted on flush.
     std::size_t num_queued_commands = 0;
diff --git a/src/video_core/renderer_opengl/gl_resource_manager.cpp b/src/video_core/renderer_opengl/gl_resource_manager.cpp
index 0e34a0f20..3428e5e21 100644
--- a/src/video_core/renderer_opengl/gl_resource_manager.cpp
+++ b/src/video_core/renderer_opengl/gl_resource_manager.cpp
@@ -171,12 +171,6 @@ void OGLBuffer::Release() {
     handle = 0;
 }
 
-void OGLBuffer::MakeStreamCopy(std::size_t buffer_size) {
-    ASSERT_OR_EXECUTE((handle != 0 && buffer_size != 0), { return; });
-
-    glNamedBufferData(handle, buffer_size, nullptr, GL_STREAM_COPY);
-}
-
 void OGLSync::Create() {
     if (handle != 0)
         return;
diff --git a/src/video_core/renderer_opengl/gl_resource_manager.h b/src/video_core/renderer_opengl/gl_resource_manager.h
index f48398669..552d79db4 100644
--- a/src/video_core/renderer_opengl/gl_resource_manager.h
+++ b/src/video_core/renderer_opengl/gl_resource_manager.h
@@ -234,9 +234,6 @@ public:
     /// Deletes the internal OpenGL resource
     void Release();
 
-    // Converts the buffer into a stream copy buffer with a fixed size
-    void MakeStreamCopy(std::size_t buffer_size);
-
     GLuint handle = 0;
 };
 
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
index c35b71b6b..ac78d344c 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
@@ -64,7 +64,7 @@ using TextureIR = std::variant<TextureOffset, TextureDerivates, TextureArgument>
 constexpr u32 MAX_CONSTBUFFER_SCALARS = static_cast<u32>(Maxwell::MaxConstBufferSize) / sizeof(u32);
 constexpr u32 MAX_CONSTBUFFER_ELEMENTS = MAX_CONSTBUFFER_SCALARS / sizeof(u32);
 
-constexpr std::string_view CommonDeclarations = R"(#define ftoi floatBitsToInt
+constexpr std::string_view COMMON_DECLARATIONS = R"(#define ftoi floatBitsToInt
 #define ftou floatBitsToUint
 #define itof intBitsToFloat
 #define utof uintBitsToFloat
@@ -77,10 +77,6 @@ bvec2 HalfFloatNanComparison(bvec2 comparison, vec2 pair1, vec2 pair2) {{
 
 const float fswzadd_modifiers_a[] = float[4](-1.0f,  1.0f, -1.0f,  0.0f );
 const float fswzadd_modifiers_b[] = float[4](-1.0f, -1.0f,  1.0f, -1.0f );
-
-layout (std140, binding = {}) uniform vs_config {{
-    float y_direction;
-}};
 )";
 
 class ShaderWriter final {
@@ -402,13 +398,6 @@ std::string FlowStackTopName(MetaStackClass stack) {
     return fmt::format("{}_flow_stack_top", GetFlowStackPrefix(stack));
 }
 
-bool UseUnifiedUniforms(const Device& device, const ShaderIR& ir, ShaderType stage) {
-    const u32 num_ubos = static_cast<u32>(ir.GetConstantBuffers().size());
-    // We waste one UBO for emulation
-    const u32 num_available_ubos = device.GetMaxUniformBuffers(stage) - 1;
-    return num_ubos > num_available_ubos;
-}
-
 struct GenericVaryingDescription {
     std::string name;
     u8 first_element = 0;
@@ -420,9 +409,8 @@ public:
     explicit GLSLDecompiler(const Device& device_, const ShaderIR& ir_, const Registry& registry_,
                             ShaderType stage_, std::string_view identifier_,
                             std::string_view suffix_)
-        : device{device_}, ir{ir_}, registry{registry_}, stage{stage_}, identifier{identifier_},
-          suffix{suffix_}, header{ir.GetHeader()}, use_unified_uniforms{
-                                                       UseUnifiedUniforms(device_, ir_, stage_)} {
+        : device{device_}, ir{ir_}, registry{registry_}, stage{stage_},
+          identifier{identifier_}, suffix{suffix_}, header{ir.GetHeader()} {
         if (stage != ShaderType::Compute) {
             transform_feedback = BuildTransformFeedback(registry.GetGraphicsInfo());
         }
@@ -516,7 +504,8 @@ private:
         if (!identifier.empty()) {
             code.AddLine("// {}", identifier);
         }
-        code.AddLine("#version 440 {}", ir.UsesLegacyVaryings() ? "compatibility" : "core");
+        const bool use_compatibility = ir.UsesLegacyVaryings() || ir.UsesYNegate();
+        code.AddLine("#version 440 {}", use_compatibility ? "compatibility" : "core");
         code.AddLine("#extension GL_ARB_separate_shader_objects : enable");
         if (device.HasShaderBallot()) {
             code.AddLine("#extension GL_ARB_shader_ballot : require");
@@ -542,7 +531,7 @@ private:
 
         code.AddNewLine();
 
-        code.AddLine(CommonDeclarations, EmulationUniformBlockBinding);
+        code.AddLine(COMMON_DECLARATIONS);
     }
 
     void DeclareVertex() {
@@ -865,17 +854,6 @@ private:
     }
 
     void DeclareConstantBuffers() {
-        if (use_unified_uniforms) {
-            const u32 binding = device.GetBaseBindings(stage).shader_storage_buffer +
-                                static_cast<u32>(ir.GetGlobalMemory().size());
-            code.AddLine("layout (std430, binding = {}) readonly buffer UnifiedUniforms {{",
-                         binding);
-            code.AddLine("    uint cbufs[];");
-            code.AddLine("}};");
-            code.AddNewLine();
-            return;
-        }
-
         u32 binding = device.GetBaseBindings(stage).uniform_buffer;
         for (const auto& [index, info] : ir.GetConstantBuffers()) {
             const u32 num_elements = Common::DivCeil(info.GetSize(), 4 * sizeof(u32));
@@ -1081,29 +1059,17 @@ private:
 
         if (const auto cbuf = std::get_if<CbufNode>(&*node)) {
             const Node offset = cbuf->GetOffset();
-            const u32 base_unified_offset = cbuf->GetIndex() * MAX_CONSTBUFFER_SCALARS;
 
             if (const auto immediate = std::get_if<ImmediateNode>(&*offset)) {
                 // Direct access
                 const u32 offset_imm = immediate->GetValue();
                 ASSERT_MSG(offset_imm % 4 == 0, "Unaligned cbuf direct access");
-                if (use_unified_uniforms) {
-                    return {fmt::format("cbufs[{}]", base_unified_offset + offset_imm / 4),
-                            Type::Uint};
-                } else {
-                    return {fmt::format("{}[{}][{}]", GetConstBuffer(cbuf->GetIndex()),
-                                        offset_imm / (4 * 4), (offset_imm / 4) % 4),
-                            Type::Uint};
-                }
-            }
-
-            // Indirect access
-            if (use_unified_uniforms) {
-                return {fmt::format("cbufs[{} + ({} >> 2)]", base_unified_offset,
-                                    Visit(offset).AsUint()),
+                return {fmt::format("{}[{}][{}]", GetConstBuffer(cbuf->GetIndex()),
+                                    offset_imm / (4 * 4), (offset_imm / 4) % 4),
                         Type::Uint};
             }
 
+            // Indirect access
             const std::string final_offset = code.GenerateTemporary();
             code.AddLine("uint {} = {} >> 2;", final_offset, Visit(offset).AsUint());
 
@@ -2293,7 +2259,6 @@ private:
                 }
             }
         }
-
         if (header.ps.omap.depth) {
             // The depth output is always 2 registers after the last color output, and current_reg
             // already contains one past the last color register.
@@ -2337,7 +2302,8 @@ private:
     }
 
     Expression YNegate(Operation operation) {
-        return {"y_direction", Type::Float};
+        // Y_NEGATE is mapped to this uniform value
+        return {"gl_FrontMaterial.ambient.a", Type::Float};
     }
 
     template <u32 element>
@@ -2787,7 +2753,6 @@ private:
     const std::string_view identifier;
     const std::string_view suffix;
     const Header header;
-    const bool use_unified_uniforms;
     std::unordered_map<u8, VaryingTFB> transform_feedback;
 
     ShaderWriter code;
@@ -3003,8 +2968,10 @@ ShaderEntries MakeEntries(const Device& device, const ShaderIR& ir, ShaderType s
     for (std::size_t i = 0; i < std::size(clip_distances); ++i) {
         entries.clip_distances = (clip_distances[i] ? 1U : 0U) << i;
     }
+    for (const auto& buffer : entries.const_buffers) {
+        entries.enabled_uniform_buffers |= 1U << buffer.GetIndex();
+    }
     entries.shader_length = ir.GetLength();
-    entries.use_unified_uniforms = UseUnifiedUniforms(device, ir, stage);
     return entries;
 }
 
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.h b/src/video_core/renderer_opengl/gl_shader_decompiler.h
index be68994bb..0397a000c 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.h
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.h
@@ -55,7 +55,7 @@ struct ShaderEntries {
     std::vector<ImageEntry> images;
     std::size_t shader_length{};
     u32 clip_distances{};
-    bool use_unified_uniforms{};
+    u32 enabled_uniform_buffers{};
 };
 
 ShaderEntries MakeEntries(const Device& device, const VideoCommon::Shader::ShaderIR& ir,
diff --git a/src/video_core/renderer_opengl/gl_state_tracker.cpp b/src/video_core/renderer_opengl/gl_state_tracker.cpp
index 60e6fa39f..dbdf5230f 100644
--- a/src/video_core/renderer_opengl/gl_state_tracker.cpp
+++ b/src/video_core/renderer_opengl/gl_state_tracker.cpp
@@ -36,16 +36,10 @@ void SetupDirtyColorMasks(Tables& tables) {
     FillBlock(tables[1], OFF(color_mask), NUM(color_mask), ColorMasks);
 }
 
-void SetupDirtyVertexArrays(Tables& tables) {
-    static constexpr std::size_t num_array = 3;
+void SetupDirtyVertexInstances(Tables& tables) {
     static constexpr std::size_t instance_base_offset = 3;
     for (std::size_t i = 0; i < Regs::NumVertexArrays; ++i) {
         const std::size_t array_offset = OFF(vertex_array) + i * NUM(vertex_array[0]);
-        const std::size_t limit_offset = OFF(vertex_array_limit) + i * NUM(vertex_array_limit[0]);
-
-        FillBlock(tables, array_offset, num_array, VertexBuffer0 + i, VertexBuffers);
-        FillBlock(tables, limit_offset, NUM(vertex_array_limit), VertexBuffer0 + i, VertexBuffers);
-
         const std::size_t instance_array_offset = array_offset + instance_base_offset;
         tables[0][instance_array_offset] = static_cast<u8>(VertexInstance0 + i);
         tables[1][instance_array_offset] = VertexInstances;
@@ -217,11 +211,11 @@ void SetupDirtyMisc(Tables& tables) {
 StateTracker::StateTracker(Tegra::GPU& gpu) : flags{gpu.Maxwell3D().dirty.flags} {
     auto& dirty = gpu.Maxwell3D().dirty;
     auto& tables = dirty.tables;
-    SetupDirtyRenderTargets(tables);
+    SetupDirtyFlags(tables);
     SetupDirtyColorMasks(tables);
     SetupDirtyViewports(tables);
     SetupDirtyScissors(tables);
-    SetupDirtyVertexArrays(tables);
+    SetupDirtyVertexInstances(tables);
     SetupDirtyVertexFormat(tables);
     SetupDirtyShaders(tables);
     SetupDirtyPolygonModes(tables);
@@ -241,19 +235,6 @@ StateTracker::StateTracker(Tegra::GPU& gpu) : flags{gpu.Maxwell3D().dirty.flags}
     SetupDirtyClipControl(tables);
     SetupDirtyDepthClampEnabled(tables);
     SetupDirtyMisc(tables);
-
-    auto& store = dirty.on_write_stores;
-    store[VertexBuffers] = true;
-    for (std::size_t i = 0; i < Regs::NumVertexArrays; ++i) {
-        store[VertexBuffer0 + i] = true;
-    }
-}
-
-void StateTracker::InvalidateStreamBuffer() {
-    flags[Dirty::VertexBuffers] = true;
-    for (int index = Dirty::VertexBuffer0; index <= Dirty::VertexBuffer31; ++index) {
-        flags[index] = true;
-    }
 }
 
 } // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_state_tracker.h b/src/video_core/renderer_opengl/gl_state_tracker.h
index 574615d3c..94c905116 100644
--- a/src/video_core/renderer_opengl/gl_state_tracker.h
+++ b/src/video_core/renderer_opengl/gl_state_tracker.h
@@ -28,10 +28,6 @@ enum : u8 {
     VertexFormat0,
     VertexFormat31 = VertexFormat0 + 31,
 
-    VertexBuffers,
-    VertexBuffer0,
-    VertexBuffer31 = VertexBuffer0 + 31,
-
     VertexInstances,
     VertexInstance0,
     VertexInstance31 = VertexInstance0 + 31,
@@ -92,8 +88,6 @@ class StateTracker {
 public:
     explicit StateTracker(Tegra::GPU& gpu);
 
-    void InvalidateStreamBuffer();
-
     void BindIndexBuffer(GLuint new_index_buffer) {
         if (index_buffer == new_index_buffer) {
             return;
@@ -110,13 +104,32 @@ public:
         glBindFramebuffer(GL_DRAW_FRAMEBUFFER, framebuffer);
     }
 
+    void ClipControl(GLenum new_origin, GLenum new_depth) {
+        if (new_origin == origin && new_depth == depth) {
+            return;
+        }
+        origin = new_origin;
+        depth = new_depth;
+        glClipControl(origin, depth);
+    }
+
+    void SetYNegate(bool new_y_negate) {
+        if (new_y_negate == y_negate) {
+            return;
+        }
+        // Y_NEGATE is mapped to gl_FrontMaterial.ambient.a
+        y_negate = new_y_negate;
+        const std::array ambient{0.0f, 0.0f, 0.0f, y_negate ? -1.0f : 1.0f};
+        glMaterialfv(GL_FRONT, GL_AMBIENT, ambient.data());
+    }
+
     void NotifyScreenDrawVertexArray() {
         flags[OpenGL::Dirty::VertexFormats] = true;
         flags[OpenGL::Dirty::VertexFormat0 + 0] = true;
         flags[OpenGL::Dirty::VertexFormat0 + 1] = true;
 
-        flags[OpenGL::Dirty::VertexBuffers] = true;
-        flags[OpenGL::Dirty::VertexBuffer0] = true;
+        flags[VideoCommon::Dirty::VertexBuffers] = true;
+        flags[VideoCommon::Dirty::VertexBuffer0] = true;
 
         flags[OpenGL::Dirty::VertexInstances] = true;
         flags[OpenGL::Dirty::VertexInstance0 + 0] = true;
@@ -202,6 +215,9 @@ private:
 
     GLuint framebuffer = 0;
     GLuint index_buffer = 0;
+    GLenum origin = GL_LOWER_LEFT;
+    GLenum depth = GL_NEGATIVE_ONE_TO_ONE;
+    bool y_negate = false;
 };
 
 } // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_stream_buffer.cpp b/src/video_core/renderer_opengl/gl_stream_buffer.cpp
index e0819cdf2..bfb992a79 100644
--- a/src/video_core/renderer_opengl/gl_stream_buffer.cpp
+++ b/src/video_core/renderer_opengl/gl_stream_buffer.cpp
@@ -1,70 +1,64 @@
-// Copyright 2018 Citra Emulator Project
+// Copyright 2021 yuzu Emulator Project
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include <tuple>
-#include <vector>
+#include <array>
+#include <memory>
+#include <span>
+
+#include <glad/glad.h>
 
 #include "common/alignment.h"
 #include "common/assert.h"
-#include "common/microprofile.h"
-#include "video_core/renderer_opengl/gl_device.h"
-#include "video_core/renderer_opengl/gl_state_tracker.h"
 #include "video_core/renderer_opengl/gl_stream_buffer.h"
 
-MICROPROFILE_DEFINE(OpenGL_StreamBuffer, "OpenGL", "Stream Buffer Orphaning",
-                    MP_RGB(128, 128, 192));
-
 namespace OpenGL {
 
-OGLStreamBuffer::OGLStreamBuffer(const Device& device, StateTracker& state_tracker_)
-    : state_tracker{state_tracker_} {
-    gl_buffer.Create();
-
-    static constexpr GLbitfield flags = GL_MAP_WRITE_BIT | GL_MAP_PERSISTENT_BIT;
-    glNamedBufferStorage(gl_buffer.handle, BUFFER_SIZE, nullptr, flags);
-    mapped_ptr = static_cast<u8*>(
-        glMapNamedBufferRange(gl_buffer.handle, 0, BUFFER_SIZE, flags | GL_MAP_FLUSH_EXPLICIT_BIT));
-
-    if (device.UseAssemblyShaders() || device.HasVertexBufferUnifiedMemory()) {
-        glMakeNamedBufferResidentNV(gl_buffer.handle, GL_READ_ONLY);
-        glGetNamedBufferParameterui64vNV(gl_buffer.handle, GL_BUFFER_GPU_ADDRESS_NV, &gpu_address);
+StreamBuffer::StreamBuffer() {
+    static constexpr GLenum flags = GL_MAP_WRITE_BIT | GL_MAP_PERSISTENT_BIT | GL_MAP_COHERENT_BIT;
+    buffer.Create();
+    glObjectLabel(GL_BUFFER, buffer.handle, -1, "Stream Buffer");
+    glNamedBufferStorage(buffer.handle, STREAM_BUFFER_SIZE, nullptr, flags);
+    mapped_pointer =
+        static_cast<u8*>(glMapNamedBufferRange(buffer.handle, 0, STREAM_BUFFER_SIZE, flags));
+    for (OGLSync& sync : fences) {
+        sync.Create();
     }
 }
 
-OGLStreamBuffer::~OGLStreamBuffer() {
-    glUnmapNamedBuffer(gl_buffer.handle);
-    gl_buffer.Release();
-}
-
-std::pair<u8*, GLintptr> OGLStreamBuffer::Map(GLsizeiptr size, GLintptr alignment) {
-    ASSERT(size <= BUFFER_SIZE);
-    ASSERT(alignment <= BUFFER_SIZE);
-    mapped_size = size;
-
-    if (alignment > 0) {
-        buffer_pos = Common::AlignUp<std::size_t>(buffer_pos, alignment);
+std::pair<std::span<u8>, size_t> StreamBuffer::Request(size_t size) noexcept {
+    ASSERT(size < REGION_SIZE);
+    for (size_t region = Region(used_iterator), region_end = Region(iterator); region < region_end;
+         ++region) {
+        fences[region].Create();
     }
+    used_iterator = iterator;
 
-    if (buffer_pos + size > BUFFER_SIZE) {
-        MICROPROFILE_SCOPE(OpenGL_StreamBuffer);
-        glInvalidateBufferData(gl_buffer.handle);
-        state_tracker.InvalidateStreamBuffer();
-
-        buffer_pos = 0;
+    for (size_t region = Region(free_iterator) + 1,
+                region_end = std::min(Region(iterator + size) + 1, NUM_SYNCS);
+         region < region_end; ++region) {
+        glClientWaitSync(fences[region].handle, 0, GL_TIMEOUT_IGNORED);
+        fences[region].Release();
     }
-
-    return std::make_pair(mapped_ptr + buffer_pos, buffer_pos);
-}
-
-void OGLStreamBuffer::Unmap(GLsizeiptr size) {
-    ASSERT(size <= mapped_size);
-
-    if (size > 0) {
-        glFlushMappedNamedBufferRange(gl_buffer.handle, buffer_pos, size);
+    if (iterator + size > free_iterator) {
+        free_iterator = iterator + size;
     }
-
-    buffer_pos += size;
+    if (iterator + size > STREAM_BUFFER_SIZE) {
+        for (size_t region = Region(used_iterator); region < NUM_SYNCS; ++region) {
+            fences[region].Create();
+        }
+        used_iterator = 0;
+        iterator = 0;
+        free_iterator = size;
+
+        for (size_t region = 0, region_end = Region(size); region <= region_end; ++region) {
+            glClientWaitSync(fences[region].handle, 0, GL_TIMEOUT_IGNORED);
+            fences[region].Release();
+        }
+    }
+    const size_t offset = iterator;
+    iterator = Common::AlignUp(iterator + size, MAX_ALIGNMENT);
+    return {std::span(mapped_pointer + offset, size), offset};
 }
 
 } // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_stream_buffer.h b/src/video_core/renderer_opengl/gl_stream_buffer.h
index dd9cf67eb..6dbb6bfba 100644
--- a/src/video_core/renderer_opengl/gl_stream_buffer.h
+++ b/src/video_core/renderer_opengl/gl_stream_buffer.h
@@ -1,9 +1,12 @@
-// Copyright 2018 Citra Emulator Project
+// Copyright 2021 yuzu Emulator Project
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
 #pragma once
 
+#include <array>
+#include <memory>
+#include <span>
 #include <utility>
 
 #include <glad/glad.h>
@@ -13,48 +16,35 @@
 
 namespace OpenGL {
 
-class Device;
-class StateTracker;
+class StreamBuffer {
+    static constexpr size_t STREAM_BUFFER_SIZE = 64 * 1024 * 1024;
+    static constexpr size_t NUM_SYNCS = 16;
+    static constexpr size_t REGION_SIZE = STREAM_BUFFER_SIZE / NUM_SYNCS;
+    static constexpr size_t MAX_ALIGNMENT = 256;
+    static_assert(STREAM_BUFFER_SIZE % MAX_ALIGNMENT == 0);
+    static_assert(STREAM_BUFFER_SIZE % NUM_SYNCS == 0);
+    static_assert(REGION_SIZE % MAX_ALIGNMENT == 0);
 
-class OGLStreamBuffer : private NonCopyable {
 public:
-    explicit OGLStreamBuffer(const Device& device, StateTracker& state_tracker_);
-    ~OGLStreamBuffer();
-
-    /*
-     * Allocates a linear chunk of memory in the GPU buffer with at least "size" bytes
-     * and the optional alignment requirement.
-     * If the buffer is full, the whole buffer is reallocated which invalidates old chunks.
-     * The return values are the pointer to the new chunk, and the offset within the buffer.
-     * The actual used size must be specified on unmapping the chunk.
-     */
-    std::pair<u8*, GLintptr> Map(GLsizeiptr size, GLintptr alignment = 0);
-
-    void Unmap(GLsizeiptr size);
-
-    GLuint Handle() const {
-        return gl_buffer.handle;
-    }
+    explicit StreamBuffer();
 
-    u64 Address() const {
-        return gpu_address;
-    }
+    [[nodiscard]] std::pair<std::span<u8>, size_t> Request(size_t size) noexcept;
 
-    GLsizeiptr Size() const noexcept {
-        return BUFFER_SIZE;
+    [[nodiscard]] GLuint Handle() const noexcept {
+        return buffer.handle;
     }
 
 private:
-    static constexpr GLsizeiptr BUFFER_SIZE = 256 * 1024 * 1024;
-
-    StateTracker& state_tracker;
-
-    OGLBuffer gl_buffer;
+    [[nodiscard]] static size_t Region(size_t offset) noexcept {
+        return offset / REGION_SIZE;
+    }
 
-    GLuint64EXT gpu_address = 0;
-    GLintptr buffer_pos = 0;
-    GLsizeiptr mapped_size = 0;
-    u8* mapped_ptr = nullptr;
+    size_t iterator = 0;
+    size_t used_iterator = 0;
+    size_t free_iterator = 0;
+    u8* mapped_pointer = nullptr;
+    OGLBuffer buffer;
+    std::array<OGLSync, NUM_SYNCS> fences;
 };
 
 } // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp
index 546cb6d00..37572ab28 100644
--- a/src/video_core/renderer_opengl/gl_texture_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp
@@ -398,9 +398,6 @@ void AttachTexture(GLuint fbo, GLenum attachment, const ImageView* image_view) {
 
 } // Anonymous namespace
 
-ImageBufferMap::ImageBufferMap(GLuint handle_, u8* map, size_t size, OGLSync* sync_)
-    : span(map, size), sync{sync_}, handle{handle_} {}
-
 ImageBufferMap::~ImageBufferMap() {
     if (sync) {
         sync->Create();
@@ -487,11 +484,11 @@ void TextureCacheRuntime::Finish() {
     glFinish();
 }
 
-ImageBufferMap TextureCacheRuntime::MapUploadBuffer(size_t size) {
+ImageBufferMap TextureCacheRuntime::UploadStagingBuffer(size_t size) {
     return upload_buffers.RequestMap(size, true);
 }
 
-ImageBufferMap TextureCacheRuntime::MapDownloadBuffer(size_t size) {
+ImageBufferMap TextureCacheRuntime::DownloadStagingBuffer(size_t size) {
     return download_buffers.RequestMap(size, false);
 }
 
@@ -596,7 +593,11 @@ ImageBufferMap TextureCacheRuntime::StagingBuffers::RequestMap(size_t requested_
                                                                bool insert_fence) {
     const size_t index = RequestBuffer(requested_size);
     OGLSync* const sync = insert_fence ? &syncs[index] : nullptr;
-    return ImageBufferMap(buffers[index].handle, maps[index], requested_size, sync);
+    return ImageBufferMap{
+        .mapped_span = std::span(maps[index], requested_size),
+        .sync = sync,
+        .buffer = buffers[index].handle,
+    };
 }
 
 size_t TextureCacheRuntime::StagingBuffers::RequestBuffer(size_t requested_size) {
@@ -711,7 +712,7 @@ Image::Image(TextureCacheRuntime& runtime, const VideoCommon::ImageInfo& info_,
 
 void Image::UploadMemory(const ImageBufferMap& map, size_t buffer_offset,
                          std::span<const VideoCommon::BufferImageCopy> copies) {
-    glBindBuffer(GL_PIXEL_UNPACK_BUFFER, map.Handle());
+    glBindBuffer(GL_PIXEL_UNPACK_BUFFER, map.buffer);
     glFlushMappedBufferRange(GL_PIXEL_UNPACK_BUFFER, buffer_offset, unswizzled_size_bytes);
 
     glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
@@ -735,7 +736,7 @@ void Image::UploadMemory(const ImageBufferMap& map, size_t buffer_offset,
 void Image::UploadMemory(const ImageBufferMap& map, size_t buffer_offset,
                          std::span<const VideoCommon::BufferCopy> copies) {
     for (const VideoCommon::BufferCopy& copy : copies) {
-        glCopyNamedBufferSubData(map.Handle(), buffer.handle, copy.src_offset + buffer_offset,
+        glCopyNamedBufferSubData(map.buffer, buffer.handle, copy.src_offset + buffer_offset,
                                  copy.dst_offset, copy.size);
     }
 }
@@ -744,7 +745,7 @@ void Image::DownloadMemory(ImageBufferMap& map, size_t buffer_offset,
                            std::span<const VideoCommon::BufferImageCopy> copies) {
     glMemoryBarrier(GL_PIXEL_BUFFER_BARRIER_BIT); // TODO: Move this to its own API
 
-    glBindBuffer(GL_PIXEL_PACK_BUFFER, map.Handle());
+    glBindBuffer(GL_PIXEL_PACK_BUFFER, map.buffer);
     glPixelStorei(GL_PACK_ALIGNMENT, 1);
 
     u32 current_row_length = std::numeric_limits<u32>::max();
diff --git a/src/video_core/renderer_opengl/gl_texture_cache.h b/src/video_core/renderer_opengl/gl_texture_cache.h
index 15b7c3676..60d08d6d6 100644
--- a/src/video_core/renderer_opengl/gl_texture_cache.h
+++ b/src/video_core/renderer_opengl/gl_texture_cache.h
@@ -31,23 +31,12 @@ using VideoCommon::NUM_RT;
 using VideoCommon::Offset2D;
 using VideoCommon::RenderTargets;
 
-class ImageBufferMap {
-public:
-    explicit ImageBufferMap(GLuint handle, u8* map, size_t size, OGLSync* sync);
+struct ImageBufferMap {
     ~ImageBufferMap();
 
-    GLuint Handle() const noexcept {
-        return handle;
-    }
-
-    std::span<u8> Span() const noexcept {
-        return span;
-    }
-
-private:
-    std::span<u8> span;
+    std::span<u8> mapped_span;
     OGLSync* sync;
-    GLuint handle;
+    GLuint buffer;
 };
 
 struct FormatProperties {
@@ -69,9 +58,9 @@ public:
 
     void Finish();
 
-    ImageBufferMap MapUploadBuffer(size_t size);
+    ImageBufferMap UploadStagingBuffer(size_t size);
 
-    ImageBufferMap MapDownloadBuffer(size_t size);
+    ImageBufferMap DownloadStagingBuffer(size_t size);
 
     void CopyImage(Image& dst, Image& src, std::span<const VideoCommon::ImageCopy> copies);
 
diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp
index 7eb5ab17a..8fcb86581 100644
--- a/src/video_core/renderer_opengl/renderer_opengl.cpp
+++ b/src/video_core/renderer_opengl/renderer_opengl.cpp
@@ -27,11 +27,14 @@
 #include "video_core/renderer_opengl/gl_shader_manager.h"
 #include "video_core/renderer_opengl/renderer_opengl.h"
 #include "video_core/textures/decoders.h"
+#include "video_core/vulkan_common/vulkan_debug_callback.h"
+#include "video_core/vulkan_common/vulkan_device.h"
+#include "video_core/vulkan_common/vulkan_instance.h"
+#include "video_core/vulkan_common/vulkan_library.h"
+#include "video_core/vulkan_common/vulkan_memory_allocator.h"
 
 namespace OpenGL {
-
 namespace {
-
 constexpr GLint PositionLocation = 0;
 constexpr GLint TexCoordLocation = 1;
 constexpr GLint ModelViewMatrixLocation = 0;
@@ -125,25 +128,98 @@ void APIENTRY DebugHandler(GLenum source, GLenum type, GLuint id, GLenum severit
     }
 }
 
+Vulkan::vk::PhysicalDevice FindPhysicalDevice(Vulkan::vk::Instance& instance) {
+    using namespace Vulkan;
+    using UUID = std::array<GLubyte, GL_UUID_SIZE_EXT>;
+
+    GLint num_device_uuids;
+    glGetIntegerv(GL_NUM_DEVICE_UUIDS_EXT, &num_device_uuids);
+    std::vector<UUID> device_uuids(num_device_uuids);
+    for (GLint index = 0; index < num_device_uuids; ++index) {
+        glGetUnsignedBytei_vEXT(GL_DEVICE_UUID_EXT, 0, device_uuids[index].data());
+    }
+    UUID driver_uuid;
+    glGetUnsignedBytevEXT(GL_DRIVER_UUID_EXT, driver_uuid.data());
+
+    for (const VkPhysicalDevice raw_physical_device : instance.EnumeratePhysicalDevices()) {
+        VkPhysicalDeviceIDProperties device_id_properties{};
+        device_id_properties.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ID_PROPERTIES;
+
+        VkPhysicalDeviceProperties2KHR properties{
+            .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2_KHR,
+            .pNext = &device_id_properties,
+            .properties{},
+        };
+        vk::PhysicalDevice physical_device(raw_physical_device, instance.Dispatch());
+        physical_device.GetProperties2KHR(properties);
+        if (!std::ranges::equal(device_id_properties.driverUUID, driver_uuid)) {
+            continue;
+        }
+        const auto it =
+            std::ranges::find_if(device_uuids, [&device_id_properties, driver_uuid](UUID uuid) {
+                return std::ranges::equal(device_id_properties.deviceUUID, uuid);
+            });
+        if (it != device_uuids.end()) {
+            return physical_device;
+        }
+    }
+    throw vk::Exception(VK_ERROR_INCOMPATIBLE_DRIVER);
+}
 } // Anonymous namespace
 
+struct VulkanObjects {
+    static std::unique_ptr<VulkanObjects> TryCreate() {
+        if (!GLAD_GL_EXT_memory_object) {
+            // Interop is not present
+            return nullptr;
+        }
+        const std::string_view vendor{reinterpret_cast<const char*>(glGetString(GL_VENDOR))};
+        if (vendor == "ATI Technologies Inc.") {
+            // Avoid using GL_EXT_memory_object on AMD, as it makes the GL driver crash
+            return nullptr;
+        }
+        if (!Settings::values.use_assembly_shaders.GetValue()) {
+            // We only need interop when assembly shaders are enabled
+            return nullptr;
+        }
+#ifdef __linux__
+        LOG_WARNING(Render_OpenGL, "Interop doesn't work on Linux at the moment");
+        return nullptr;
+#endif
+        try {
+            return std::make_unique<VulkanObjects>();
+        } catch (const Vulkan::vk::Exception& exception) {
+            LOG_ERROR(Render_OpenGL, "Failed to initialize Vulkan objects with error: {}",
+                      exception.what());
+            return nullptr;
+        }
+    }
+
+    Common::DynamicLibrary library{Vulkan::OpenLibrary()};
+    Vulkan::vk::InstanceDispatch dld;
+    Vulkan::vk::Instance instance{Vulkan::CreateInstance(library, dld, VK_API_VERSION_1_1)};
+    Vulkan::Device device{*instance, FindPhysicalDevice(instance), nullptr, dld};
+    Vulkan::MemoryAllocator memory_allocator{device, true};
+};
+
 RendererOpenGL::RendererOpenGL(Core::TelemetrySession& telemetry_session_,
                                Core::Frontend::EmuWindow& emu_window_,
                                Core::Memory::Memory& cpu_memory_, Tegra::GPU& gpu_,
                                std::unique_ptr<Core::Frontend::GraphicsContext> context_)
     : RendererBase{emu_window_, std::move(context_)}, telemetry_session{telemetry_session_},
-      emu_window{emu_window_}, cpu_memory{cpu_memory_}, gpu{gpu_}, program_manager{device},
-      rasterizer{emu_window, gpu, cpu_memory, device, screen_info, program_manager, state_tracker} {
+      emu_window{emu_window_}, cpu_memory{cpu_memory_}, gpu{gpu_},
+      vulkan_objects{VulkanObjects::TryCreate()}, device{vulkan_objects != nullptr},
+      state_tracker{gpu}, program_manager{device},
+      rasterizer(emu_window, gpu, cpu_memory, device,
+                 vulkan_objects ? &vulkan_objects->device : nullptr,
+                 vulkan_objects ? &vulkan_objects->memory_allocator : nullptr, screen_info,
+                 program_manager, state_tracker) {
     if (Settings::values.renderer_debug && GLAD_GL_KHR_debug) {
         glEnable(GL_DEBUG_OUTPUT);
         glEnable(GL_DEBUG_OUTPUT_SYNCHRONOUS);
         glDebugMessageCallback(DebugHandler, nullptr);
     }
     AddTelemetryFields();
-
-    if (!GLAD_GL_VERSION_4_6) {
-        throw std::runtime_error{"OpenGL 4.3 is not available"};
-    }
     InitOpenGLObjects();
 }
 
@@ -280,6 +356,7 @@ void RendererOpenGL::InitOpenGLObjects() {
     // Enable unified vertex attributes and query vertex buffer address when the driver supports it
     if (device.HasVertexBufferUnifiedMemory()) {
         glEnableClientState(GL_VERTEX_ATTRIB_ARRAY_UNIFIED_NV);
+        glEnableClientState(GL_ELEMENT_ARRAY_UNIFIED_NV);
 
         glMakeNamedBufferResidentNV(vertex_buffer.handle, GL_READ_ONLY);
         glGetNamedBufferParameterui64vNV(vertex_buffer.handle, GL_BUFFER_GPU_ADDRESS_NV,
@@ -412,6 +489,7 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) {
 
     program_manager.BindHostPipeline(pipeline.handle);
 
+    state_tracker.ClipControl(GL_LOWER_LEFT, GL_ZERO_TO_ONE);
     glEnable(GL_CULL_FACE);
     if (screen_info.display_srgb) {
         glEnable(GL_FRAMEBUFFER_SRGB);
@@ -430,7 +508,6 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) {
     glCullFace(GL_BACK);
     glFrontFace(GL_CW);
     glColorMaski(0, GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE);
-    glClipControl(GL_LOWER_LEFT, GL_ZERO_TO_ONE);
     glViewportIndexedf(0, 0.0f, 0.0f, static_cast<GLfloat>(layout.width),
                        static_cast<GLfloat>(layout.height));
     glDepthRangeIndexed(0, 0.0, 0.0);
diff --git a/src/video_core/renderer_opengl/renderer_opengl.h b/src/video_core/renderer_opengl/renderer_opengl.h
index e043a0ccb..f210190dd 100644
--- a/src/video_core/renderer_opengl/renderer_opengl.h
+++ b/src/video_core/renderer_opengl/renderer_opengl.h
@@ -38,6 +38,8 @@ class GPU;
 
 namespace OpenGL {
 
+struct VulkanObjects;
+
 /// Structure used for storing information about the textures for the Switch screen
 struct TextureInfo {
     OGLTexture resource;
@@ -99,8 +101,11 @@ private:
     Core::Memory::Memory& cpu_memory;
     Tegra::GPU& gpu;
 
-    const Device device;
-    StateTracker state_tracker{gpu};
+    std::unique_ptr<VulkanObjects> vulkan_objects;
+    Device device;
+    StateTracker state_tracker;
+    ProgramManager program_manager;
+    RasterizerOpenGL rasterizer;
 
     // OpenGL object IDs
     OGLSampler present_sampler;
@@ -116,11 +121,6 @@ private:
     /// Display information for Switch screen
     ScreenInfo screen_info;
 
-    /// Global dummy shader pipeline
-    ProgramManager program_manager;
-
-    RasterizerOpenGL rasterizer;
-
     /// OpenGL framebuffer data
     std::vector<u8> gl_framebuffer_data;
 
diff --git a/src/video_core/renderer_opengl/util_shaders.cpp b/src/video_core/renderer_opengl/util_shaders.cpp
index eb849cbf2..aeb36551c 100644
--- a/src/video_core/renderer_opengl/util_shaders.cpp
+++ b/src/video_core/renderer_opengl/util_shaders.cpp
@@ -71,7 +71,7 @@ void UtilShaders::BlockLinearUpload2D(Image& image, const ImageBufferMap& map, s
     static constexpr GLuint BINDING_OUTPUT_IMAGE = 0;
 
     program_manager.BindHostCompute(block_linear_unswizzle_2d_program.handle);
-    glFlushMappedNamedBufferRange(map.Handle(), buffer_offset, image.guest_size_bytes);
+    glFlushMappedNamedBufferRange(map.buffer, buffer_offset, image.guest_size_bytes);
     glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_SWIZZLE_BUFFER, swizzle_table_buffer.handle);
 
     const GLenum store_format = StoreFormat(BytesPerBlock(image.info.format));
@@ -91,8 +91,8 @@ void UtilShaders::BlockLinearUpload2D(Image& image, const ImageBufferMap& map, s
         glUniform1ui(5, params.x_shift);
         glUniform1ui(6, params.block_height);
         glUniform1ui(7, params.block_height_mask);
-        glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.Handle(),
-                          input_offset, image.guest_size_bytes - swizzle.buffer_offset);
+        glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.buffer, input_offset,
+                          image.guest_size_bytes - swizzle.buffer_offset);
         glBindImageTexture(BINDING_OUTPUT_IMAGE, image.Handle(), swizzle.level, GL_TRUE, 0,
                            GL_WRITE_ONLY, store_format);
         glDispatchCompute(num_dispatches_x, num_dispatches_y, image.info.resources.layers);
@@ -108,7 +108,7 @@ void UtilShaders::BlockLinearUpload3D(Image& image, const ImageBufferMap& map, s
     static constexpr GLuint BINDING_INPUT_BUFFER = 1;
     static constexpr GLuint BINDING_OUTPUT_IMAGE = 0;
 
-    glFlushMappedNamedBufferRange(map.Handle(), buffer_offset, image.guest_size_bytes);
+    glFlushMappedNamedBufferRange(map.buffer, buffer_offset, image.guest_size_bytes);
     program_manager.BindHostCompute(block_linear_unswizzle_3d_program.handle);
     glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_SWIZZLE_BUFFER, swizzle_table_buffer.handle);
 
@@ -132,8 +132,8 @@ void UtilShaders::BlockLinearUpload3D(Image& image, const ImageBufferMap& map, s
         glUniform1ui(7, params.block_height_mask);
         glUniform1ui(8, params.block_depth);
         glUniform1ui(9, params.block_depth_mask);
-        glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.Handle(),
-                          input_offset, image.guest_size_bytes - swizzle.buffer_offset);
+        glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.buffer, input_offset,
+                          image.guest_size_bytes - swizzle.buffer_offset);
         glBindImageTexture(BINDING_OUTPUT_IMAGE, image.Handle(), swizzle.level, GL_TRUE, 0,
                            GL_WRITE_ONLY, store_format);
         glDispatchCompute(num_dispatches_x, num_dispatches_y, num_dispatches_z);
@@ -159,7 +159,7 @@ void UtilShaders::PitchUpload(Image& image, const ImageBufferMap& map, size_t bu
                          "Non-power of two images are not implemented");
 
     program_manager.BindHostCompute(pitch_unswizzle_program.handle);
-    glFlushMappedNamedBufferRange(map.Handle(), buffer_offset, image.guest_size_bytes);
+    glFlushMappedNamedBufferRange(map.buffer, buffer_offset, image.guest_size_bytes);
     glUniform2ui(LOC_ORIGIN, 0, 0);
     glUniform2i(LOC_DESTINATION, 0, 0);
     glUniform1ui(LOC_BYTES_PER_BLOCK, bytes_per_block);
@@ -172,8 +172,8 @@ void UtilShaders::PitchUpload(Image& image, const ImageBufferMap& map, size_t bu
         const u32 num_dispatches_x = Common::DivCeil(num_tiles.width, WORKGROUP_SIZE.width);
         const u32 num_dispatches_y = Common::DivCeil(num_tiles.height, WORKGROUP_SIZE.height);
 
-        glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.Handle(),
-                          input_offset, image.guest_size_bytes - swizzle.buffer_offset);
+        glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.buffer, input_offset,
+                          image.guest_size_bytes - swizzle.buffer_offset);
         glDispatchCompute(num_dispatches_x, num_dispatches_y, 1);
     }
     program_manager.RestoreGuestCompute();
diff --git a/src/video_core/renderer_opengl/util_shaders.h b/src/video_core/renderer_opengl/util_shaders.h
index 359997255..bec026bc3 100644
--- a/src/video_core/renderer_opengl/util_shaders.h
+++ b/src/video_core/renderer_opengl/util_shaders.h
@@ -15,9 +15,10 @@
 namespace OpenGL {
 
 class Image;
-class ImageBufferMap;
 class ProgramManager;
 
+struct ImageBufferMap;
+
 class UtilShaders {
 public:
     explicit UtilShaders(ProgramManager& program_manager);
diff --git a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp
index 85121d9fd..19aaf034f 100644
--- a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp
+++ b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp
@@ -531,13 +531,9 @@ VkCompareOp ComparisonOp(Maxwell::ComparisonOp comparison) {
     return {};
 }
 
-VkIndexType IndexFormat(const Device& device, Maxwell::IndexFormat index_format) {
+VkIndexType IndexFormat(Maxwell::IndexFormat index_format) {
     switch (index_format) {
     case Maxwell::IndexFormat::UnsignedByte:
-        if (!device.IsExtIndexTypeUint8Supported()) {
-            UNIMPLEMENTED_MSG("Native uint8 indices are not supported on this device");
-            return VK_INDEX_TYPE_UINT16;
-        }
         return VK_INDEX_TYPE_UINT8_EXT;
     case Maxwell::IndexFormat::UnsignedShort:
         return VK_INDEX_TYPE_UINT16;
diff --git a/src/video_core/renderer_vulkan/maxwell_to_vk.h b/src/video_core/renderer_vulkan/maxwell_to_vk.h
index 7c34b47dc..e3e06ba38 100644
--- a/src/video_core/renderer_vulkan/maxwell_to_vk.h
+++ b/src/video_core/renderer_vulkan/maxwell_to_vk.h
@@ -53,7 +53,7 @@ VkFormat VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttrib
 
 VkCompareOp ComparisonOp(Maxwell::ComparisonOp comparison);
 
-VkIndexType IndexFormat(const Device& device, Maxwell::IndexFormat index_format);
+VkIndexType IndexFormat(Maxwell::IndexFormat index_format);
 
 VkStencilOp StencilOp(Maxwell::StencilOp stencil_op);
 
diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.cpp b/src/video_core/renderer_vulkan/renderer_vulkan.cpp
index 6909576cb..1cc720ddd 100644
--- a/src/video_core/renderer_vulkan/renderer_vulkan.cpp
+++ b/src/video_core/renderer_vulkan/renderer_vulkan.cpp
@@ -107,7 +107,7 @@ RendererVulkan::RendererVulkan(Core::TelemetrySession& telemetry_session_,
       debug_callback(Settings::values.renderer_debug ? CreateDebugCallback(instance) : nullptr),
       surface(CreateSurface(instance, render_window)),
       device(CreateDevice(instance, dld, *surface)),
-      memory_allocator(device),
+      memory_allocator(device, false),
       state_tracker(gpu),
       scheduler(device, state_tracker),
       swapchain(*surface, device, scheduler, render_window.GetFramebufferLayout().width,
diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.h b/src/video_core/renderer_vulkan/renderer_vulkan.h
index 1efaf3b77..72071316c 100644
--- a/src/video_core/renderer_vulkan/renderer_vulkan.h
+++ b/src/video_core/renderer_vulkan/renderer_vulkan.h
@@ -58,12 +58,11 @@ private:
     vk::InstanceDispatch dld;
 
     vk::Instance instance;
-
+    vk::DebugUtilsMessenger debug_callback;
     vk::SurfaceKHR surface;
 
     VKScreenInfo screen_info;
 
-    vk::DebugUtilsMessenger debug_callback;
     Device device;
     MemoryAllocator memory_allocator;
     StateTracker state_tracker;
diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.cpp b/src/video_core/renderer_vulkan/vk_blit_screen.cpp
index df8992528..a1a32aabe 100644
--- a/src/video_core/renderer_vulkan/vk_blit_screen.cpp
+++ b/src/video_core/renderer_vulkan/vk_blit_screen.cpp
@@ -148,8 +148,8 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool
     SetUniformData(data, framebuffer);
     SetVertexData(data, framebuffer);
 
-    const std::span<u8> map = buffer_commit.Map();
-    std::memcpy(map.data(), &data, sizeof(data));
+    const std::span<u8> mapped_span = buffer_commit.Map();
+    std::memcpy(mapped_span.data(), &data, sizeof(data));
 
     if (!use_accelerated) {
         const u64 image_offset = GetRawImageOffset(framebuffer, image_index);
@@ -162,8 +162,8 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool
         constexpr u32 block_height_log2 = 4;
         const u32 bytes_per_pixel = GetBytesPerPixel(framebuffer);
         Tegra::Texture::UnswizzleTexture(
-            map.subspan(image_offset, size_bytes), std::span(host_ptr, size_bytes), bytes_per_pixel,
-            framebuffer.width, framebuffer.height, 1, block_height_log2, 0);
+            mapped_span.subspan(image_offset, size_bytes), std::span(host_ptr, size_bytes),
+            bytes_per_pixel, framebuffer.width, framebuffer.height, 1, block_height_log2, 0);
 
         const VkBufferImageCopy copy{
             .bufferOffset = image_offset,
@@ -263,7 +263,6 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool
         cmdbuf.Draw(4, 1, 0, 0);
         cmdbuf.EndRenderPass();
     });
-
     return *semaphores[image_index];
 }
 
diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
index d8ad40a0f..48fc5d966 100644
--- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
@@ -3,188 +3,276 @@
 // Refer to the license.txt file included.
 
 #include <algorithm>
+#include <array>
 #include <cstring>
-#include <memory>
+#include <span>
+#include <vector>
 
-#include "core/core.h"
 #include "video_core/buffer_cache/buffer_cache.h"
+#include "video_core/renderer_vulkan/maxwell_to_vk.h"
 #include "video_core/renderer_vulkan/vk_buffer_cache.h"
 #include "video_core/renderer_vulkan/vk_scheduler.h"
-#include "video_core/renderer_vulkan/vk_stream_buffer.h"
+#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
+#include "video_core/renderer_vulkan/vk_update_descriptor.h"
 #include "video_core/vulkan_common/vulkan_device.h"
+#include "video_core/vulkan_common/vulkan_memory_allocator.h"
 #include "video_core/vulkan_common/vulkan_wrapper.h"
 
 namespace Vulkan {
-
 namespace {
+VkBufferCopy MakeBufferCopy(const VideoCommon::BufferCopy& copy) {
+    return VkBufferCopy{
+        .srcOffset = copy.src_offset,
+        .dstOffset = copy.dst_offset,
+        .size = copy.size,
+    };
+}
 
-constexpr VkBufferUsageFlags BUFFER_USAGE =
-    VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT |
-    VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT;
-
-constexpr VkPipelineStageFlags UPLOAD_PIPELINE_STAGE =
-    VK_PIPELINE_STAGE_TRANSFER_BIT | VK_PIPELINE_STAGE_VERTEX_INPUT_BIT |
-    VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-    VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT;
-
-constexpr VkAccessFlags UPLOAD_ACCESS_BARRIERS =
-    VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_UNIFORM_READ_BIT |
-    VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_INDEX_READ_BIT;
+VkIndexType IndexTypeFromNumElements(const Device& device, u32 num_elements) {
+    if (num_elements <= 0xff && device.IsExtIndexTypeUint8Supported()) {
+        return VK_INDEX_TYPE_UINT8_EXT;
+    }
+    if (num_elements <= 0xffff) {
+        return VK_INDEX_TYPE_UINT16;
+    }
+    return VK_INDEX_TYPE_UINT32;
+}
 
-constexpr VkAccessFlags TRANSFORM_FEEDBACK_WRITE_ACCESS =
-    VK_ACCESS_TRANSFORM_FEEDBACK_WRITE_BIT_EXT | VK_ACCESS_TRANSFORM_FEEDBACK_COUNTER_WRITE_BIT_EXT;
+size_t BytesPerIndex(VkIndexType index_type) {
+    switch (index_type) {
+    case VK_INDEX_TYPE_UINT8_EXT:
+        return 1;
+    case VK_INDEX_TYPE_UINT16:
+        return 2;
+    case VK_INDEX_TYPE_UINT32:
+        return 4;
+    default:
+        UNREACHABLE_MSG("Invalid index type={}", index_type);
+        return 1;
+    }
+}
 
+template <typename T>
+std::array<T, 6> MakeQuadIndices(u32 quad, u32 first) {
+    std::array<T, 6> indices{0, 1, 2, 0, 2, 3};
+    std::ranges::transform(indices, indices.begin(),
+                           [quad, first](u32 index) { return first + index + quad * 4; });
+    return indices;
+}
 } // Anonymous namespace
 
-Buffer::Buffer(const Device& device_, MemoryAllocator& memory_allocator, VKScheduler& scheduler_,
-               StagingBufferPool& staging_pool_, VAddr cpu_addr_, std::size_t size_)
-    : BufferBlock{cpu_addr_, size_}, device{device_}, scheduler{scheduler_}, staging_pool{
-                                                                                 staging_pool_} {
-    buffer = device.GetLogical().CreateBuffer(VkBufferCreateInfo{
+Buffer::Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams null_params)
+    : VideoCommon::BufferBase<VideoCore::RasterizerInterface>(null_params) {}
+
+Buffer::Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rasterizer_,
+               VAddr cpu_addr_, u64 size_bytes_)
+    : VideoCommon::BufferBase<VideoCore::RasterizerInterface>(rasterizer_, cpu_addr_, size_bytes_) {
+    buffer = runtime.device.GetLogical().CreateBuffer(VkBufferCreateInfo{
         .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
         .pNext = nullptr,
         .flags = 0,
-        .size = static_cast<VkDeviceSize>(size_),
-        .usage = BUFFER_USAGE | VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT,
+        .size = SizeBytes(),
+        .usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT |
+                 VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT |
+                 VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT |
+                 VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT |
+                 VK_BUFFER_USAGE_VERTEX_BUFFER_BIT,
         .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
         .queueFamilyIndexCount = 0,
         .pQueueFamilyIndices = nullptr,
     });
-    commit = memory_allocator.Commit(buffer, MemoryUsage::DeviceLocal);
+    if (runtime.device.HasDebuggingToolAttached()) {
+        buffer.SetObjectNameEXT(fmt::format("Buffer 0x{:x}", CpuAddr()).c_str());
+    }
+    commit = runtime.memory_allocator.Commit(buffer, MemoryUsage::DeviceLocal);
 }
 
-Buffer::~Buffer() = default;
+BufferCacheRuntime::BufferCacheRuntime(const Device& device_, MemoryAllocator& memory_allocator_,
+                                       VKScheduler& scheduler_, StagingBufferPool& staging_pool_,
+                                       VKUpdateDescriptorQueue& update_descriptor_queue_,
+                                       VKDescriptorPool& descriptor_pool)
+    : device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_},
+      staging_pool{staging_pool_}, update_descriptor_queue{update_descriptor_queue_},
+      uint8_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue),
+      quad_index_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue) {}
 
-void Buffer::Upload(std::size_t offset, std::size_t data_size, const u8* data) {
-    const auto& staging = staging_pool.Request(data_size, MemoryUsage::Upload);
-    std::memcpy(staging.mapped_span.data(), data, data_size);
+StagingBufferRef BufferCacheRuntime::UploadStagingBuffer(size_t size) {
+    return staging_pool.Request(size, MemoryUsage::Upload);
+}
 
-    scheduler.RequestOutsideRenderPassOperationContext();
+StagingBufferRef BufferCacheRuntime::DownloadStagingBuffer(size_t size) {
+    return staging_pool.Request(size, MemoryUsage::Download);
+}
 
-    const VkBuffer handle = Handle();
-    scheduler.Record([staging = staging.buffer, handle, offset, data_size,
-                      &device = device](vk::CommandBuffer cmdbuf) {
-        const VkBufferMemoryBarrier read_barrier{
-            .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
-            .pNext = nullptr,
-            .srcAccessMask =
-                VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT |
-                VK_ACCESS_HOST_WRITE_BIT |
-                (device.IsExtTransformFeedbackSupported() ? TRANSFORM_FEEDBACK_WRITE_ACCESS : 0),
-            .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
-            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-            .buffer = handle,
-            .offset = offset,
-            .size = data_size,
-        };
-        const VkBufferMemoryBarrier write_barrier{
-            .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
-            .pNext = nullptr,
-            .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
-            .dstAccessMask = UPLOAD_ACCESS_BARRIERS,
-            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-            .buffer = handle,
-            .offset = offset,
-            .size = data_size,
-        };
+void BufferCacheRuntime::Finish() {
+    scheduler.Finish();
+}
+
+void BufferCacheRuntime::CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer,
+                                    std::span<const VideoCommon::BufferCopy> copies) {
+    static constexpr VkMemoryBarrier READ_BARRIER{
+        .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+        .pNext = nullptr,
+        .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
+        .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT,
+    };
+    static constexpr VkMemoryBarrier WRITE_BARRIER{
+        .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+        .pNext = nullptr,
+        .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+        .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
+    };
+    // Measuring a popular game, this number never exceeds the specified size once data is warmed up
+    boost::container::small_vector<VkBufferCopy, 3> vk_copies(copies.size());
+    std::ranges::transform(copies, vk_copies.begin(), MakeBufferCopy);
+    scheduler.RequestOutsideRenderPassOperationContext();
+    scheduler.Record([src_buffer, dst_buffer, vk_copies](vk::CommandBuffer cmdbuf) {
         cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
-                               0, read_barrier);
-        cmdbuf.CopyBuffer(staging, handle, VkBufferCopy{0, offset, data_size});
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, UPLOAD_PIPELINE_STAGE, 0,
-                               write_barrier);
+                               0, READ_BARRIER);
+        cmdbuf.CopyBuffer(src_buffer, dst_buffer, vk_copies);
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
+                               0, WRITE_BARRIER);
     });
 }
 
-void Buffer::Download(std::size_t offset, std::size_t data_size, u8* data) {
-    auto staging = staging_pool.Request(data_size, MemoryUsage::Download);
-    scheduler.RequestOutsideRenderPassOperationContext();
+void BufferCacheRuntime::BindIndexBuffer(PrimitiveTopology topology, IndexFormat index_format,
+                                         u32 base_vertex, u32 num_indices, VkBuffer buffer,
+                                         u32 offset, [[maybe_unused]] u32 size) {
+    VkIndexType index_type = MaxwellToVK::IndexFormat(index_format);
+    if (topology == PrimitiveTopology::Quads) {
+        index_type = VK_INDEX_TYPE_UINT32;
+        std::tie(buffer, offset) =
+            quad_index_pass.Assemble(index_format, num_indices, base_vertex, buffer, offset);
+    } else if (index_type == VK_INDEX_TYPE_UINT8_EXT && !device.IsExtIndexTypeUint8Supported()) {
+        index_type = VK_INDEX_TYPE_UINT16;
+        std::tie(buffer, offset) = uint8_pass.Assemble(num_indices, buffer, offset);
+    }
+    scheduler.Record([buffer, offset, index_type](vk::CommandBuffer cmdbuf) {
+        cmdbuf.BindIndexBuffer(buffer, offset, index_type);
+    });
+}
 
-    const VkBuffer handle = Handle();
-    scheduler.Record(
-        [staging = staging.buffer, handle, offset, data_size](vk::CommandBuffer cmdbuf) {
-            const VkBufferMemoryBarrier barrier{
-                .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
-                .pNext = nullptr,
-                .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
-                .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
-                .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-                .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-                .buffer = handle,
-                .offset = offset,
-                .size = data_size,
-            };
-
-            cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_VERTEX_SHADER_BIT |
-                                       VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
-                                       VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
-                                   VK_PIPELINE_STAGE_TRANSFER_BIT, 0, {}, barrier, {});
-            cmdbuf.CopyBuffer(handle, staging, VkBufferCopy{offset, 0, data_size});
-        });
-    scheduler.Finish();
+void BufferCacheRuntime::BindQuadArrayIndexBuffer(u32 first, u32 count) {
+    ReserveQuadArrayLUT(first + count, true);
 
-    std::memcpy(data, staging.mapped_span.data(), data_size);
+    // The LUT has the indices 0, 1, 2, and 3 copied as an array
+    // To apply these 'first' offsets we can apply an offset based on the modulus.
+    const VkIndexType index_type = quad_array_lut_index_type;
+    const size_t sub_first_offset = static_cast<size_t>(first % 4) * (current_num_indices / 4);
+    const size_t offset = (sub_first_offset + first / 4) * 6ULL * BytesPerIndex(index_type);
+    scheduler.Record([buffer = *quad_array_lut, index_type, offset](vk::CommandBuffer cmdbuf) {
+        cmdbuf.BindIndexBuffer(buffer, offset, index_type);
+    });
 }
 
-void Buffer::CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst_offset,
-                      std::size_t copy_size) {
-    scheduler.RequestOutsideRenderPassOperationContext();
+void BufferCacheRuntime::BindVertexBuffer(u32 index, VkBuffer buffer, u32 offset, u32 size,
+                                          u32 stride) {
+    if (device.IsExtExtendedDynamicStateSupported()) {
+        scheduler.Record([index, buffer, offset, size, stride](vk::CommandBuffer cmdbuf) {
+            const VkDeviceSize vk_offset = offset;
+            const VkDeviceSize vk_size = buffer != VK_NULL_HANDLE ? size : VK_WHOLE_SIZE;
+            const VkDeviceSize vk_stride = stride;
+            cmdbuf.BindVertexBuffers2EXT(index, 1, &buffer, &vk_offset, &vk_size, &vk_stride);
+        });
+    } else {
+        scheduler.Record([index, buffer, offset](vk::CommandBuffer cmdbuf) {
+            cmdbuf.BindVertexBuffer(index, buffer, offset);
+        });
+    }
+}
 
-    const VkBuffer dst_buffer = Handle();
-    scheduler.Record([src_buffer = src.Handle(), dst_buffer, src_offset, dst_offset,
-                      copy_size](vk::CommandBuffer cmdbuf) {
-        cmdbuf.CopyBuffer(src_buffer, dst_buffer, VkBufferCopy{src_offset, dst_offset, copy_size});
-
-        std::array<VkBufferMemoryBarrier, 2> barriers;
-        barriers[0].sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
-        barriers[0].pNext = nullptr;
-        barriers[0].srcAccessMask = VK_ACCESS_TRANSFER_READ_BIT;
-        barriers[0].dstAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
-        barriers[0].srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
-        barriers[0].dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
-        barriers[0].buffer = src_buffer;
-        barriers[0].offset = src_offset;
-        barriers[0].size = copy_size;
-        barriers[1].sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
-        barriers[1].pNext = nullptr;
-        barriers[1].srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
-        barriers[1].dstAccessMask = UPLOAD_ACCESS_BARRIERS;
-        barriers[1].srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
-        barriers[1].dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
-        barriers[1].buffer = dst_buffer;
-        barriers[1].offset = dst_offset;
-        barriers[1].size = copy_size;
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, UPLOAD_PIPELINE_STAGE, 0, {},
-                               barriers, {});
+void BufferCacheRuntime::BindTransformFeedbackBuffer(u32 index, VkBuffer buffer, u32 offset,
+                                                     u32 size) {
+    if (!device.IsExtTransformFeedbackSupported()) {
+        // Already logged in the rasterizer
+        return;
+    }
+    scheduler.Record([index, buffer, offset, size](vk::CommandBuffer cmdbuf) {
+        const VkDeviceSize vk_offset = offset;
+        const VkDeviceSize vk_size = size;
+        cmdbuf.BindTransformFeedbackBuffersEXT(index, 1, &buffer, &vk_offset, &vk_size);
     });
 }
 
-VKBufferCache::VKBufferCache(VideoCore::RasterizerInterface& rasterizer_,
-                             Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_,
-                             const Device& device_, MemoryAllocator& memory_allocator_,
-                             VKScheduler& scheduler_, VKStreamBuffer& stream_buffer_,
-                             StagingBufferPool& staging_pool_)
-    : VideoCommon::BufferCache<Buffer, VkBuffer, VKStreamBuffer>{rasterizer_, gpu_memory_,
-                                                                 cpu_memory_, stream_buffer_},
-      device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_},
-      staging_pool{staging_pool_} {}
-
-VKBufferCache::~VKBufferCache() = default;
-
-std::shared_ptr<Buffer> VKBufferCache::CreateBlock(VAddr cpu_addr, std::size_t size) {
-    return std::make_shared<Buffer>(device, memory_allocator, scheduler, staging_pool, cpu_addr,
-                                    size);
+void BufferCacheRuntime::BindBuffer(VkBuffer buffer, u32 offset, u32 size) {
+    update_descriptor_queue.AddBuffer(buffer, offset, size);
 }
 
-VKBufferCache::BufferInfo VKBufferCache::GetEmptyBuffer(std::size_t size) {
-    size = std::max(size, std::size_t(4));
-    const auto& empty = staging_pool.Request(size, MemoryUsage::DeviceLocal);
+void BufferCacheRuntime::ReserveQuadArrayLUT(u32 num_indices, bool wait_for_idle) {
+    if (num_indices <= current_num_indices) {
+        return;
+    }
+    if (wait_for_idle) {
+        scheduler.Finish();
+    }
+    current_num_indices = num_indices;
+    quad_array_lut_index_type = IndexTypeFromNumElements(device, num_indices);
+
+    const u32 num_quads = num_indices / 4;
+    const u32 num_triangle_indices = num_quads * 6;
+    const u32 num_first_offset_copies = 4;
+    const size_t bytes_per_index = BytesPerIndex(quad_array_lut_index_type);
+    const size_t size_bytes = num_triangle_indices * bytes_per_index * num_first_offset_copies;
+    quad_array_lut = device.GetLogical().CreateBuffer(VkBufferCreateInfo{
+        .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .size = size_bytes,
+        .usage = VK_BUFFER_USAGE_INDEX_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT,
+        .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
+        .queueFamilyIndexCount = 0,
+        .pQueueFamilyIndices = nullptr,
+    });
+    if (device.HasDebuggingToolAttached()) {
+        quad_array_lut.SetObjectNameEXT("Quad LUT");
+    }
+    quad_array_lut_commit = memory_allocator.Commit(quad_array_lut, MemoryUsage::DeviceLocal);
+
+    const StagingBufferRef staging = staging_pool.Request(size_bytes, MemoryUsage::Upload);
+    u8* staging_data = staging.mapped_span.data();
+    const size_t quad_size = bytes_per_index * 6;
+    for (u32 first = 0; first < num_first_offset_copies; ++first) {
+        for (u32 quad = 0; quad < num_quads; ++quad) {
+            switch (quad_array_lut_index_type) {
+            case VK_INDEX_TYPE_UINT8_EXT:
+                std::memcpy(staging_data, MakeQuadIndices<u8>(quad, first).data(), quad_size);
+                break;
+            case VK_INDEX_TYPE_UINT16:
+                std::memcpy(staging_data, MakeQuadIndices<u16>(quad, first).data(), quad_size);
+                break;
+            case VK_INDEX_TYPE_UINT32:
+                std::memcpy(staging_data, MakeQuadIndices<u32>(quad, first).data(), quad_size);
+                break;
+            default:
+                UNREACHABLE();
+                break;
+            }
+            staging_data += quad_size;
+        }
+    }
     scheduler.RequestOutsideRenderPassOperationContext();
-    scheduler.Record([size, buffer = empty.buffer](vk::CommandBuffer cmdbuf) {
-        cmdbuf.FillBuffer(buffer, 0, size, 0);
+    scheduler.Record([src_buffer = staging.buffer, dst_buffer = *quad_array_lut,
+                      size_bytes](vk::CommandBuffer cmdbuf) {
+        const VkBufferCopy copy{
+            .srcOffset = 0,
+            .dstOffset = 0,
+            .size = size_bytes,
+        };
+        const VkBufferMemoryBarrier write_barrier{
+            .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_INDEX_READ_BIT,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .buffer = dst_buffer,
+            .offset = 0,
+            .size = size_bytes,
+        };
+        cmdbuf.CopyBuffer(src_buffer, dst_buffer, copy);
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_VERTEX_INPUT_BIT,
+                               0, write_barrier);
     });
-    return {empty.buffer, 0, 0};
 }
 
 } // namespace Vulkan
diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.h b/src/video_core/renderer_vulkan/vk_buffer_cache.h
index 41d577510..d232e1f2d 100644
--- a/src/video_core/renderer_vulkan/vk_buffer_cache.h
+++ b/src/video_core/renderer_vulkan/vk_buffer_cache.h
@@ -4,69 +4,112 @@
 
 #pragma once
 
-#include <memory>
-
-#include "common/common_types.h"
 #include "video_core/buffer_cache/buffer_cache.h"
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/renderer_vulkan/vk_compute_pass.h"
 #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
-#include "video_core/renderer_vulkan/vk_stream_buffer.h"
 #include "video_core/vulkan_common/vulkan_memory_allocator.h"
 #include "video_core/vulkan_common/vulkan_wrapper.h"
 
 namespace Vulkan {
 
 class Device;
+class VKDescriptorPool;
 class VKScheduler;
+class VKUpdateDescriptorQueue;
 
-class Buffer final : public VideoCommon::BufferBlock {
-public:
-    explicit Buffer(const Device& device, MemoryAllocator& memory_allocator, VKScheduler& scheduler,
-                    StagingBufferPool& staging_pool, VAddr cpu_addr_, std::size_t size_);
-    ~Buffer();
-
-    void Upload(std::size_t offset, std::size_t data_size, const u8* data);
-
-    void Download(std::size_t offset, std::size_t data_size, u8* data);
+class BufferCacheRuntime;
 
-    void CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst_offset,
-                  std::size_t copy_size);
+class Buffer : public VideoCommon::BufferBase<VideoCore::RasterizerInterface> {
+public:
+    explicit Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams null_params);
+    explicit Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rasterizer_,
+                    VAddr cpu_addr_, u64 size_bytes_);
 
-    VkBuffer Handle() const {
+    [[nodiscard]] VkBuffer Handle() const noexcept {
         return *buffer;
     }
 
-    u64 Address() const {
-        return 0;
+    operator VkBuffer() const noexcept {
+        return *buffer;
     }
 
 private:
-    const Device& device;
-    VKScheduler& scheduler;
-    StagingBufferPool& staging_pool;
-
     vk::Buffer buffer;
     MemoryCommit commit;
 };
 
-class VKBufferCache final : public VideoCommon::BufferCache<Buffer, VkBuffer, VKStreamBuffer> {
+class BufferCacheRuntime {
+    friend Buffer;
+
+    using PrimitiveTopology = Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology;
+    using IndexFormat = Tegra::Engines::Maxwell3D::Regs::IndexFormat;
+
 public:
-    explicit VKBufferCache(VideoCore::RasterizerInterface& rasterizer,
-                           Tegra::MemoryManager& gpu_memory, Core::Memory::Memory& cpu_memory,
-                           const Device& device, MemoryAllocator& memory_allocator,
-                           VKScheduler& scheduler, VKStreamBuffer& stream_buffer,
-                           StagingBufferPool& staging_pool);
-    ~VKBufferCache();
+    explicit BufferCacheRuntime(const Device& device_, MemoryAllocator& memory_manager_,
+                                VKScheduler& scheduler_, StagingBufferPool& staging_pool_,
+                                VKUpdateDescriptorQueue& update_descriptor_queue_,
+                                VKDescriptorPool& descriptor_pool);
+
+    void Finish();
+
+    [[nodiscard]] StagingBufferRef UploadStagingBuffer(size_t size);
+
+    [[nodiscard]] StagingBufferRef DownloadStagingBuffer(size_t size);
 
-    BufferInfo GetEmptyBuffer(std::size_t size) override;
+    void CopyBuffer(VkBuffer src_buffer, VkBuffer dst_buffer,
+                    std::span<const VideoCommon::BufferCopy> copies);
 
-protected:
-    std::shared_ptr<Buffer> CreateBlock(VAddr cpu_addr, std::size_t size) override;
+    void BindIndexBuffer(PrimitiveTopology topology, IndexFormat index_format, u32 num_indices,
+                         u32 base_vertex, VkBuffer buffer, u32 offset, u32 size);
+
+    void BindQuadArrayIndexBuffer(u32 first, u32 count);
+
+    void BindVertexBuffer(u32 index, VkBuffer buffer, u32 offset, u32 size, u32 stride);
+
+    void BindTransformFeedbackBuffer(u32 index, VkBuffer buffer, u32 offset, u32 size);
+
+    void BindUniformBuffer(VkBuffer buffer, u32 offset, u32 size) {
+        BindBuffer(buffer, offset, size);
+    }
+
+    void BindStorageBuffer(VkBuffer buffer, u32 offset, u32 size,
+                           [[maybe_unused]] bool is_written) {
+        BindBuffer(buffer, offset, size);
+    }
 
 private:
+    void BindBuffer(VkBuffer buffer, u32 offset, u32 size);
+
+    void ReserveQuadArrayLUT(u32 num_indices, bool wait_for_idle);
+
     const Device& device;
     MemoryAllocator& memory_allocator;
     VKScheduler& scheduler;
     StagingBufferPool& staging_pool;
+    VKUpdateDescriptorQueue& update_descriptor_queue;
+
+    vk::Buffer quad_array_lut;
+    MemoryCommit quad_array_lut_commit;
+    VkIndexType quad_array_lut_index_type{};
+    u32 current_num_indices = 0;
+
+    Uint8Pass uint8_pass;
+    QuadIndexedPass quad_index_pass;
 };
 
+struct BufferCacheParams {
+    using Runtime = Vulkan::BufferCacheRuntime;
+    using Buffer = Vulkan::Buffer;
+
+    static constexpr bool IS_OPENGL = false;
+    static constexpr bool HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS = false;
+    static constexpr bool HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT = false;
+    static constexpr bool NEEDS_BIND_UNIFORM_INDEX = false;
+    static constexpr bool NEEDS_BIND_STORAGE_INDEX = false;
+    static constexpr bool USE_MEMORY_MAPS = true;
+};
+
+using BufferCache = VideoCommon::BufferCache<BufferCacheParams>;
+
 } // namespace Vulkan
diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp
index 5eb6a54be..a4fdcdf81 100644
--- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp
+++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp
@@ -10,7 +10,6 @@
 #include "common/alignment.h"
 #include "common/assert.h"
 #include "common/common_types.h"
-#include "video_core/host_shaders/vulkan_quad_array_comp_spv.h"
 #include "video_core/host_shaders/vulkan_quad_indexed_comp_spv.h"
 #include "video_core/host_shaders/vulkan_uint8_comp_spv.h"
 #include "video_core/renderer_vulkan/vk_compute_pass.h"
@@ -22,30 +21,7 @@
 #include "video_core/vulkan_common/vulkan_wrapper.h"
 
 namespace Vulkan {
-
 namespace {
-
-VkDescriptorSetLayoutBinding BuildQuadArrayPassDescriptorSetLayoutBinding() {
-    return {
-        .binding = 0,
-        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
-        .descriptorCount = 1,
-        .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
-        .pImmutableSamplers = nullptr,
-    };
-}
-
-VkDescriptorUpdateTemplateEntryKHR BuildQuadArrayPassDescriptorUpdateTemplateEntry() {
-    return {
-        .dstBinding = 0,
-        .dstArrayElement = 0,
-        .descriptorCount = 1,
-        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
-        .offset = 0,
-        .stride = sizeof(DescriptorUpdateEntry),
-    };
-}
-
 VkPushConstantRange BuildComputePushConstantRange(std::size_t size) {
     return {
         .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
@@ -162,55 +138,6 @@ VkDescriptorSet VKComputePass::CommitDescriptorSet(
     return set;
 }
 
-QuadArrayPass::QuadArrayPass(const Device& device_, VKScheduler& scheduler_,
-                             VKDescriptorPool& descriptor_pool_,
-                             StagingBufferPool& staging_buffer_pool_,
-                             VKUpdateDescriptorQueue& update_descriptor_queue_)
-    : VKComputePass(device_, descriptor_pool_, BuildQuadArrayPassDescriptorSetLayoutBinding(),
-                    BuildQuadArrayPassDescriptorUpdateTemplateEntry(),
-                    BuildComputePushConstantRange(sizeof(u32)), VULKAN_QUAD_ARRAY_COMP_SPV),
-      scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
-      update_descriptor_queue{update_descriptor_queue_} {}
-
-QuadArrayPass::~QuadArrayPass() = default;
-
-std::pair<VkBuffer, VkDeviceSize> QuadArrayPass::Assemble(u32 num_vertices, u32 first) {
-    const u32 num_triangle_vertices = (num_vertices / 4) * 6;
-    const std::size_t staging_size = num_triangle_vertices * sizeof(u32);
-    const auto staging_ref = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal);
-
-    update_descriptor_queue.Acquire();
-    update_descriptor_queue.AddBuffer(staging_ref.buffer, 0, staging_size);
-    const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue);
-
-    scheduler.RequestOutsideRenderPassOperationContext();
-
-    ASSERT(num_vertices % 4 == 0);
-    const u32 num_quads = num_vertices / 4;
-    scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = staging_ref.buffer,
-                      num_quads, first, set](vk::CommandBuffer cmdbuf) {
-        constexpr u32 dispatch_size = 1024;
-        cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
-        cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, layout, 0, set, {});
-        cmdbuf.PushConstants(layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(first), &first);
-        cmdbuf.Dispatch(Common::AlignUp(num_quads, dispatch_size) / dispatch_size, 1, 1);
-
-        VkBufferMemoryBarrier barrier;
-        barrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
-        barrier.pNext = nullptr;
-        barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
-        barrier.dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT;
-        barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
-        barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
-        barrier.buffer = buffer;
-        barrier.offset = 0;
-        barrier.size = static_cast<VkDeviceSize>(num_quads) * 6 * sizeof(u32);
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
-                               VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, {}, {barrier}, {});
-    });
-    return {staging_ref.buffer, 0};
-}
-
 Uint8Pass::Uint8Pass(const Device& device, VKScheduler& scheduler_,
                      VKDescriptorPool& descriptor_pool, StagingBufferPool& staging_buffer_pool_,
                      VKUpdateDescriptorQueue& update_descriptor_queue_)
@@ -221,18 +148,18 @@ Uint8Pass::Uint8Pass(const Device& device, VKScheduler& scheduler_,
 
 Uint8Pass::~Uint8Pass() = default;
 
-std::pair<VkBuffer, u64> Uint8Pass::Assemble(u32 num_vertices, VkBuffer src_buffer,
-                                             u64 src_offset) {
+std::pair<VkBuffer, u32> Uint8Pass::Assemble(u32 num_vertices, VkBuffer src_buffer,
+                                             u32 src_offset) {
     const u32 staging_size = static_cast<u32>(num_vertices * sizeof(u16));
-    const auto staging_ref = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal);
+    const auto staging = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal);
 
     update_descriptor_queue.Acquire();
     update_descriptor_queue.AddBuffer(src_buffer, src_offset, num_vertices);
-    update_descriptor_queue.AddBuffer(staging_ref.buffer, 0, staging_size);
+    update_descriptor_queue.AddBuffer(staging.buffer, 0, staging_size);
     const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue);
 
     scheduler.RequestOutsideRenderPassOperationContext();
-    scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = staging_ref.buffer, set,
+    scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = staging.buffer, set,
                       num_vertices](vk::CommandBuffer cmdbuf) {
         constexpr u32 dispatch_size = 1024;
         cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
@@ -252,7 +179,7 @@ std::pair<VkBuffer, u64> Uint8Pass::Assemble(u32 num_vertices, VkBuffer src_buff
         cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
                                VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, {}, barrier, {});
     });
-    return {staging_ref.buffer, 0};
+    return {staging.buffer, 0};
 }
 
 QuadIndexedPass::QuadIndexedPass(const Device& device_, VKScheduler& scheduler_,
@@ -267,9 +194,9 @@ QuadIndexedPass::QuadIndexedPass(const Device& device_, VKScheduler& scheduler_,
 
 QuadIndexedPass::~QuadIndexedPass() = default;
 
-std::pair<VkBuffer, u64> QuadIndexedPass::Assemble(
+std::pair<VkBuffer, u32> QuadIndexedPass::Assemble(
     Tegra::Engines::Maxwell3D::Regs::IndexFormat index_format, u32 num_vertices, u32 base_vertex,
-    VkBuffer src_buffer, u64 src_offset) {
+    VkBuffer src_buffer, u32 src_offset) {
     const u32 index_shift = [index_format] {
         switch (index_format) {
         case Tegra::Engines::Maxwell3D::Regs::IndexFormat::UnsignedByte:
@@ -286,15 +213,15 @@ std::pair<VkBuffer, u64> QuadIndexedPass::Assemble(
     const u32 num_tri_vertices = (num_vertices / 4) * 6;
 
     const std::size_t staging_size = num_tri_vertices * sizeof(u32);
-    const auto staging_ref = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal);
+    const auto staging = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal);
 
     update_descriptor_queue.Acquire();
     update_descriptor_queue.AddBuffer(src_buffer, src_offset, input_size);
-    update_descriptor_queue.AddBuffer(staging_ref.buffer, 0, staging_size);
+    update_descriptor_queue.AddBuffer(staging.buffer, 0, staging_size);
     const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue);
 
     scheduler.RequestOutsideRenderPassOperationContext();
-    scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = staging_ref.buffer, set,
+    scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = staging.buffer, set,
                       num_tri_vertices, base_vertex, index_shift](vk::CommandBuffer cmdbuf) {
         static constexpr u32 dispatch_size = 1024;
         const std::array push_constants = {base_vertex, index_shift};
@@ -317,7 +244,7 @@ std::pair<VkBuffer, u64> QuadIndexedPass::Assemble(
         cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
                                VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, {}, barrier, {});
     });
-    return {staging_ref.buffer, 0};
+    return {staging.buffer, 0};
 }
 
 } // namespace Vulkan
diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.h b/src/video_core/renderer_vulkan/vk_compute_pass.h
index f5c6f5f17..4904019f5 100644
--- a/src/video_core/renderer_vulkan/vk_compute_pass.h
+++ b/src/video_core/renderer_vulkan/vk_compute_pass.h
@@ -41,22 +41,6 @@ private:
     vk::ShaderModule module;
 };
 
-class QuadArrayPass final : public VKComputePass {
-public:
-    explicit QuadArrayPass(const Device& device_, VKScheduler& scheduler_,
-                           VKDescriptorPool& descriptor_pool_,
-                           StagingBufferPool& staging_buffer_pool_,
-                           VKUpdateDescriptorQueue& update_descriptor_queue_);
-    ~QuadArrayPass();
-
-    std::pair<VkBuffer, VkDeviceSize> Assemble(u32 num_vertices, u32 first);
-
-private:
-    VKScheduler& scheduler;
-    StagingBufferPool& staging_buffer_pool;
-    VKUpdateDescriptorQueue& update_descriptor_queue;
-};
-
 class Uint8Pass final : public VKComputePass {
 public:
     explicit Uint8Pass(const Device& device_, VKScheduler& scheduler_,
@@ -64,7 +48,9 @@ public:
                        VKUpdateDescriptorQueue& update_descriptor_queue_);
     ~Uint8Pass();
 
-    std::pair<VkBuffer, u64> Assemble(u32 num_vertices, VkBuffer src_buffer, u64 src_offset);
+    /// Assemble uint8 indices into an uint16 index buffer
+    /// Returns a pair with the staging buffer, and the offset where the assembled data is
+    std::pair<VkBuffer, u32> Assemble(u32 num_vertices, VkBuffer src_buffer, u32 src_offset);
 
 private:
     VKScheduler& scheduler;
@@ -80,9 +66,9 @@ public:
                              VKUpdateDescriptorQueue& update_descriptor_queue_);
     ~QuadIndexedPass();
 
-    std::pair<VkBuffer, u64> Assemble(Tegra::Engines::Maxwell3D::Regs::IndexFormat index_format,
+    std::pair<VkBuffer, u32> Assemble(Tegra::Engines::Maxwell3D::Regs::IndexFormat index_format,
                                       u32 num_vertices, u32 base_vertex, VkBuffer src_buffer,
-                                      u64 src_offset);
+                                      u32 src_offset);
 
 private:
     VKScheduler& scheduler;
diff --git a/src/video_core/renderer_vulkan/vk_fence_manager.cpp b/src/video_core/renderer_vulkan/vk_fence_manager.cpp
index 6cd00884d..3bec48d14 100644
--- a/src/video_core/renderer_vulkan/vk_fence_manager.cpp
+++ b/src/video_core/renderer_vulkan/vk_fence_manager.cpp
@@ -45,8 +45,8 @@ void InnerFence::Wait() {
 }
 
 VKFenceManager::VKFenceManager(VideoCore::RasterizerInterface& rasterizer_, Tegra::GPU& gpu_,
-                               Tegra::MemoryManager& memory_manager_, TextureCache& texture_cache_,
-                               VKBufferCache& buffer_cache_, VKQueryCache& query_cache_,
+                               TextureCache& texture_cache_, BufferCache& buffer_cache_,
+                               VKQueryCache& query_cache_, const Device& device_,
                                VKScheduler& scheduler_)
     : GenericFenceManager{rasterizer_, gpu_, texture_cache_, buffer_cache_, query_cache_},
       scheduler{scheduler_} {}
diff --git a/src/video_core/renderer_vulkan/vk_fence_manager.h b/src/video_core/renderer_vulkan/vk_fence_manager.h
index 9c5e5aa8f..2f8322d29 100644
--- a/src/video_core/renderer_vulkan/vk_fence_manager.h
+++ b/src/video_core/renderer_vulkan/vk_fence_manager.h
@@ -22,7 +22,6 @@ class RasterizerInterface;
 namespace Vulkan {
 
 class Device;
-class VKBufferCache;
 class VKQueryCache;
 class VKScheduler;
 
@@ -45,14 +44,14 @@ private:
 using Fence = std::shared_ptr<InnerFence>;
 
 using GenericFenceManager =
-    VideoCommon::FenceManager<Fence, TextureCache, VKBufferCache, VKQueryCache>;
+    VideoCommon::FenceManager<Fence, TextureCache, BufferCache, VKQueryCache>;
 
 class VKFenceManager final : public GenericFenceManager {
 public:
-    explicit VKFenceManager(VideoCore::RasterizerInterface& rasterizer_, Tegra::GPU& gpu_,
-                            Tegra::MemoryManager& memory_manager_, TextureCache& texture_cache_,
-                            VKBufferCache& buffer_cache_, VKQueryCache& query_cache_,
-                            VKScheduler& scheduler_);
+    explicit VKFenceManager(VideoCore::RasterizerInterface& rasterizer, Tegra::GPU& gpu,
+                            TextureCache& texture_cache, BufferCache& buffer_cache,
+                            VKQueryCache& query_cache, const Device& device,
+                            VKScheduler& scheduler);
 
 protected:
     Fence CreateFence(u32 value, bool is_stubbed) override;
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
index f0a111829..684d4e3a6 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
@@ -8,8 +8,6 @@
 #include <mutex>
 #include <vector>
 
-#include <boost/container/static_vector.hpp>
-
 #include "common/alignment.h"
 #include "common/assert.h"
 #include "common/logging/log.h"
@@ -24,7 +22,6 @@
 #include "video_core/renderer_vulkan/maxwell_to_vk.h"
 #include "video_core/renderer_vulkan/renderer_vulkan.h"
 #include "video_core/renderer_vulkan/vk_buffer_cache.h"
-#include "video_core/renderer_vulkan/vk_compute_pass.h"
 #include "video_core/renderer_vulkan/vk_compute_pipeline.h"
 #include "video_core/renderer_vulkan/vk_descriptor_pool.h"
 #include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
@@ -50,15 +47,16 @@ MICROPROFILE_DEFINE(Vulkan_WaitForWorker, "Vulkan", "Wait for worker", MP_RGB(25
 MICROPROFILE_DEFINE(Vulkan_Drawing, "Vulkan", "Record drawing", MP_RGB(192, 128, 128));
 MICROPROFILE_DEFINE(Vulkan_Compute, "Vulkan", "Record compute", MP_RGB(192, 128, 128));
 MICROPROFILE_DEFINE(Vulkan_Clearing, "Vulkan", "Record clearing", MP_RGB(192, 128, 128));
-MICROPROFILE_DEFINE(Vulkan_Geometry, "Vulkan", "Setup geometry", MP_RGB(192, 128, 128));
-MICROPROFILE_DEFINE(Vulkan_ConstBuffers, "Vulkan", "Setup constant buffers", MP_RGB(192, 128, 128));
-MICROPROFILE_DEFINE(Vulkan_GlobalBuffers, "Vulkan", "Setup global buffers", MP_RGB(192, 128, 128));
-MICROPROFILE_DEFINE(Vulkan_RenderTargets, "Vulkan", "Setup render targets", MP_RGB(192, 128, 128));
-MICROPROFILE_DEFINE(Vulkan_Textures, "Vulkan", "Setup textures", MP_RGB(192, 128, 128));
-MICROPROFILE_DEFINE(Vulkan_Images, "Vulkan", "Setup images", MP_RGB(192, 128, 128));
 MICROPROFILE_DEFINE(Vulkan_PipelineCache, "Vulkan", "Pipeline cache", MP_RGB(192, 128, 128));
 
 namespace {
+struct DrawParams {
+    u32 base_instance;
+    u32 num_instances;
+    u32 base_vertex;
+    u32 num_vertices;
+    bool is_indexed;
+};
 
 constexpr auto COMPUTE_SHADER_INDEX = static_cast<size_t>(Tegra::Engines::ShaderType::Compute);
 
@@ -67,7 +65,6 @@ VkViewport GetViewportState(const Device& device, const Maxwell& regs, size_t in
     const float width = src.scale_x * 2.0f;
     const float height = src.scale_y * 2.0f;
     const float reduce_z = regs.depth_mode == Maxwell::DepthMode::MinusOneToOne ? 1.0f : 0.0f;
-
     VkViewport viewport{
         .x = src.translate_x - src.scale_x,
         .y = src.translate_y - src.scale_y,
@@ -76,12 +73,10 @@ VkViewport GetViewportState(const Device& device, const Maxwell& regs, size_t in
         .minDepth = src.translate_z - src.scale_z * reduce_z,
         .maxDepth = src.translate_z + src.scale_z,
     };
-
     if (!device.IsExtDepthRangeUnrestrictedSupported()) {
         viewport.minDepth = std::clamp(viewport.minDepth, 0.0f, 1.0f);
         viewport.maxDepth = std::clamp(viewport.maxDepth, 0.0f, 1.0f);
     }
-
     return viewport;
 }
 
@@ -146,13 +141,6 @@ TextureHandle GetTextureInfo(const Engine& engine, bool via_header_index, const
     return TextureHandle(engine.AccessConstBuffer32(shader_type, buffer, offset), via_header_index);
 }
 
-template <size_t N>
-std::array<VkDeviceSize, N> ExpandStrides(const std::array<u16, N>& strides) {
-    std::array<VkDeviceSize, N> expanded;
-    std::copy(strides.begin(), strides.end(), expanded.begin());
-    return expanded;
-}
-
 ImageViewType ImageViewTypeFromEntry(const SamplerEntry& entry) {
     if (entry.is_buffer) {
         return ImageViewType::e2D;
@@ -221,190 +209,25 @@ void PushImageDescriptors(const ShaderEntries& entries, TextureCache& texture_ca
     }
 }
 
-} // Anonymous namespace
-
-class BufferBindings final {
-public:
-    void AddVertexBinding(VkBuffer buffer, VkDeviceSize offset, VkDeviceSize size, u32 stride) {
-        vertex.buffers[vertex.num_buffers] = buffer;
-        vertex.offsets[vertex.num_buffers] = offset;
-        vertex.sizes[vertex.num_buffers] = size;
-        vertex.strides[vertex.num_buffers] = static_cast<u16>(stride);
-        ++vertex.num_buffers;
-    }
-
-    void SetIndexBinding(VkBuffer buffer, VkDeviceSize offset, VkIndexType type) {
-        index.buffer = buffer;
-        index.offset = offset;
-        index.type = type;
-    }
-
-    void Bind(const Device& device, VKScheduler& scheduler) const {
-        // Use this large switch case to avoid dispatching more memory in the record lambda than
-        // what we need. It looks horrible, but it's the best we can do on standard C++.
-        switch (vertex.num_buffers) {
-        case 0:
-            return BindStatic<0>(device, scheduler);
-        case 1:
-            return BindStatic<1>(device, scheduler);
-        case 2:
-            return BindStatic<2>(device, scheduler);
-        case 3:
-            return BindStatic<3>(device, scheduler);
-        case 4:
-            return BindStatic<4>(device, scheduler);
-        case 5:
-            return BindStatic<5>(device, scheduler);
-        case 6:
-            return BindStatic<6>(device, scheduler);
-        case 7:
-            return BindStatic<7>(device, scheduler);
-        case 8:
-            return BindStatic<8>(device, scheduler);
-        case 9:
-            return BindStatic<9>(device, scheduler);
-        case 10:
-            return BindStatic<10>(device, scheduler);
-        case 11:
-            return BindStatic<11>(device, scheduler);
-        case 12:
-            return BindStatic<12>(device, scheduler);
-        case 13:
-            return BindStatic<13>(device, scheduler);
-        case 14:
-            return BindStatic<14>(device, scheduler);
-        case 15:
-            return BindStatic<15>(device, scheduler);
-        case 16:
-            return BindStatic<16>(device, scheduler);
-        case 17:
-            return BindStatic<17>(device, scheduler);
-        case 18:
-            return BindStatic<18>(device, scheduler);
-        case 19:
-            return BindStatic<19>(device, scheduler);
-        case 20:
-            return BindStatic<20>(device, scheduler);
-        case 21:
-            return BindStatic<21>(device, scheduler);
-        case 22:
-            return BindStatic<22>(device, scheduler);
-        case 23:
-            return BindStatic<23>(device, scheduler);
-        case 24:
-            return BindStatic<24>(device, scheduler);
-        case 25:
-            return BindStatic<25>(device, scheduler);
-        case 26:
-            return BindStatic<26>(device, scheduler);
-        case 27:
-            return BindStatic<27>(device, scheduler);
-        case 28:
-            return BindStatic<28>(device, scheduler);
-        case 29:
-            return BindStatic<29>(device, scheduler);
-        case 30:
-            return BindStatic<30>(device, scheduler);
-        case 31:
-            return BindStatic<31>(device, scheduler);
-        case 32:
-            return BindStatic<32>(device, scheduler);
-        }
-        UNREACHABLE();
-    }
-
-private:
-    // Some of these fields are intentionally left uninitialized to avoid initializing them twice.
-    struct {
-        size_t num_buffers = 0;
-        std::array<VkBuffer, Maxwell::NumVertexArrays> buffers;
-        std::array<VkDeviceSize, Maxwell::NumVertexArrays> offsets;
-        std::array<VkDeviceSize, Maxwell::NumVertexArrays> sizes;
-        std::array<u16, Maxwell::NumVertexArrays> strides;
-    } vertex;
-
-    struct {
-        VkBuffer buffer = nullptr;
-        VkDeviceSize offset;
-        VkIndexType type;
-    } index;
-
-    template <size_t N>
-    void BindStatic(const Device& device, VKScheduler& scheduler) const {
-        if (device.IsExtExtendedDynamicStateSupported()) {
-            if (index.buffer) {
-                BindStatic<N, true, true>(scheduler);
-            } else {
-                BindStatic<N, false, true>(scheduler);
-            }
-        } else {
-            if (index.buffer) {
-                BindStatic<N, true, false>(scheduler);
-            } else {
-                BindStatic<N, false, false>(scheduler);
-            }
-        }
-    }
-
-    template <size_t N, bool is_indexed, bool has_extended_dynamic_state>
-    void BindStatic(VKScheduler& scheduler) const {
-        static_assert(N <= Maxwell::NumVertexArrays);
-        if constexpr (N == 0) {
-            return;
-        }
-
-        std::array<VkBuffer, N> buffers;
-        std::array<VkDeviceSize, N> offsets;
-        std::copy(vertex.buffers.begin(), vertex.buffers.begin() + N, buffers.begin());
-        std::copy(vertex.offsets.begin(), vertex.offsets.begin() + N, offsets.begin());
-
-        if constexpr (has_extended_dynamic_state) {
-            // With extended dynamic states we can specify the length and stride of a vertex buffer
-            std::array<VkDeviceSize, N> sizes;
-            std::array<u16, N> strides;
-            std::copy(vertex.sizes.begin(), vertex.sizes.begin() + N, sizes.begin());
-            std::copy(vertex.strides.begin(), vertex.strides.begin() + N, strides.begin());
-
-            if constexpr (is_indexed) {
-                scheduler.Record(
-                    [buffers, offsets, sizes, strides, index = index](vk::CommandBuffer cmdbuf) {
-                        cmdbuf.BindIndexBuffer(index.buffer, index.offset, index.type);
-                        cmdbuf.BindVertexBuffers2EXT(0, static_cast<u32>(N), buffers.data(),
-                                                     offsets.data(), sizes.data(),
-                                                     ExpandStrides(strides).data());
-                    });
-            } else {
-                scheduler.Record([buffers, offsets, sizes, strides](vk::CommandBuffer cmdbuf) {
-                    cmdbuf.BindVertexBuffers2EXT(0, static_cast<u32>(N), buffers.data(),
-                                                 offsets.data(), sizes.data(),
-                                                 ExpandStrides(strides).data());
-                });
-            }
-            return;
-        }
-
-        if constexpr (is_indexed) {
-            // Indexed draw
-            scheduler.Record([buffers, offsets, index = index](vk::CommandBuffer cmdbuf) {
-                cmdbuf.BindIndexBuffer(index.buffer, index.offset, index.type);
-                cmdbuf.BindVertexBuffers(0, static_cast<u32>(N), buffers.data(), offsets.data());
-            });
-        } else {
-            // Array draw
-            scheduler.Record([buffers, offsets](vk::CommandBuffer cmdbuf) {
-                cmdbuf.BindVertexBuffers(0, static_cast<u32>(N), buffers.data(), offsets.data());
-            });
-        }
-    }
-};
-
-void RasterizerVulkan::DrawParameters::Draw(vk::CommandBuffer cmdbuf) const {
-    if (is_indexed) {
-        cmdbuf.DrawIndexed(num_vertices, num_instances, 0, base_vertex, base_instance);
-    } else {
-        cmdbuf.Draw(num_vertices, num_instances, base_vertex, base_instance);
+DrawParams MakeDrawParams(const Maxwell& regs, u32 num_instances, bool is_instanced,
+                          bool is_indexed) {
+    DrawParams params{
+        .base_instance = regs.vb_base_instance,
+        .num_instances = is_instanced ? num_instances : 1,
+        .base_vertex = is_indexed ? regs.vb_element_base : regs.vertex_buffer.first,
+        .num_vertices = is_indexed ? regs.index_array.count : regs.vertex_buffer.count,
+        .is_indexed = is_indexed,
+    };
+    if (regs.draw.topology == Maxwell::PrimitiveTopology::Quads) {
+        // 6 triangle vertices per quad, base vertex is part of the index
+        // See BindQuadArrayIndexBuffer for more details
+        params.num_vertices = (params.num_vertices / 4) * 6;
+        params.base_vertex = 0;
+        params.is_indexed = true;
     }
+    return params;
 }
+} // Anonymous namespace
 
 RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_,
                                    Tegra::MemoryManager& gpu_memory_,
@@ -414,21 +237,19 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra
     : RasterizerAccelerated{cpu_memory_}, gpu{gpu_},
       gpu_memory{gpu_memory_}, maxwell3d{gpu.Maxwell3D()}, kepler_compute{gpu.KeplerCompute()},
       screen_info{screen_info_}, device{device_}, memory_allocator{memory_allocator_},
-      state_tracker{state_tracker_}, scheduler{scheduler_}, stream_buffer(device, scheduler),
+      state_tracker{state_tracker_}, scheduler{scheduler_},
       staging_pool(device, memory_allocator, scheduler), descriptor_pool(device, scheduler),
       update_descriptor_queue(device, scheduler),
       blit_image(device, scheduler, state_tracker, descriptor_pool),
-      quad_array_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue),
-      quad_indexed_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue),
-      uint8_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue),
       texture_cache_runtime{device, scheduler, memory_allocator, staging_pool, blit_image},
       texture_cache(texture_cache_runtime, *this, maxwell3d, kepler_compute, gpu_memory),
+      buffer_cache_runtime(device, memory_allocator, scheduler, staging_pool,
+                           update_descriptor_queue, descriptor_pool),
+      buffer_cache(*this, maxwell3d, kepler_compute, gpu_memory, cpu_memory_, buffer_cache_runtime),
       pipeline_cache(*this, gpu, maxwell3d, kepler_compute, gpu_memory, device, scheduler,
                      descriptor_pool, update_descriptor_queue),
-      buffer_cache(*this, gpu_memory, cpu_memory_, device, memory_allocator, scheduler,
-                   stream_buffer, staging_pool),
       query_cache{*this, maxwell3d, gpu_memory, device, scheduler},
-      fence_manager(*this, gpu, gpu_memory, texture_cache, buffer_cache, query_cache, scheduler),
+      fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache, device, scheduler),
       wfi_event(device.GetLogical().CreateEvent()), async_shaders(emu_window_) {
     scheduler.SetQueryCache(query_cache);
     if (device.UseAsynchronousShaders()) {
@@ -449,22 +270,14 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) {
     GraphicsPipelineCacheKey key;
     key.fixed_state.Fill(maxwell3d.regs, device.IsExtExtendedDynamicStateSupported());
 
-    buffer_cache.Map(CalculateGraphicsStreamBufferSize(is_indexed));
-
-    BufferBindings buffer_bindings;
-    const DrawParameters draw_params =
-        SetupGeometry(key.fixed_state, buffer_bindings, is_indexed, is_instanced);
+    std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
 
-    auto lock = texture_cache.AcquireLock();
     texture_cache.SynchronizeGraphicsDescriptors();
-
     texture_cache.UpdateRenderTargets(false);
 
     const auto shaders = pipeline_cache.GetShaders();
     key.shaders = GetShaderAddresses(shaders);
-    SetupShaderDescriptors(shaders);
-
-    buffer_cache.Unmap();
+    SetupShaderDescriptors(shaders, is_indexed);
 
     const Framebuffer* const framebuffer = texture_cache.GetFramebuffer();
     key.renderpass = framebuffer->RenderPass();
@@ -476,22 +289,29 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) {
         return;
     }
 
-    buffer_bindings.Bind(device, scheduler);
-
     BeginTransformFeedback();
 
     scheduler.RequestRenderpass(framebuffer);
     scheduler.BindGraphicsPipeline(pipeline->GetHandle());
     UpdateDynamicStates();
 
-    const auto pipeline_layout = pipeline->GetLayout();
-    const auto descriptor_set = pipeline->CommitDescriptorSet();
+    const auto& regs = maxwell3d.regs;
+    const u32 num_instances = maxwell3d.mme_draw.instance_count;
+    const DrawParams draw_params = MakeDrawParams(regs, num_instances, is_instanced, is_indexed);
+    const VkPipelineLayout pipeline_layout = pipeline->GetLayout();
+    const VkDescriptorSet descriptor_set = pipeline->CommitDescriptorSet();
     scheduler.Record([pipeline_layout, descriptor_set, draw_params](vk::CommandBuffer cmdbuf) {
         if (descriptor_set) {
             cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline_layout,
-                                      DESCRIPTOR_SET, descriptor_set, {});
+                                      DESCRIPTOR_SET, descriptor_set, nullptr);
+        }
+        if (draw_params.is_indexed) {
+            cmdbuf.DrawIndexed(draw_params.num_vertices, draw_params.num_instances, 0,
+                               draw_params.base_vertex, draw_params.base_instance);
+        } else {
+            cmdbuf.Draw(draw_params.num_vertices, draw_params.num_instances,
+                        draw_params.base_vertex, draw_params.base_instance);
         }
-        draw_params.Draw(cmdbuf);
     });
 
     EndTransformFeedback();
@@ -515,7 +335,7 @@ void RasterizerVulkan::Clear() {
         return;
     }
 
-    auto lock = texture_cache.AcquireLock();
+    std::scoped_lock lock{texture_cache.mutex};
     texture_cache.UpdateRenderTargets(true);
     const Framebuffer* const framebuffer = texture_cache.GetFramebuffer();
     const VkExtent2D render_area = framebuffer->RenderArea();
@@ -559,7 +379,6 @@ void RasterizerVulkan::Clear() {
     if (use_stencil) {
         aspect_flags |= VK_IMAGE_ASPECT_STENCIL_BIT;
     }
-
     scheduler.Record([clear_depth = regs.clear_depth, clear_stencil = regs.clear_stencil,
                       clear_rect, aspect_flags](vk::CommandBuffer cmdbuf) {
         VkClearAttachment attachment;
@@ -580,12 +399,11 @@ void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) {
     auto& pipeline = pipeline_cache.GetComputePipeline({
         .shader = code_addr,
         .shared_memory_size = launch_desc.shared_alloc,
-        .workgroup_size =
-            {
-                launch_desc.block_dim_x,
-                launch_desc.block_dim_y,
-                launch_desc.block_dim_z,
-            },
+        .workgroup_size{
+            launch_desc.block_dim_x,
+            launch_desc.block_dim_y,
+            launch_desc.block_dim_z,
+        },
     });
 
     // Compute dispatches can't be executed inside a renderpass
@@ -594,10 +412,21 @@ void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) {
     image_view_indices.clear();
     sampler_handles.clear();
 
-    auto lock = texture_cache.AcquireLock();
-    texture_cache.SynchronizeComputeDescriptors();
+    std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
 
     const auto& entries = pipeline.GetEntries();
+    buffer_cache.SetEnabledComputeUniformBuffers(entries.enabled_uniform_buffers);
+    buffer_cache.UnbindComputeStorageBuffers();
+    u32 ssbo_index = 0;
+    for (const auto& buffer : entries.global_buffers) {
+        buffer_cache.BindComputeStorageBuffer(ssbo_index, buffer.cbuf_index, buffer.cbuf_offset,
+                                              buffer.is_written);
+        ++ssbo_index;
+    }
+    buffer_cache.UpdateComputeBuffers();
+
+    texture_cache.SynchronizeComputeDescriptors();
+
     SetupComputeUniformTexels(entries);
     SetupComputeTextures(entries);
     SetupComputeStorageTexels(entries);
@@ -606,20 +435,15 @@ void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) {
     const std::span indices_span(image_view_indices.data(), image_view_indices.size());
     texture_cache.FillComputeImageViews(indices_span, image_view_ids);
 
-    buffer_cache.Map(CalculateComputeStreamBufferSize());
-
     update_descriptor_queue.Acquire();
 
-    SetupComputeConstBuffers(entries);
-    SetupComputeGlobalBuffers(entries);
+    buffer_cache.BindHostComputeBuffers();
 
     ImageViewId* image_view_id_ptr = image_view_ids.data();
     VkSampler* sampler_ptr = sampler_handles.data();
     PushImageDescriptors(entries, texture_cache, update_descriptor_queue, image_view_id_ptr,
                          sampler_ptr);
 
-    buffer_cache.Unmap();
-
     const VkPipeline pipeline_handle = pipeline.GetHandle();
     const VkPipelineLayout pipeline_layout = pipeline.GetLayout();
     const VkDescriptorSet descriptor_set = pipeline.CommitDescriptorSet();
@@ -644,6 +468,11 @@ void RasterizerVulkan::Query(GPUVAddr gpu_addr, VideoCore::QueryType type,
     query_cache.Query(gpu_addr, type, timestamp);
 }
 
+void RasterizerVulkan::BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr,
+                                                 u32 size) {
+    buffer_cache.BindGraphicsUniformBuffer(stage, index, gpu_addr, size);
+}
+
 void RasterizerVulkan::FlushAll() {}
 
 void RasterizerVulkan::FlushRegion(VAddr addr, u64 size) {
@@ -651,19 +480,23 @@ void RasterizerVulkan::FlushRegion(VAddr addr, u64 size) {
         return;
     }
     {
-        auto lock = texture_cache.AcquireLock();
+        std::scoped_lock lock{texture_cache.mutex};
         texture_cache.DownloadMemory(addr, size);
     }
-    buffer_cache.FlushRegion(addr, size);
+    {
+        std::scoped_lock lock{buffer_cache.mutex};
+        buffer_cache.DownloadMemory(addr, size);
+    }
     query_cache.FlushRegion(addr, size);
 }
 
 bool RasterizerVulkan::MustFlushRegion(VAddr addr, u64 size) {
+    std::scoped_lock lock{texture_cache.mutex, buffer_cache.mutex};
     if (!Settings::IsGPULevelHigh()) {
-        return buffer_cache.MustFlushRegion(addr, size);
+        return buffer_cache.IsRegionGpuModified(addr, size);
     }
     return texture_cache.IsRegionGpuModified(addr, size) ||
-           buffer_cache.MustFlushRegion(addr, size);
+           buffer_cache.IsRegionGpuModified(addr, size);
 }
 
 void RasterizerVulkan::InvalidateRegion(VAddr addr, u64 size) {
@@ -671,11 +504,14 @@ void RasterizerVulkan::InvalidateRegion(VAddr addr, u64 size) {
         return;
     }
     {
-        auto lock = texture_cache.AcquireLock();
+        std::scoped_lock lock{texture_cache.mutex};
         texture_cache.WriteMemory(addr, size);
     }
+    {
+        std::scoped_lock lock{buffer_cache.mutex};
+        buffer_cache.WriteMemory(addr, size);
+    }
     pipeline_cache.InvalidateRegion(addr, size);
-    buffer_cache.InvalidateRegion(addr, size);
     query_cache.InvalidateRegion(addr, size);
 }
 
@@ -683,25 +519,34 @@ void RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) {
     if (addr == 0 || size == 0) {
         return;
     }
+    pipeline_cache.OnCPUWrite(addr, size);
     {
-        auto lock = texture_cache.AcquireLock();
+        std::scoped_lock lock{texture_cache.mutex};
         texture_cache.WriteMemory(addr, size);
     }
-    pipeline_cache.OnCPUWrite(addr, size);
-    buffer_cache.OnCPUWrite(addr, size);
+    {
+        std::scoped_lock lock{buffer_cache.mutex};
+        buffer_cache.CachedWriteMemory(addr, size);
+    }
 }
 
 void RasterizerVulkan::SyncGuestHost() {
-    buffer_cache.SyncGuestHost();
     pipeline_cache.SyncGuestHost();
+    {
+        std::scoped_lock lock{buffer_cache.mutex};
+        buffer_cache.FlushCachedWrites();
+    }
 }
 
 void RasterizerVulkan::UnmapMemory(VAddr addr, u64 size) {
     {
-        auto lock = texture_cache.AcquireLock();
+        std::scoped_lock lock{texture_cache.mutex};
         texture_cache.UnmapMemory(addr, size);
     }
-    buffer_cache.OnCPUWrite(addr, size);
+    {
+        std::scoped_lock lock{buffer_cache.mutex};
+        buffer_cache.WriteMemory(addr, size);
+    }
     pipeline_cache.OnCPUWrite(addr, size);
 }
 
@@ -774,18 +619,21 @@ void RasterizerVulkan::TickFrame() {
     draw_counter = 0;
     update_descriptor_queue.TickFrame();
     fence_manager.TickFrame();
-    buffer_cache.TickFrame();
     staging_pool.TickFrame();
     {
-        auto lock = texture_cache.AcquireLock();
+        std::scoped_lock lock{texture_cache.mutex};
         texture_cache.TickFrame();
     }
+    {
+        std::scoped_lock lock{buffer_cache.mutex};
+        buffer_cache.TickFrame();
+    }
 }
 
 bool RasterizerVulkan::AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Surface& src,
                                              const Tegra::Engines::Fermi2D::Surface& dst,
                                              const Tegra::Engines::Fermi2D::Config& copy_config) {
-    auto lock = texture_cache.AcquireLock();
+    std::scoped_lock lock{texture_cache.mutex};
     texture_cache.BlitImage(dst, src, copy_config);
     return true;
 }
@@ -795,13 +643,11 @@ bool RasterizerVulkan::AccelerateDisplay(const Tegra::FramebufferConfig& config,
     if (!framebuffer_addr) {
         return false;
     }
-
-    auto lock = texture_cache.AcquireLock();
+    std::scoped_lock lock{texture_cache.mutex};
     ImageView* const image_view = texture_cache.TryFindFramebufferImageView(framebuffer_addr);
     if (!image_view) {
         return false;
     }
-
     screen_info.image_view = image_view->Handle(VideoCommon::ImageViewType::e2D);
     screen_info.width = image_view->size.width;
     screen_info.height = image_view->size.height;
@@ -830,29 +676,8 @@ void RasterizerVulkan::FlushWork() {
     draw_counter = 0;
 }
 
-RasterizerVulkan::DrawParameters RasterizerVulkan::SetupGeometry(FixedPipelineState& fixed_state,
-                                                                 BufferBindings& buffer_bindings,
-                                                                 bool is_indexed,
-                                                                 bool is_instanced) {
-    MICROPROFILE_SCOPE(Vulkan_Geometry);
-
-    const auto& regs = maxwell3d.regs;
-
-    SetupVertexArrays(buffer_bindings);
-
-    const u32 base_instance = regs.vb_base_instance;
-    const u32 num_instances = is_instanced ? maxwell3d.mme_draw.instance_count : 1;
-    const u32 base_vertex = is_indexed ? regs.vb_element_base : regs.vertex_buffer.first;
-    const u32 num_vertices = is_indexed ? regs.index_array.count : regs.vertex_buffer.count;
-
-    DrawParameters params{base_instance, num_instances, base_vertex, num_vertices, is_indexed};
-    SetupIndexBuffer(buffer_bindings, params, is_indexed);
-
-    return params;
-}
-
 void RasterizerVulkan::SetupShaderDescriptors(
-    const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders) {
+    const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders, bool is_indexed) {
     image_view_indices.clear();
     sampler_handles.clear();
     for (size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) {
@@ -860,15 +685,27 @@ void RasterizerVulkan::SetupShaderDescriptors(
         if (!shader) {
             continue;
         }
-        const auto& entries = shader->GetEntries();
+        const ShaderEntries& entries = shader->GetEntries();
         SetupGraphicsUniformTexels(entries, stage);
         SetupGraphicsTextures(entries, stage);
         SetupGraphicsStorageTexels(entries, stage);
         SetupGraphicsImages(entries, stage);
+
+        buffer_cache.SetEnabledUniformBuffers(stage, entries.enabled_uniform_buffers);
+        buffer_cache.UnbindGraphicsStorageBuffers(stage);
+        u32 ssbo_index = 0;
+        for (const auto& buffer : entries.global_buffers) {
+            buffer_cache.BindGraphicsStorageBuffer(stage, ssbo_index, buffer.cbuf_index,
+                                                   buffer.cbuf_offset, buffer.is_written);
+            ++ssbo_index;
+        }
     }
     const std::span indices_span(image_view_indices.data(), image_view_indices.size());
+    buffer_cache.UpdateGraphicsBuffers(is_indexed);
     texture_cache.FillGraphicsImageViews(indices_span, image_view_ids);
 
+    buffer_cache.BindHostGeometryBuffers(is_indexed);
+
     update_descriptor_queue.Acquire();
 
     ImageViewId* image_view_id_ptr = image_view_ids.data();
@@ -879,11 +716,9 @@ void RasterizerVulkan::SetupShaderDescriptors(
         if (!shader) {
             continue;
         }
-        const auto& entries = shader->GetEntries();
-        SetupGraphicsConstBuffers(entries, stage);
-        SetupGraphicsGlobalBuffers(entries, stage);
-        PushImageDescriptors(entries, texture_cache, update_descriptor_queue, image_view_id_ptr,
-                             sampler_ptr);
+        buffer_cache.BindHostStageBuffers(stage);
+        PushImageDescriptors(shader->GetEntries(), texture_cache, update_descriptor_queue,
+                             image_view_id_ptr, sampler_ptr);
     }
 }
 
@@ -916,27 +751,11 @@ void RasterizerVulkan::BeginTransformFeedback() {
         LOG_ERROR(Render_Vulkan, "Transform feedbacks used but not supported");
         return;
     }
-
     UNIMPLEMENTED_IF(regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationControl) ||
                      regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::TesselationEval) ||
                      regs.IsShaderConfigEnabled(Maxwell::ShaderProgram::Geometry));
-
-    UNIMPLEMENTED_IF(regs.tfb_bindings[1].buffer_enable);
-    UNIMPLEMENTED_IF(regs.tfb_bindings[2].buffer_enable);
-    UNIMPLEMENTED_IF(regs.tfb_bindings[3].buffer_enable);
-
-    const auto& binding = regs.tfb_bindings[0];
-    UNIMPLEMENTED_IF(binding.buffer_enable == 0);
-    UNIMPLEMENTED_IF(binding.buffer_offset != 0);
-
-    const GPUVAddr gpu_addr = binding.Address();
-    const VkDeviceSize size = static_cast<VkDeviceSize>(binding.buffer_size);
-    const auto info = buffer_cache.UploadMemory(gpu_addr, size, 4, true);
-
-    scheduler.Record([buffer = info.handle, offset = info.offset, size](vk::CommandBuffer cmdbuf) {
-        cmdbuf.BindTransformFeedbackBuffersEXT(0, 1, &buffer, &offset, &size);
-        cmdbuf.BeginTransformFeedbackEXT(0, 0, nullptr, nullptr);
-    });
+    scheduler.Record(
+        [](vk::CommandBuffer cmdbuf) { cmdbuf.BeginTransformFeedbackEXT(0, 0, nullptr, nullptr); });
 }
 
 void RasterizerVulkan::EndTransformFeedback() {
@@ -947,104 +766,11 @@ void RasterizerVulkan::EndTransformFeedback() {
     if (!device.IsExtTransformFeedbackSupported()) {
         return;
     }
-
     scheduler.Record(
         [](vk::CommandBuffer cmdbuf) { cmdbuf.EndTransformFeedbackEXT(0, 0, nullptr, nullptr); });
 }
 
-void RasterizerVulkan::SetupVertexArrays(BufferBindings& buffer_bindings) {
-    const auto& regs = maxwell3d.regs;
-
-    for (size_t index = 0; index < Maxwell::NumVertexArrays; ++index) {
-        const auto& vertex_array = regs.vertex_array[index];
-        if (!vertex_array.IsEnabled()) {
-            continue;
-        }
-        const GPUVAddr start{vertex_array.StartAddress()};
-        const GPUVAddr end{regs.vertex_array_limit[index].LimitAddress()};
-
-        ASSERT(end >= start);
-        const size_t size = end - start;
-        if (size == 0) {
-            buffer_bindings.AddVertexBinding(DefaultBuffer(), 0, DEFAULT_BUFFER_SIZE, 0);
-            continue;
-        }
-        const auto info = buffer_cache.UploadMemory(start, size);
-        buffer_bindings.AddVertexBinding(info.handle, info.offset, size, vertex_array.stride);
-    }
-}
-
-void RasterizerVulkan::SetupIndexBuffer(BufferBindings& buffer_bindings, DrawParameters& params,
-                                        bool is_indexed) {
-    if (params.num_vertices == 0) {
-        return;
-    }
-    const auto& regs = maxwell3d.regs;
-    switch (regs.draw.topology) {
-    case Maxwell::PrimitiveTopology::Quads: {
-        if (!params.is_indexed) {
-            const auto [buffer, offset] =
-                quad_array_pass.Assemble(params.num_vertices, params.base_vertex);
-            buffer_bindings.SetIndexBinding(buffer, offset, VK_INDEX_TYPE_UINT32);
-            params.base_vertex = 0;
-            params.num_vertices = params.num_vertices * 6 / 4;
-            params.is_indexed = true;
-            break;
-        }
-        const GPUVAddr gpu_addr = regs.index_array.IndexStart();
-        const auto info = buffer_cache.UploadMemory(gpu_addr, CalculateIndexBufferSize());
-        VkBuffer buffer = info.handle;
-        u64 offset = info.offset;
-        std::tie(buffer, offset) = quad_indexed_pass.Assemble(
-            regs.index_array.format, params.num_vertices, params.base_vertex, buffer, offset);
-
-        buffer_bindings.SetIndexBinding(buffer, offset, VK_INDEX_TYPE_UINT32);
-        params.num_vertices = (params.num_vertices / 4) * 6;
-        params.base_vertex = 0;
-        break;
-    }
-    default: {
-        if (!is_indexed) {
-            break;
-        }
-        const GPUVAddr gpu_addr = regs.index_array.IndexStart();
-        const auto info = buffer_cache.UploadMemory(gpu_addr, CalculateIndexBufferSize());
-        VkBuffer buffer = info.handle;
-        u64 offset = info.offset;
-
-        auto format = regs.index_array.format;
-        const bool is_uint8 = format == Maxwell::IndexFormat::UnsignedByte;
-        if (is_uint8 && !device.IsExtIndexTypeUint8Supported()) {
-            std::tie(buffer, offset) = uint8_pass.Assemble(params.num_vertices, buffer, offset);
-            format = Maxwell::IndexFormat::UnsignedShort;
-        }
-
-        buffer_bindings.SetIndexBinding(buffer, offset, MaxwellToVK::IndexFormat(device, format));
-        break;
-    }
-    }
-}
-
-void RasterizerVulkan::SetupGraphicsConstBuffers(const ShaderEntries& entries, size_t stage) {
-    MICROPROFILE_SCOPE(Vulkan_ConstBuffers);
-    const auto& shader_stage = maxwell3d.state.shader_stages[stage];
-    for (const auto& entry : entries.const_buffers) {
-        SetupConstBuffer(entry, shader_stage.const_buffers[entry.GetIndex()]);
-    }
-}
-
-void RasterizerVulkan::SetupGraphicsGlobalBuffers(const ShaderEntries& entries, size_t stage) {
-    MICROPROFILE_SCOPE(Vulkan_GlobalBuffers);
-    const auto& cbufs{maxwell3d.state.shader_stages[stage]};
-
-    for (const auto& entry : entries.global_buffers) {
-        const auto addr = cbufs.const_buffers[entry.GetCbufIndex()].address + entry.GetCbufOffset();
-        SetupGlobalBuffer(entry, addr);
-    }
-}
-
 void RasterizerVulkan::SetupGraphicsUniformTexels(const ShaderEntries& entries, size_t stage) {
-    MICROPROFILE_SCOPE(Vulkan_Textures);
     const auto& regs = maxwell3d.regs;
     const bool via_header_index = regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex;
     for (const auto& entry : entries.uniform_texels) {
@@ -1054,7 +780,6 @@ void RasterizerVulkan::SetupGraphicsUniformTexels(const ShaderEntries& entries,
 }
 
 void RasterizerVulkan::SetupGraphicsTextures(const ShaderEntries& entries, size_t stage) {
-    MICROPROFILE_SCOPE(Vulkan_Textures);
     const auto& regs = maxwell3d.regs;
     const bool via_header_index = regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex;
     for (const auto& entry : entries.samplers) {
@@ -1070,7 +795,6 @@ void RasterizerVulkan::SetupGraphicsTextures(const ShaderEntries& entries, size_
 }
 
 void RasterizerVulkan::SetupGraphicsStorageTexels(const ShaderEntries& entries, size_t stage) {
-    MICROPROFILE_SCOPE(Vulkan_Textures);
     const auto& regs = maxwell3d.regs;
     const bool via_header_index = regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex;
     for (const auto& entry : entries.storage_texels) {
@@ -1080,7 +804,6 @@ void RasterizerVulkan::SetupGraphicsStorageTexels(const ShaderEntries& entries,
 }
 
 void RasterizerVulkan::SetupGraphicsImages(const ShaderEntries& entries, size_t stage) {
-    MICROPROFILE_SCOPE(Vulkan_Images);
     const auto& regs = maxwell3d.regs;
     const bool via_header_index = regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex;
     for (const auto& entry : entries.images) {
@@ -1089,32 +812,7 @@ void RasterizerVulkan::SetupGraphicsImages(const ShaderEntries& entries, size_t
     }
 }
 
-void RasterizerVulkan::SetupComputeConstBuffers(const ShaderEntries& entries) {
-    MICROPROFILE_SCOPE(Vulkan_ConstBuffers);
-    const auto& launch_desc = kepler_compute.launch_description;
-    for (const auto& entry : entries.const_buffers) {
-        const auto& config = launch_desc.const_buffer_config[entry.GetIndex()];
-        const std::bitset<8> mask = launch_desc.const_buffer_enable_mask.Value();
-        const Tegra::Engines::ConstBufferInfo info{
-            .address = config.Address(),
-            .size = config.size,
-            .enabled = mask[entry.GetIndex()],
-        };
-        SetupConstBuffer(entry, info);
-    }
-}
-
-void RasterizerVulkan::SetupComputeGlobalBuffers(const ShaderEntries& entries) {
-    MICROPROFILE_SCOPE(Vulkan_GlobalBuffers);
-    const auto& cbufs{kepler_compute.launch_description.const_buffer_config};
-    for (const auto& entry : entries.global_buffers) {
-        const auto addr{cbufs[entry.GetCbufIndex()].Address() + entry.GetCbufOffset()};
-        SetupGlobalBuffer(entry, addr);
-    }
-}
-
 void RasterizerVulkan::SetupComputeUniformTexels(const ShaderEntries& entries) {
-    MICROPROFILE_SCOPE(Vulkan_Textures);
     const bool via_header_index = kepler_compute.launch_description.linked_tsc;
     for (const auto& entry : entries.uniform_texels) {
         const TextureHandle handle =
@@ -1124,7 +822,6 @@ void RasterizerVulkan::SetupComputeUniformTexels(const ShaderEntries& entries) {
 }
 
 void RasterizerVulkan::SetupComputeTextures(const ShaderEntries& entries) {
-    MICROPROFILE_SCOPE(Vulkan_Textures);
     const bool via_header_index = kepler_compute.launch_description.linked_tsc;
     for (const auto& entry : entries.samplers) {
         for (size_t index = 0; index < entry.size; ++index) {
@@ -1139,7 +836,6 @@ void RasterizerVulkan::SetupComputeTextures(const ShaderEntries& entries) {
 }
 
 void RasterizerVulkan::SetupComputeStorageTexels(const ShaderEntries& entries) {
-    MICROPROFILE_SCOPE(Vulkan_Textures);
     const bool via_header_index = kepler_compute.launch_description.linked_tsc;
     for (const auto& entry : entries.storage_texels) {
         const TextureHandle handle =
@@ -1149,7 +845,6 @@ void RasterizerVulkan::SetupComputeStorageTexels(const ShaderEntries& entries) {
 }
 
 void RasterizerVulkan::SetupComputeImages(const ShaderEntries& entries) {
-    MICROPROFILE_SCOPE(Vulkan_Images);
     const bool via_header_index = kepler_compute.launch_description.linked_tsc;
     for (const auto& entry : entries.images) {
         const TextureHandle handle =
@@ -1158,42 +853,6 @@ void RasterizerVulkan::SetupComputeImages(const ShaderEntries& entries) {
     }
 }
 
-void RasterizerVulkan::SetupConstBuffer(const ConstBufferEntry& entry,
-                                        const Tegra::Engines::ConstBufferInfo& buffer) {
-    if (!buffer.enabled) {
-        // Set values to zero to unbind buffers
-        update_descriptor_queue.AddBuffer(DefaultBuffer(), 0, DEFAULT_BUFFER_SIZE);
-        return;
-    }
-    // Align the size to avoid bad std140 interactions
-    const size_t size = Common::AlignUp(CalculateConstBufferSize(entry, buffer), 4 * sizeof(float));
-    ASSERT(size <= MaxConstbufferSize);
-
-    const u64 alignment = device.GetUniformBufferAlignment();
-    const auto info = buffer_cache.UploadMemory(buffer.address, size, alignment);
-    update_descriptor_queue.AddBuffer(info.handle, info.offset, size);
-}
-
-void RasterizerVulkan::SetupGlobalBuffer(const GlobalBufferEntry& entry, GPUVAddr address) {
-    const u64 actual_addr = gpu_memory.Read<u64>(address);
-    const u32 size = gpu_memory.Read<u32>(address + 8);
-
-    if (size == 0) {
-        // Sometimes global memory pointers don't have a proper size. Upload a dummy entry
-        // because Vulkan doesn't like empty buffers.
-        // Note: Do *not* use DefaultBuffer() here, storage buffers can be written breaking the
-        // default buffer.
-        static constexpr size_t dummy_size = 4;
-        const auto info = buffer_cache.GetEmptyBuffer(dummy_size);
-        update_descriptor_queue.AddBuffer(info.handle, info.offset, dummy_size);
-        return;
-    }
-
-    const auto info = buffer_cache.UploadMemory(
-        actual_addr, size, device.GetStorageBufferAlignment(), entry.IsWritten());
-    update_descriptor_queue.AddBuffer(info.handle, info.offset, size);
-}
-
 void RasterizerVulkan::UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& regs) {
     if (!state_tracker.TouchViewports()) {
         return;
@@ -1206,7 +865,8 @@ void RasterizerVulkan::UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& reg
         GetViewportState(device, regs, 8),  GetViewportState(device, regs, 9),
         GetViewportState(device, regs, 10), GetViewportState(device, regs, 11),
         GetViewportState(device, regs, 12), GetViewportState(device, regs, 13),
-        GetViewportState(device, regs, 14), GetViewportState(device, regs, 15)};
+        GetViewportState(device, regs, 14), GetViewportState(device, regs, 15),
+    };
     scheduler.Record([viewports](vk::CommandBuffer cmdbuf) { cmdbuf.SetViewport(0, viewports); });
 }
 
@@ -1214,13 +874,14 @@ void RasterizerVulkan::UpdateScissorsState(Tegra::Engines::Maxwell3D::Regs& regs
     if (!state_tracker.TouchScissors()) {
         return;
     }
-    const std::array scissors = {
+    const std::array scissors{
         GetScissorState(regs, 0),  GetScissorState(regs, 1),  GetScissorState(regs, 2),
         GetScissorState(regs, 3),  GetScissorState(regs, 4),  GetScissorState(regs, 5),
         GetScissorState(regs, 6),  GetScissorState(regs, 7),  GetScissorState(regs, 8),
         GetScissorState(regs, 9),  GetScissorState(regs, 10), GetScissorState(regs, 11),
         GetScissorState(regs, 12), GetScissorState(regs, 13), GetScissorState(regs, 14),
-        GetScissorState(regs, 15)};
+        GetScissorState(regs, 15),
+    };
     scheduler.Record([scissors](vk::CommandBuffer cmdbuf) { cmdbuf.SetScissor(0, scissors); });
 }
 
@@ -1385,73 +1046,4 @@ void RasterizerVulkan::UpdateStencilTestEnable(Tegra::Engines::Maxwell3D::Regs&
     });
 }
 
-size_t RasterizerVulkan::CalculateGraphicsStreamBufferSize(bool is_indexed) const {
-    size_t size = CalculateVertexArraysSize();
-    if (is_indexed) {
-        size = Common::AlignUp(size, 4) + CalculateIndexBufferSize();
-    }
-    size += Maxwell::MaxConstBuffers * (MaxConstbufferSize + device.GetUniformBufferAlignment());
-    return size;
-}
-
-size_t RasterizerVulkan::CalculateComputeStreamBufferSize() const {
-    return Tegra::Engines::KeplerCompute::NumConstBuffers *
-           (Maxwell::MaxConstBufferSize + device.GetUniformBufferAlignment());
-}
-
-size_t RasterizerVulkan::CalculateVertexArraysSize() const {
-    const auto& regs = maxwell3d.regs;
-
-    size_t size = 0;
-    for (u32 index = 0; index < Maxwell::NumVertexArrays; ++index) {
-        // This implementation assumes that all attributes are used in the shader.
-        const GPUVAddr start{regs.vertex_array[index].StartAddress()};
-        const GPUVAddr end{regs.vertex_array_limit[index].LimitAddress()};
-        DEBUG_ASSERT(end >= start);
-
-        size += (end - start) * regs.vertex_array[index].enable;
-    }
-    return size;
-}
-
-size_t RasterizerVulkan::CalculateIndexBufferSize() const {
-    return static_cast<size_t>(maxwell3d.regs.index_array.count) *
-           static_cast<size_t>(maxwell3d.regs.index_array.FormatSizeInBytes());
-}
-
-size_t RasterizerVulkan::CalculateConstBufferSize(
-    const ConstBufferEntry& entry, const Tegra::Engines::ConstBufferInfo& buffer) const {
-    if (entry.IsIndirect()) {
-        // Buffer is accessed indirectly, so upload the entire thing
-        return buffer.size;
-    } else {
-        // Buffer is accessed directly, upload just what we use
-        return entry.GetSize();
-    }
-}
-
-VkBuffer RasterizerVulkan::DefaultBuffer() {
-    if (default_buffer) {
-        return *default_buffer;
-    }
-    default_buffer = device.GetLogical().CreateBuffer({
-        .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
-        .pNext = nullptr,
-        .flags = 0,
-        .size = DEFAULT_BUFFER_SIZE,
-        .usage = VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_VERTEX_BUFFER_BIT |
-                 VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT,
-        .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
-        .queueFamilyIndexCount = 0,
-        .pQueueFamilyIndices = nullptr,
-    });
-    default_buffer_commit = memory_allocator.Commit(default_buffer, MemoryUsage::DeviceLocal);
-
-    scheduler.RequestOutsideRenderPassOperationContext();
-    scheduler.Record([buffer = *default_buffer](vk::CommandBuffer cmdbuf) {
-        cmdbuf.FillBuffer(buffer, 0, DEFAULT_BUFFER_SIZE, 0);
-    });
-    return *default_buffer;
-}
-
 } // namespace Vulkan
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h
index 8e261b9bd..7fc6741da 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.h
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.h
@@ -18,14 +18,12 @@
 #include "video_core/renderer_vulkan/blit_image.h"
 #include "video_core/renderer_vulkan/fixed_pipeline_state.h"
 #include "video_core/renderer_vulkan/vk_buffer_cache.h"
-#include "video_core/renderer_vulkan/vk_compute_pass.h"
 #include "video_core/renderer_vulkan/vk_descriptor_pool.h"
 #include "video_core/renderer_vulkan/vk_fence_manager.h"
 #include "video_core/renderer_vulkan/vk_pipeline_cache.h"
 #include "video_core/renderer_vulkan/vk_query_cache.h"
 #include "video_core/renderer_vulkan/vk_scheduler.h"
 #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
-#include "video_core/renderer_vulkan/vk_stream_buffer.h"
 #include "video_core/renderer_vulkan/vk_texture_cache.h"
 #include "video_core/renderer_vulkan/vk_update_descriptor.h"
 #include "video_core/shader/async_shaders.h"
@@ -49,7 +47,6 @@ namespace Vulkan {
 struct VKScreenInfo;
 
 class StateTracker;
-class BufferBindings;
 
 class RasterizerVulkan final : public VideoCore::RasterizerAccelerated {
 public:
@@ -65,6 +62,7 @@ public:
     void DispatchCompute(GPUVAddr code_addr) override;
     void ResetCounter(VideoCore::QueryType type) override;
     void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) override;
+    void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size) override;
     void FlushAll() override;
     void FlushRegion(VAddr addr, u64 size) override;
     bool MustFlushRegion(VAddr addr, u64 size) override;
@@ -107,24 +105,11 @@ private:
 
     static constexpr VkDeviceSize DEFAULT_BUFFER_SIZE = 4 * sizeof(float);
 
-    struct DrawParameters {
-        void Draw(vk::CommandBuffer cmdbuf) const;
-
-        u32 base_instance = 0;
-        u32 num_instances = 0;
-        u32 base_vertex = 0;
-        u32 num_vertices = 0;
-        bool is_indexed = 0;
-    };
-
     void FlushWork();
 
-    /// Setups geometry buffers and state.
-    DrawParameters SetupGeometry(FixedPipelineState& fixed_state, BufferBindings& buffer_bindings,
-                                 bool is_indexed, bool is_instanced);
-
     /// Setup descriptors in the graphics pipeline.
-    void SetupShaderDescriptors(const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders);
+    void SetupShaderDescriptors(const std::array<Shader*, Maxwell::MaxShaderProgram>& shaders,
+                                bool is_indexed);
 
     void UpdateDynamicStates();
 
@@ -132,16 +117,6 @@ private:
 
     void EndTransformFeedback();
 
-    void SetupVertexArrays(BufferBindings& buffer_bindings);
-
-    void SetupIndexBuffer(BufferBindings& buffer_bindings, DrawParameters& params, bool is_indexed);
-
-    /// Setup constant buffers in the graphics pipeline.
-    void SetupGraphicsConstBuffers(const ShaderEntries& entries, std::size_t stage);
-
-    /// Setup global buffers in the graphics pipeline.
-    void SetupGraphicsGlobalBuffers(const ShaderEntries& entries, std::size_t stage);
-
     /// Setup uniform texels in the graphics pipeline.
     void SetupGraphicsUniformTexels(const ShaderEntries& entries, std::size_t stage);
 
@@ -154,12 +129,6 @@ private:
     /// Setup images in the graphics pipeline.
     void SetupGraphicsImages(const ShaderEntries& entries, std::size_t stage);
 
-    /// Setup constant buffers in the compute pipeline.
-    void SetupComputeConstBuffers(const ShaderEntries& entries);
-
-    /// Setup global buffers in the compute pipeline.
-    void SetupComputeGlobalBuffers(const ShaderEntries& entries);
-
     /// Setup texel buffers in the compute pipeline.
     void SetupComputeUniformTexels(const ShaderEntries& entries);
 
@@ -172,11 +141,6 @@ private:
     /// Setup images in the compute pipeline.
     void SetupComputeImages(const ShaderEntries& entries);
 
-    void SetupConstBuffer(const ConstBufferEntry& entry,
-                          const Tegra::Engines::ConstBufferInfo& buffer);
-
-    void SetupGlobalBuffer(const GlobalBufferEntry& entry, GPUVAddr address);
-
     void UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& regs);
     void UpdateScissorsState(Tegra::Engines::Maxwell3D::Regs& regs);
     void UpdateDepthBias(Tegra::Engines::Maxwell3D::Regs& regs);
@@ -193,19 +157,6 @@ private:
     void UpdateStencilOp(Tegra::Engines::Maxwell3D::Regs& regs);
     void UpdateStencilTestEnable(Tegra::Engines::Maxwell3D::Regs& regs);
 
-    size_t CalculateGraphicsStreamBufferSize(bool is_indexed) const;
-
-    size_t CalculateComputeStreamBufferSize() const;
-
-    size_t CalculateVertexArraysSize() const;
-
-    size_t CalculateIndexBufferSize() const;
-
-    size_t CalculateConstBufferSize(const ConstBufferEntry& entry,
-                                    const Tegra::Engines::ConstBufferInfo& buffer) const;
-
-    VkBuffer DefaultBuffer();
-
     Tegra::GPU& gpu;
     Tegra::MemoryManager& gpu_memory;
     Tegra::Engines::Maxwell3D& maxwell3d;
@@ -217,24 +168,19 @@ private:
     StateTracker& state_tracker;
     VKScheduler& scheduler;
 
-    VKStreamBuffer stream_buffer;
     StagingBufferPool staging_pool;
     VKDescriptorPool descriptor_pool;
     VKUpdateDescriptorQueue update_descriptor_queue;
     BlitImageHelper blit_image;
-    QuadArrayPass quad_array_pass;
-    QuadIndexedPass quad_indexed_pass;
-    Uint8Pass uint8_pass;
 
     TextureCacheRuntime texture_cache_runtime;
     TextureCache texture_cache;
+    BufferCacheRuntime buffer_cache_runtime;
+    BufferCache buffer_cache;
     VKPipelineCache pipeline_cache;
-    VKBufferCache buffer_cache;
     VKQueryCache query_cache;
     VKFenceManager fence_manager;
 
-    vk::Buffer default_buffer;
-    MemoryCommit default_buffer_commit;
     vk::Event wfi_event;
     VideoCommon::Shader::AsyncShaders async_shaders;
 
diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp
index 66004f9c0..f35c120b0 100644
--- a/src/video_core/renderer_vulkan/vk_scheduler.cpp
+++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp
@@ -52,18 +52,6 @@ VKScheduler::~VKScheduler() {
     worker_thread.join();
 }
 
-u64 VKScheduler::CurrentTick() const noexcept {
-    return master_semaphore->CurrentTick();
-}
-
-bool VKScheduler::IsFree(u64 tick) const noexcept {
-    return master_semaphore->IsFree(tick);
-}
-
-void VKScheduler::Wait(u64 tick) {
-    master_semaphore->Wait(tick);
-}
-
 void VKScheduler::Flush(VkSemaphore semaphore) {
     SubmitExecution(semaphore);
     AllocateNewContext();
@@ -269,7 +257,7 @@ void VKScheduler::EndRenderPass() {
         cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
                                    VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
                                    VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
-                               VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT, 0, nullptr, nullptr,
+                               VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, nullptr, nullptr,
                                vk::Span(barriers.data(), num_images));
     });
     state.renderpass = nullptr;
diff --git a/src/video_core/renderer_vulkan/vk_scheduler.h b/src/video_core/renderer_vulkan/vk_scheduler.h
index 15f2987eb..3ce48e9d2 100644
--- a/src/video_core/renderer_vulkan/vk_scheduler.h
+++ b/src/video_core/renderer_vulkan/vk_scheduler.h
@@ -14,6 +14,7 @@
 #include "common/alignment.h"
 #include "common/common_types.h"
 #include "common/threadsafe_queue.h"
+#include "video_core/renderer_vulkan/vk_master_semaphore.h"
 #include "video_core/vulkan_common/vulkan_wrapper.h"
 
 namespace Vulkan {
@@ -21,7 +22,6 @@ namespace Vulkan {
 class CommandPool;
 class Device;
 class Framebuffer;
-class MasterSemaphore;
 class StateTracker;
 class VKQueryCache;
 
@@ -32,15 +32,6 @@ public:
     explicit VKScheduler(const Device& device, StateTracker& state_tracker);
     ~VKScheduler();
 
-    /// Returns the current command buffer tick.
-    [[nodiscard]] u64 CurrentTick() const noexcept;
-
-    /// Returns true when a tick has been triggered by the GPU.
-    [[nodiscard]] bool IsFree(u64 tick) const noexcept;
-
-    /// Waits for the given tick to trigger on the GPU.
-    void Wait(u64 tick);
-
     /// Sends the current execution context to the GPU.
     void Flush(VkSemaphore semaphore = nullptr);
 
@@ -82,6 +73,21 @@ public:
         (void)chunk->Record(command);
     }
 
+    /// Returns the current command buffer tick.
+    [[nodiscard]] u64 CurrentTick() const noexcept {
+        return master_semaphore->CurrentTick();
+    }
+
+    /// Returns true when a tick has been triggered by the GPU.
+    [[nodiscard]] bool IsFree(u64 tick) const noexcept {
+        return master_semaphore->IsFree(tick);
+    }
+
+    /// Waits for the given tick to trigger on the GPU.
+    void Wait(u64 tick) {
+        master_semaphore->Wait(tick);
+    }
+
     /// Returns the master timeline semaphore.
     [[nodiscard]] MasterSemaphore& GetMasterSemaphore() const noexcept {
         return *master_semaphore;
diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
index 61d52b961..e165a6987 100644
--- a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
+++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
@@ -3127,6 +3127,9 @@ ShaderEntries GenerateShaderEntries(const VideoCommon::Shader::ShaderIR& ir) {
             entries.attributes.insert(GetGenericAttributeLocation(attribute));
         }
     }
+    for (const auto& buffer : entries.const_buffers) {
+        entries.enabled_uniform_buffers |= 1U << buffer.GetIndex();
+    }
     entries.clip_distances = ir.GetClipDistances();
     entries.shader_length = ir.GetLength();
     entries.uses_warps = ir.UsesWarps();
diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.h b/src/video_core/renderer_vulkan/vk_shader_decompiler.h
index 26381e444..5d94132a5 100644
--- a/src/video_core/renderer_vulkan/vk_shader_decompiler.h
+++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.h
@@ -39,24 +39,7 @@ private:
     u32 index{};
 };
 
-class GlobalBufferEntry {
-public:
-    constexpr explicit GlobalBufferEntry(u32 cbuf_index_, u32 cbuf_offset_, bool is_written_)
-        : cbuf_index{cbuf_index_}, cbuf_offset{cbuf_offset_}, is_written{is_written_} {}
-
-    constexpr u32 GetCbufIndex() const {
-        return cbuf_index;
-    }
-
-    constexpr u32 GetCbufOffset() const {
-        return cbuf_offset;
-    }
-
-    constexpr bool IsWritten() const {
-        return is_written;
-    }
-
-private:
+struct GlobalBufferEntry {
     u32 cbuf_index{};
     u32 cbuf_offset{};
     bool is_written{};
@@ -78,6 +61,7 @@ struct ShaderEntries {
     std::set<u32> attributes;
     std::array<bool, Maxwell::NumClipDistances> clip_distances{};
     std::size_t shader_length{};
+    u32 enabled_uniform_buffers{};
     bool uses_warps{};
 };
 
diff --git a/src/video_core/renderer_vulkan/vk_state_tracker.cpp b/src/video_core/renderer_vulkan/vk_state_tracker.cpp
index 1779a2e30..e81fad007 100644
--- a/src/video_core/renderer_vulkan/vk_state_tracker.cpp
+++ b/src/video_core/renderer_vulkan/vk_state_tracker.cpp
@@ -30,15 +30,18 @@ using Table = Maxwell3D::DirtyState::Table;
 using Flags = Maxwell3D::DirtyState::Flags;
 
 Flags MakeInvalidationFlags() {
-    static constexpr std::array INVALIDATION_FLAGS{
+    static constexpr int INVALIDATION_FLAGS[]{
         Viewports,         Scissors,  DepthBias,         BlendConstants,    DepthBounds,
         StencilProperties, CullMode,  DepthBoundsEnable, DepthTestEnable,   DepthWriteEnable,
-        DepthCompareOp,    FrontFace, StencilOp,         StencilTestEnable,
+        DepthCompareOp,    FrontFace, StencilOp,         StencilTestEnable, VertexBuffers,
     };
     Flags flags{};
     for (const int flag : INVALIDATION_FLAGS) {
         flags[flag] = true;
     }
+    for (int index = VertexBuffer0; index <= VertexBuffer31; ++index) {
+        flags[index] = true;
+    }
     return flags;
 }
 
@@ -130,7 +133,7 @@ void SetupDirtyStencilTestEnable(Tables& tables) {
 StateTracker::StateTracker(Tegra::GPU& gpu)
     : flags{gpu.Maxwell3D().dirty.flags}, invalidation_flags{MakeInvalidationFlags()} {
     auto& tables = gpu.Maxwell3D().dirty.tables;
-    SetupDirtyRenderTargets(tables);
+    SetupDirtyFlags(tables);
     SetupDirtyViewports(tables);
     SetupDirtyScissors(tables);
     SetupDirtyDepthBias(tables);
diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
index aa7c5d7c6..1eeb45ca9 100644
--- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
@@ -426,46 +426,47 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
 void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage image,
                        VkImageAspectFlags aspect_mask, bool is_initialized,
                        std::span<const VkBufferImageCopy> copies) {
-    static constexpr VkAccessFlags ACCESS_FLAGS = VK_ACCESS_SHADER_WRITE_BIT |
-                                                  VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
-                                                  VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
+    static constexpr VkAccessFlags WRITE_ACCESS_FLAGS =
+        VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
+        VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
+    static constexpr VkAccessFlags READ_ACCESS_FLAGS = VK_ACCESS_SHADER_READ_BIT |
+                                                       VK_ACCESS_COLOR_ATTACHMENT_READ_BIT |
+                                                       VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT;
     const VkImageMemoryBarrier read_barrier{
         .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
         .pNext = nullptr,
-        .srcAccessMask = ACCESS_FLAGS,
+        .srcAccessMask = WRITE_ACCESS_FLAGS,
         .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
         .oldLayout = is_initialized ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED,
         .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
         .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
         .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
         .image = image,
-        .subresourceRange =
-            {
-                .aspectMask = aspect_mask,
-                .baseMipLevel = 0,
-                .levelCount = VK_REMAINING_MIP_LEVELS,
-                .baseArrayLayer = 0,
-                .layerCount = VK_REMAINING_ARRAY_LAYERS,
-            },
+        .subresourceRange{
+            .aspectMask = aspect_mask,
+            .baseMipLevel = 0,
+            .levelCount = VK_REMAINING_MIP_LEVELS,
+            .baseArrayLayer = 0,
+            .layerCount = VK_REMAINING_ARRAY_LAYERS,
+        },
     };
     const VkImageMemoryBarrier write_barrier{
         .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
         .pNext = nullptr,
         .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
-        .dstAccessMask = ACCESS_FLAGS,
+        .dstAccessMask = WRITE_ACCESS_FLAGS | READ_ACCESS_FLAGS,
         .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
         .newLayout = VK_IMAGE_LAYOUT_GENERAL,
         .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
         .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
         .image = image,
-        .subresourceRange =
-            {
-                .aspectMask = aspect_mask,
-                .baseMipLevel = 0,
-                .levelCount = VK_REMAINING_MIP_LEVELS,
-                .baseArrayLayer = 0,
-                .layerCount = VK_REMAINING_ARRAY_LAYERS,
-            },
+        .subresourceRange{
+            .aspectMask = aspect_mask,
+            .baseMipLevel = 0,
+            .levelCount = VK_REMAINING_MIP_LEVELS,
+            .baseArrayLayer = 0,
+            .layerCount = VK_REMAINING_ARRAY_LAYERS,
+        },
     };
     cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
                            read_barrier);
@@ -569,20 +570,12 @@ void TextureCacheRuntime::Finish() {
     scheduler.Finish();
 }
 
-ImageBufferMap TextureCacheRuntime::MapUploadBuffer(size_t size) {
-    const auto staging_ref = staging_buffer_pool.Request(size, MemoryUsage::Upload);
-    return {
-        .handle = staging_ref.buffer,
-        .span = staging_ref.mapped_span,
-    };
+StagingBufferRef TextureCacheRuntime::UploadStagingBuffer(size_t size) {
+    return staging_buffer_pool.Request(size, MemoryUsage::Upload);
 }
 
-ImageBufferMap TextureCacheRuntime::MapDownloadBuffer(size_t size) {
-    const auto staging_ref = staging_buffer_pool.Request(size, MemoryUsage::Download);
-    return {
-        .handle = staging_ref.buffer,
-        .span = staging_ref.mapped_span,
-    };
+StagingBufferRef TextureCacheRuntime::DownloadStagingBuffer(size_t size) {
+    return staging_buffer_pool.Request(size, MemoryUsage::Download);
 }
 
 void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst, ImageView& src,
@@ -754,7 +747,7 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src,
                 .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
                                  VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT |
                                  VK_ACCESS_TRANSFER_WRITE_BIT,
-                .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+                .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
                 .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
                 .newLayout = VK_IMAGE_LAYOUT_GENERAL,
                 .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
@@ -765,12 +758,9 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src,
             VkImageMemoryBarrier{
                 .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
                 .pNext = nullptr,
-                .srcAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT |
-                                 VK_ACCESS_COLOR_ATTACHMENT_READ_BIT |
-                                 VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
-                                 VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT |
+                .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
                                  VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT |
-                                 VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT,
+                                 VK_ACCESS_TRANSFER_WRITE_BIT,
                 .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
                 .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
                 .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
@@ -828,12 +818,12 @@ Image::Image(TextureCacheRuntime& runtime, const ImageInfo& info_, GPUVAddr gpu_
     }
 }
 
-void Image::UploadMemory(const ImageBufferMap& map, size_t buffer_offset,
+void Image::UploadMemory(const StagingBufferRef& map, size_t buffer_offset,
                          std::span<const BufferImageCopy> copies) {
     // TODO: Move this to another API
     scheduler->RequestOutsideRenderPassOperationContext();
     std::vector vk_copies = TransformBufferImageCopies(copies, buffer_offset, aspect_mask);
-    const VkBuffer src_buffer = map.handle;
+    const VkBuffer src_buffer = map.buffer;
     const VkImage vk_image = *image;
     const VkImageAspectFlags vk_aspect_mask = aspect_mask;
     const bool is_initialized = std::exchange(initialized, true);
@@ -843,12 +833,12 @@ void Image::UploadMemory(const ImageBufferMap& map, size_t buffer_offset,
     });
 }
 
-void Image::UploadMemory(const ImageBufferMap& map, size_t buffer_offset,
+void Image::UploadMemory(const StagingBufferRef& map, size_t buffer_offset,
                          std::span<const VideoCommon::BufferCopy> copies) {
     // TODO: Move this to another API
     scheduler->RequestOutsideRenderPassOperationContext();
     std::vector vk_copies = TransformBufferCopies(copies, buffer_offset);
-    const VkBuffer src_buffer = map.handle;
+    const VkBuffer src_buffer = map.buffer;
     const VkBuffer dst_buffer = *buffer;
     scheduler->Record([src_buffer, dst_buffer, vk_copies](vk::CommandBuffer cmdbuf) {
         // TODO: Barriers
@@ -856,13 +846,58 @@ void Image::UploadMemory(const ImageBufferMap& map, size_t buffer_offset,
     });
 }
 
-void Image::DownloadMemory(const ImageBufferMap& map, size_t buffer_offset,
+void Image::DownloadMemory(const StagingBufferRef& map, size_t buffer_offset,
                            std::span<const BufferImageCopy> copies) {
     std::vector vk_copies = TransformBufferImageCopies(copies, buffer_offset, aspect_mask);
-    scheduler->Record([buffer = map.handle, image = *image, aspect_mask = aspect_mask,
+    scheduler->Record([buffer = map.buffer, image = *image, aspect_mask = aspect_mask,
                        vk_copies](vk::CommandBuffer cmdbuf) {
-        // TODO: Barriers
-        cmdbuf.CopyImageToBuffer(image, VK_IMAGE_LAYOUT_GENERAL, buffer, vk_copies);
+        const VkImageMemoryBarrier read_barrier{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
+            .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
+            .newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = image,
+            .subresourceRange{
+                .aspectMask = aspect_mask,
+                .baseMipLevel = 0,
+                .levelCount = VK_REMAINING_MIP_LEVELS,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            },
+        };
+        const VkImageMemoryBarrier image_write_barrier{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = 0,
+            .dstAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
+            .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
+            .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = image,
+            .subresourceRange{
+                .aspectMask = aspect_mask,
+                .baseMipLevel = 0,
+                .levelCount = VK_REMAINING_MIP_LEVELS,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            },
+        };
+        const VkMemoryBarrier memory_write_barrier{
+            .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
+        };
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
+                               0, read_barrier);
+        cmdbuf.CopyImageToBuffer(image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, buffer, vk_copies);
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
+                               0, memory_write_barrier, nullptr, image_write_barrier);
     });
 }
 
@@ -1127,7 +1162,7 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM
         .pAttachments = attachments.data(),
         .width = key.size.width,
         .height = key.size.height,
-        .layers = static_cast<u32>(num_layers),
+        .layers = static_cast<u32>(std::max(num_layers, 1)),
     });
     if (runtime.device.HasDebuggingToolAttached()) {
         framebuffer.SetObjectNameEXT(VideoCommon::Name(key).c_str());
diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h
index 8d29361a1..4558c3297 100644
--- a/src/video_core/renderer_vulkan/vk_texture_cache.h
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.h
@@ -7,6 +7,7 @@
 #include <compare>
 #include <span>
 
+#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
 #include "video_core/texture_cache/texture_cache.h"
 #include "video_core/vulkan_common/vulkan_memory_allocator.h"
 #include "video_core/vulkan_common/vulkan_wrapper.h"
@@ -53,19 +54,6 @@ struct hash<Vulkan::RenderPassKey> {
 
 namespace Vulkan {
 
-struct ImageBufferMap {
-    [[nodiscard]] VkBuffer Handle() const noexcept {
-        return handle;
-    }
-
-    [[nodiscard]] std::span<u8> Span() const noexcept {
-        return span;
-    }
-
-    VkBuffer handle;
-    std::span<u8> span;
-};
-
 struct TextureCacheRuntime {
     const Device& device;
     VKScheduler& scheduler;
@@ -76,9 +64,9 @@ struct TextureCacheRuntime {
 
     void Finish();
 
-    [[nodiscard]] ImageBufferMap MapUploadBuffer(size_t size);
+    [[nodiscard]] StagingBufferRef UploadStagingBuffer(size_t size);
 
-    [[nodiscard]] ImageBufferMap MapDownloadBuffer(size_t size);
+    [[nodiscard]] StagingBufferRef DownloadStagingBuffer(size_t size);
 
     void BlitImage(Framebuffer* dst_framebuffer, ImageView& dst, ImageView& src,
                    const std::array<Offset2D, 2>& dst_region,
@@ -94,7 +82,7 @@ struct TextureCacheRuntime {
         return false;
     }
 
-    void AccelerateImageUpload(Image&, const ImageBufferMap&, size_t,
+    void AccelerateImageUpload(Image&, const StagingBufferRef&, size_t,
                                std::span<const VideoCommon::SwizzleParameters>) {
         UNREACHABLE();
     }
@@ -112,13 +100,13 @@ public:
     explicit Image(TextureCacheRuntime&, const VideoCommon::ImageInfo& info, GPUVAddr gpu_addr,
                    VAddr cpu_addr);
 
-    void UploadMemory(const ImageBufferMap& map, size_t buffer_offset,
+    void UploadMemory(const StagingBufferRef& map, size_t buffer_offset,
                       std::span<const VideoCommon::BufferImageCopy> copies);
 
-    void UploadMemory(const ImageBufferMap& map, size_t buffer_offset,
+    void UploadMemory(const StagingBufferRef& map, size_t buffer_offset,
                       std::span<const VideoCommon::BufferCopy> copies);
 
-    void DownloadMemory(const ImageBufferMap& map, size_t buffer_offset,
+    void DownloadMemory(const StagingBufferRef& map, size_t buffer_offset,
                         std::span<const VideoCommon::BufferImageCopy> copies);
 
     [[nodiscard]] VkImage Handle() const noexcept {
diff --git a/src/video_core/shader/async_shaders.h b/src/video_core/shader/async_shaders.h
index 0dbb1a31f..7fdff6e56 100644
--- a/src/video_core/shader/async_shaders.h
+++ b/src/video_core/shader/async_shaders.h
@@ -9,16 +9,7 @@
 #include <shared_mutex>
 #include <thread>
 
-// This header includes both Vulkan and OpenGL headers, this has to be fixed
-// Unfortunately, including OpenGL will include Windows.h that defines macros that can cause issues.
-// Forcefully include glad early and undefine macros
 #include <glad/glad.h>
-#ifdef CreateEvent
-#undef CreateEvent
-#endif
-#ifdef CreateSemaphore
-#undef CreateSemaphore
-#endif
 
 #include "common/common_types.h"
 #include "video_core/renderer_opengl/gl_device.h"
diff --git a/src/video_core/shader/decode/other.cpp b/src/video_core/shader/decode/other.cpp
index d3ea07aac..5f88537bc 100644
--- a/src/video_core/shader/decode/other.cpp
+++ b/src/video_core/shader/decode/other.cpp
@@ -76,6 +76,7 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
             case SystemVariable::InvocationId:
                 return Operation(OperationCode::InvocationId);
             case SystemVariable::Ydirection:
+                uses_y_negate = true;
                 return Operation(OperationCode::YNegate);
             case SystemVariable::InvocationInfo:
                 LOG_WARNING(HW_GPU, "S2R instruction with InvocationInfo is incomplete");
diff --git a/src/video_core/shader/shader_ir.h b/src/video_core/shader/shader_ir.h
index 0c6ab0f07..1cd7c14d7 100644
--- a/src/video_core/shader/shader_ir.h
+++ b/src/video_core/shader/shader_ir.h
@@ -139,6 +139,10 @@ public:
         return uses_legacy_varyings;
     }
 
+    bool UsesYNegate() const {
+        return uses_y_negate;
+    }
+
     bool UsesWarps() const {
         return uses_warps;
     }
@@ -465,6 +469,7 @@ private:
     bool uses_instance_id{};
     bool uses_vertex_id{};
     bool uses_legacy_varyings{};
+    bool uses_y_negate{};
     bool uses_warps{};
     bool uses_indexed_samplers{};
 
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index d1080300f..f336b705f 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -103,9 +103,6 @@ public:
     /// Notify the cache that a new frame has been queued
     void TickFrame();
 
-    /// Return an unique mutually exclusive lock for the cache
-    [[nodiscard]] std::unique_lock<std::mutex> AcquireLock();
-
     /// Return a constant reference to the given image view id
     [[nodiscard]] const ImageView& GetImageView(ImageViewId id) const noexcept;
 
@@ -179,6 +176,8 @@ public:
     /// Return true when a CPU region is modified from the GPU
     [[nodiscard]] bool IsRegionGpuModified(VAddr addr, size_t size);
 
+    std::mutex mutex;
+
 private:
     /// Iterate over all page indices in a range
     template <typename Func>
@@ -212,8 +211,8 @@ private:
     void RefreshContents(Image& image);
 
     /// Upload data from guest to an image
-    template <typename MapBuffer>
-    void UploadImageContents(Image& image, MapBuffer& map, size_t buffer_offset);
+    template <typename StagingBuffer>
+    void UploadImageContents(Image& image, StagingBuffer& staging_buffer, size_t buffer_offset);
 
     /// Find or create an image view from a guest descriptor
     [[nodiscard]] ImageViewId FindImageView(const TICEntry& config);
@@ -325,8 +324,6 @@ private:
 
     RenderTargets render_targets;
 
-    std::mutex mutex;
-
     std::unordered_map<TICEntry, ImageViewId> image_views;
     std::unordered_map<TSCEntry, SamplerId> samplers;
     std::unordered_map<RenderTargets, FramebufferId> framebuffers;
@@ -386,11 +383,6 @@ void TextureCache<P>::TickFrame() {
 }
 
 template <class P>
-std::unique_lock<std::mutex> TextureCache<P>::AcquireLock() {
-    return std::unique_lock{mutex};
-}
-
-template <class P>
 const typename P::ImageView& TextureCache<P>::GetImageView(ImageViewId id) const noexcept {
     return slot_image_views[id];
 }
@@ -598,11 +590,11 @@ void TextureCache<P>::DownloadMemory(VAddr cpu_addr, size_t size) {
     });
     for (const ImageId image_id : images) {
         Image& image = slot_images[image_id];
-        auto map = runtime.MapDownloadBuffer(image.unswizzled_size_bytes);
+        auto map = runtime.DownloadStagingBuffer(image.unswizzled_size_bytes);
         const auto copies = FullDownloadCopies(image.info);
         image.DownloadMemory(map, 0, copies);
         runtime.Finish();
-        SwizzleImage(gpu_memory, image.gpu_addr, image.info, copies, map.Span());
+        SwizzleImage(gpu_memory, image.gpu_addr, image.info, copies, map.mapped_span);
     }
 }
 
@@ -757,7 +749,7 @@ void TextureCache<P>::PopAsyncFlushes() {
     for (const ImageId image_id : download_ids) {
         total_size_bytes += slot_images[image_id].unswizzled_size_bytes;
     }
-    auto download_map = runtime.MapDownloadBuffer(total_size_bytes);
+    auto download_map = runtime.DownloadStagingBuffer(total_size_bytes);
     size_t buffer_offset = 0;
     for (const ImageId image_id : download_ids) {
         Image& image = slot_images[image_id];
@@ -769,7 +761,7 @@ void TextureCache<P>::PopAsyncFlushes() {
     runtime.Finish();
 
     buffer_offset = 0;
-    const std::span<u8> download_span = download_map.Span();
+    const std::span<u8> download_span = download_map.mapped_span;
     for (const ImageId image_id : download_ids) {
         const ImageBase& image = slot_images[image_id];
         const auto copies = FullDownloadCopies(image.info);
@@ -806,7 +798,7 @@ void TextureCache<P>::RefreshContents(Image& image) {
         LOG_WARNING(HW_GPU, "MSAA image uploads are not implemented");
         return;
     }
-    auto map = runtime.MapUploadBuffer(MapSizeBytes(image));
+    auto map = runtime.UploadStagingBuffer(MapSizeBytes(image));
     UploadImageContents(image, map, 0);
     runtime.InsertUploadMemoryBarrier();
 }
@@ -814,7 +806,7 @@ void TextureCache<P>::RefreshContents(Image& image) {
 template <class P>
 template <typename MapBuffer>
 void TextureCache<P>::UploadImageContents(Image& image, MapBuffer& map, size_t buffer_offset) {
-    const std::span<u8> mapped_span = map.Span().subspan(buffer_offset);
+    const std::span<u8> mapped_span = map.mapped_span.subspan(buffer_offset);
     const GPUVAddr gpu_addr = image.gpu_addr;
 
     if (True(image.flags & ImageFlagBits::AcceleratedUpload)) {
diff --git a/src/video_core/vulkan_common/vulkan_memory_allocator.h b/src/video_core/vulkan_common/vulkan_memory_allocator.h
index 86393310a..d1ce29450 100644
--- a/src/video_core/vulkan_common/vulkan_memory_allocator.h
+++ b/src/video_core/vulkan_common/vulkan_memory_allocator.h
@@ -78,7 +78,7 @@ public:
      *
      * @throw vk::Exception on failure
      */
-    explicit MemoryAllocator(const Device& device_, bool export_allocations_ = false);
+    explicit MemoryAllocator(const Device& device_, bool export_allocations_);
     ~MemoryAllocator();
 
     MemoryAllocator& operator=(const MemoryAllocator&) = delete;