4 files changed, 29 insertions, 0 deletions

diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h
index 53874e75cc..70d79c3535 100644
--- a/src/video_core/buffer_cache/buffer_cache.h
+++ b/src/video_core/buffer_cache/buffer_cache.h
@@ -100,6 +100,12 @@ void BufferCache<P>::TickFrame() {
         return;
     }
     runtime.TickFrame(slot_buffers);
+    if (frame_tick % 600 == 0 && !uploaded_buffer_ranges.empty()) {
+        for (const auto& [address, size] : uploaded_buffer_ranges) {
+            device_memory.HintCold(address, size);
+        }
+        uploaded_buffer_ranges.clear();
+    }
     if (static u64 frames{}; ++frames % 600 == 0) {
         u64 total{};
         size_t count{};
@@ -1928,6 +1934,11 @@ void BufferCache<P>::MappedUploadMemory([[maybe_unused]] Buffer& buffer,
             u8* const src_pointer = staging_pointer.data() + copy.src_offset;
             const DAddr device_addr = buffer.CpuAddr() + copy.dst_offset;
             device_memory.ReadBlockUnsafe(device_addr, src_pointer, copy.size);
+            // Large uploads are mostly geometry the host GPU now has its own copy of: TickFrame
+            // marks their guest memory cold (see TextureCache::RefreshContents).
+            if (copy.size >= 256_KiB) {
+                uploaded_buffer_ranges.emplace_back(device_addr, copy.size);
+            }
             // Apply the staging offset
             copy.src_offset += upload_staging.offset;
         }
diff --git a/src/video_core/buffer_cache/buffer_cache_base.h b/src/video_core/buffer_cache/buffer_cache_base.h
index d8a61b7913..47f342932d 100644
--- a/src/video_core/buffer_cache/buffer_cache_base.h
+++ b/src/video_core/buffer_cache/buffer_cache_base.h
@@ -580,6 +580,8 @@ private:
 
     size_t immediate_buffer_capacity = 0;
     Common::ScratchBuffer<u8> immediate_buffer_alloc;
+    /// Guest memory large uploads came from since the last cold hint (device address, size).
+    std::vector<std::pair<DAddr, u64>> uploaded_buffer_ranges;
 
     struct LRUItemParams {
         using ObjectType = BufferId;
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index 33ca051482..77485a282e 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -320,6 +320,12 @@ void TextureCache<P>::TickFrame() {
     sentenced_image_view.Tick();
     TickAsyncDecode();
     TickAsyncUnswizzle();
+    if (frame_tick % 600 == 0 && !uploaded_ranges.empty()) {
+        for (const auto& [address, size] : uploaded_ranges) {
+            device_memory.HintCold(address, size);
+        }
+        uploaded_ranges.clear();
+    }
 
     runtime.TickFrame();
     ++frame_tick;
@@ -1365,6 +1371,14 @@ void TextureCache<P>::RefreshContents(Image& image, ImageId image_id) {
     auto staging = runtime.UploadStagingBuffer(MapSizeBytes(image));
     UploadImageContents(image, staging);
     runtime.InsertUploadMemoryBarrier();
+    // The host GPU has its own copy now and the guest rarely reads this memory again: TickFrame
+    // marks it cold, so the OS pages it out before anything else when it needs RAM (title 1
+    // streams several GB of textures through it).
+    constexpr u64 ColdHintMinBytes = 64_KiB;
+    if (image.guest_size_bytes >= ColdHintMinBytes &&
+        False(image.flags & (ImageFlagBits::GpuModified | ImageFlagBits::Sparse))) {
+        uploaded_ranges.emplace_back(image.cpu_addr, image.guest_size_bytes);
+    }
 }
 
 template <class P>
diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h
index 141e5b76d5..12240287ff 100644
--- a/src/video_core/texture_cache/texture_cache_base.h
+++ b/src/video_core/texture_cache/texture_cache_base.h
@@ -578,6 +578,8 @@ private:
     Common::ThreadWorker texture_decode_worker{1, "TextureDecoder", {},
                                                Common::ThreadPlacement::Efficiency};
     std::vector<std::unique_ptr<AsyncDecodeContext>> async_decodes;
+    /// Guest memory images were uploaded from since the last cold hint (device address, size).
+    std::vector<std::pair<DAddr, u64>> uploaded_ranges;
 
     std::deque<PendingUnswizzle> unswizzle_queue;
     u8 current_unswizzle_frame;