diff --git a/GPU/Directx9/GPU_DX9.cpp b/GPU/Directx9/GPU_DX9.cpp
index ac7a596874..2967686809 100644
--- a/GPU/Directx9/GPU_DX9.cpp
+++ b/GPU/Directx9/GPU_DX9.cpp
@@ -1330,6 +1330,11 @@ bool DIRECTX9_GPU::PerformMemoryDownload(u32 dest, int size) {
 	return false;
 }
 
+bool DIRECTX9_GPU::PerformMemoryUpload(u32 dest, int size) {
+	InvalidateCache(dest, size, GPU_INVALIDATE_HINT);
+	return false;
+}
+
 bool DIRECTX9_GPU::PerformStencilUpload(u32 dest, int size) {
 	return false;
 }
diff --git a/GPU/Directx9/GPU_DX9.h b/GPU/Directx9/GPU_DX9.h
index 35ff0dc4ba..33bd0bd19b 100644
--- a/GPU/Directx9/GPU_DX9.h
+++ b/GPU/Directx9/GPU_DX9.h
@@ -49,6 +49,7 @@ public:
 	virtual bool PerformMemoryCopy(u32 dest, u32 src, int size);
 	virtual bool PerformMemorySet(u32 dest, u8 v, int size);
 	virtual bool PerformMemoryDownload(u32 dest, int size);
+	virtual bool PerformMemoryUpload(u32 dest, int size);
 	virtual bool PerformStencilUpload(u32 dest, int size);
 	virtual void ClearCacheNextFrame();
 	virtual void DeviceLost();  // Only happens on Android. Drop all textures and shaders.
diff --git a/GPU/GLES/GLES_GPU.cpp b/GPU/GLES/GLES_GPU.cpp
index 0b9b0b57bb..c251dba9b5 100644
--- a/GPU/GLES/GLES_GPU.cpp
+++ b/GPU/GLES/GLES_GPU.cpp
@@ -1997,7 +1997,11 @@ void GLES_GPU::InvalidateCacheInternal(u32 addr, int size, GPUInvalidationType t
 
 void GLES_GPU::PerformMemoryCopyInternal(u32 dest, u32 src, int size) {
 	if (!framebufferManager_.NotifyFramebufferCopy(src, dest, size)) {
-		Memory::Memcpy(dest, Memory::GetPointer(src), size);
+		// We use a little hack for Download/Upload using a VRAM mirror.
+		// Since they're identical we don't need to copy.
+		if (!Memory::IsVRAMAddress(dest) || (dest ^ 0x00400000) != src) {
+			Memory::Memcpy(dest, Memory::GetPointer(src), size);
+		}
 	}
 	InvalidateCache(dest, size, GPU_INVALIDATE_HINT);
 }
@@ -2064,6 +2068,12 @@ bool GLES_GPU::PerformMemoryDownload(u32 dest, int size) {
 	return gpu->PerformMemoryCopy(dest ^ 0x00400000, dest, size);
 }
 
+bool GLES_GPU::PerformMemoryUpload(u32 dest, int size) {
+	// Cheat a bit to force a download of the framebuffer.
+	// VRAM + 0x00400000 is simply a VRAM mirror.
+	return gpu->PerformMemoryCopy(dest, dest ^ 0x00400000, size);
+}
+
 bool GLES_GPU::PerformStencilUpload(u32 dest, int size) {
 	if (framebufferManager_.MayIntersectFramebuffer(dest)) {
 		if (IsOnSeparateCPUThread()) {
diff --git a/GPU/GLES/GLES_GPU.h b/GPU/GLES/GLES_GPU.h
index 85bcb032c1..3d3f7e07d9 100644
--- a/GPU/GLES/GLES_GPU.h
+++ b/GPU/GLES/GLES_GPU.h
@@ -48,6 +48,7 @@ public:
 	virtual bool PerformMemoryCopy(u32 dest, u32 src, int size);
 	virtual bool PerformMemorySet(u32 dest, u8 v, int size);
 	virtual bool PerformMemoryDownload(u32 dest, int size);
+	virtual bool PerformMemoryUpload(u32 dest, int size);
 	virtual bool PerformStencilUpload(u32 dest, int size);
 	virtual void ClearCacheNextFrame();
 	virtual void DeviceLost();  // Only happens on Android. Drop all textures and shaders.
diff --git a/GPU/GPUInterface.h b/GPU/GPUInterface.h
index e8f6844514..b6b418ca4a 100644
--- a/GPU/GPUInterface.h
+++ b/GPU/GPUInterface.h
@@ -250,6 +250,7 @@ public:
 	virtual bool PerformMemoryCopy(u32 dest, u32 src, int size) = 0;
 	virtual bool PerformMemorySet(u32 dest, u8 v, int size) = 0;
 	virtual bool PerformMemoryDownload(u32 dest, int size) = 0;
+	virtual bool PerformMemoryUpload(u32 dest, int size) = 0;
 	virtual bool PerformStencilUpload(u32 dest, int size) = 0;
 
 	// Will cause the texture cache to be cleared at the start of the next frame.
diff --git a/GPU/Null/NullGpu.cpp b/GPU/Null/NullGpu.cpp
index f427b909d0..cfd358f929 100644
--- a/GPU/Null/NullGpu.cpp
+++ b/GPU/Null/NullGpu.cpp
@@ -675,6 +675,12 @@ bool NullGPU::PerformMemoryDownload(u32 dest, int size) {
 	return false;
 }
 
+bool NullGPU::PerformMemoryUpload(u32 dest, int size) {
+	// Nothing to update.
+	InvalidateCache(dest, size, GPU_INVALIDATE_HINT);
+	return false;
+}
+
 bool NullGPU::PerformStencilUpload(u32 dest, int size) {
 	return false;
 }
diff --git a/GPU/Null/NullGpu.h b/GPU/Null/NullGpu.h
index a83b256e1e..aac1fb7124 100644
--- a/GPU/Null/NullGpu.h
+++ b/GPU/Null/NullGpu.h
@@ -37,6 +37,7 @@ public:
 	virtual bool PerformMemoryCopy(u32 dest, u32 src, int size);
 	virtual bool PerformMemorySet(u32 dest, u8 v, int size);
 	virtual bool PerformMemoryDownload(u32 dest, int size);
+	virtual bool PerformMemoryUpload(u32 dest, int size);
 	virtual bool PerformStencilUpload(u32 dest, int size);
 	virtual void ClearCacheNextFrame() {};
 
diff --git a/GPU/Software/SoftGpu.cpp b/GPU/Software/SoftGpu.cpp
index 2d3e38c522..e54e6082bb 100644
--- a/GPU/Software/SoftGpu.cpp
+++ b/GPU/Software/SoftGpu.cpp
@@ -876,6 +876,13 @@ bool SoftGPU::PerformMemoryDownload(u32 dest, int size)
 	return false;
 }
 
+bool SoftGPU::PerformMemoryUpload(u32 dest, int size)
+{
+	// Nothing to update.
+	InvalidateCache(dest, size, GPU_INVALIDATE_HINT);
+	return false;
+}
+
 bool SoftGPU::PerformStencilUpload(u32 dest, int size)
 {
 	return false;
diff --git a/GPU/Software/SoftGpu.h b/GPU/Software/SoftGpu.h
index 9ffe4509bd..4d028e4da8 100644
--- a/GPU/Software/SoftGpu.h
+++ b/GPU/Software/SoftGpu.h
@@ -62,6 +62,7 @@ public:
 	virtual bool PerformMemoryCopy(u32 dest, u32 src, int size);
 	virtual bool PerformMemorySet(u32 dest, u8 v, int size);
 	virtual bool PerformMemoryDownload(u32 dest, int size);
+	virtual bool PerformMemoryUpload(u32 dest, int size);
 	virtual bool PerformStencilUpload(u32 dest, int size);
 	virtual void ClearCacheNextFrame() {};