diff --git a/src/core/gpu.cpp b/src/core/gpu.cpp
index 3eaed5f06..3ff74b33a 100644
--- a/src/core/gpu.cpp
+++ b/src/core/gpu.cpp
@@ -698,7 +698,35 @@ void GPU::ReadVRAM(u32 x, u32 y, u32 width, u32 height) {}
 
 void GPU::FillVRAM(u32 x, u32 y, u32 width, u32 height, u32 color) {}
 
-void GPU::UpdateVRAM(u32 x, u32 y, u32 width, u32 height, const void* data) {}
+void GPU::UpdateVRAM(u32 x, u32 y, u32 width, u32 height, const void* data)
+{
+  // Fast path when the copy is not oversized.
+  if ((x + width) <= VRAM_WIDTH && (y + height) <= VRAM_HEIGHT)
+  {
+    const u16* src_ptr = static_cast<const u16*>(data);
+    u16* dst_ptr = &m_vram_ptr[y * VRAM_WIDTH + x];
+    for (u32 yoffs = 0; yoffs < height; yoffs++)
+    {
+      std::copy_n(src_ptr, width, dst_ptr);
+      src_ptr += width;
+      dst_ptr += VRAM_WIDTH;
+    }
+  }
+  else
+  {
+    // Slow path when we need to handle wrap-around.
+    const u16* src_ptr = static_cast<const u16*>(data);
+    for (u32 row = 0; row < height;)
+    {
+      u16* dst_row_ptr = &m_vram_ptr[((y + row++) % VRAM_HEIGHT) * VRAM_WIDTH];
+      for (u32 col = 0; col < width;)
+      {
+        // TODO: Handle unaligned reads...
+        dst_row_ptr[(x + col++) % VRAM_WIDTH] = *(src_ptr++);
+      }
+    }
+  }
+}
 
 void GPU::CopyVRAM(u32 src_x, u32 src_y, u32 dst_x, u32 dst_y, u32 width, u32 height) {}
 
diff --git a/src/core/gpu_commands.cpp b/src/core/gpu_commands.cpp
index 83067fc08..eebcf1a8f 100644
--- a/src/core/gpu_commands.cpp
+++ b/src/core/gpu_commands.cpp
@@ -286,13 +286,6 @@ bool GPU::HandleCopyRectangleCPUToVRAMCommand(const u32*& command_ptr, u32 comma
   Log_DebugPrintf("Copy rectangle from CPU to VRAM offset=(%u,%u), size=(%u,%u)", dst_x, dst_y, copy_width,
                   copy_height);
 
-  if ((dst_x + copy_width) > VRAM_WIDTH || (dst_y + copy_height) > VRAM_HEIGHT)
-  {
-    Log_ErrorPrintf("Out of bounds CPU->VRAM copy (%u,%u) @ (%u,%u)", copy_width, copy_height, dst_x, dst_y);
-    command_ptr += num_words;
-    return true;
-  }
-
   if (m_system->GetSettings().debugging.dump_cpu_to_vram_copies)
   {
     DumpVRAMToFile(SmallString::FromFormat("cpu_to_vram_copy_%u.png", s_cpu_to_vram_dump_id++), copy_width, copy_height,
@@ -321,12 +314,6 @@ bool GPU::HandleCopyRectangleVRAMToCPUCommand(const u32*& command_ptr, u32 comma
 
   Log_DebugPrintf("Copy rectangle from VRAM to CPU offset=(%u,%u), size=(%u,%u)", src_x, src_y, width, height);
 
-  if ((src_x + width) > VRAM_WIDTH || (src_y + height) > VRAM_HEIGHT)
-  {
-    Panic("Out of bounds VRAM copy");
-    return true;
-  }
-
   // all rendering should be done first...
   FlushRender();
 
diff --git a/src/core/gpu_hw_d3d11.cpp b/src/core/gpu_hw_d3d11.cpp
index e304addaf..4b48bed07 100644
--- a/src/core/gpu_hw_d3d11.cpp
+++ b/src/core/gpu_hw_d3d11.cpp
@@ -638,6 +638,16 @@ void GPU_HW_D3D11::UpdateVRAM(u32 x, u32 y, u32 width, u32 height, const void* d
 {
   GPU_HW::UpdateVRAM(x, y, width, height, data);
 
+  if ((x + width) > VRAM_WIDTH || (y + height) > VRAM_HEIGHT)
+  {
+    // CPU round trip if oversized for now.
+    Log_WarningPrintf("Oversized VRAM update (%u-%u, %u-%u), CPU round trip", x, x + width, y, y + height);
+    GPU::UpdateVRAM(x, y, width, height, data);
+    ReadVRAM(0, 0, VRAM_WIDTH, VRAM_HEIGHT);
+    UpdateVRAM(0, 0, VRAM_WIDTH, VRAM_HEIGHT, m_vram_shadow.data());
+    return;
+  }
+
   const u32 num_pixels = width * height;
   const auto map_result = m_texture_stream_buffer.Map(m_context.Get(), sizeof(u16), num_pixels * sizeof(u16));
   std::memcpy(map_result.pointer, data, num_pixels * sizeof(u16));
diff --git a/src/core/gpu_hw_opengl.cpp b/src/core/gpu_hw_opengl.cpp
index 1f442c7ab..e95600124 100644
--- a/src/core/gpu_hw_opengl.cpp
+++ b/src/core/gpu_hw_opengl.cpp
@@ -583,6 +583,16 @@ void GPU_HW_OpenGL::UpdateVRAM(u32 x, u32 y, u32 width, u32 height, const void*
 {
   GPU_HW::UpdateVRAM(x, y, width, height, data);
 
+  if ((x + width) > VRAM_WIDTH || (y + height) > VRAM_HEIGHT)
+  {
+    // CPU round trip if oversized for now.
+    Log_WarningPrintf("Oversized VRAM update (%u-%u, %u-%u), CPU round trip", x, x + width, y, y + height);
+    GPU::UpdateVRAM(x, y, width, height, data);
+    ReadVRAM(0, 0, VRAM_WIDTH, VRAM_HEIGHT);
+    UpdateVRAM(0, 0, VRAM_WIDTH, VRAM_HEIGHT, m_vram_shadow.data());
+    return;
+  }
+
   const u32 num_pixels = width * height;
   if (num_pixels < m_max_texture_buffer_size)
   {
diff --git a/src/core/gpu_sw.cpp b/src/core/gpu_sw.cpp
index d2458d038..825b11dc5 100644
--- a/src/core/gpu_sw.cpp
+++ b/src/core/gpu_sw.cpp
@@ -50,17 +50,6 @@ void GPU_SW::FillVRAM(u32 x, u32 y, u32 width, u32 height, u32 color)
     std::fill_n(GetPixelPtr(x, y + yoffs), width, color16);
 }
 
-void GPU_SW::UpdateVRAM(u32 x, u32 y, u32 width, u32 height, const void* data)
-{
-  const u16* src_ptr = static_cast<const u16*>(data);
-  for (u32 yoffs = 0; yoffs < height; yoffs++)
-  {
-    u16* dst_ptr = GetPixelPtr(x, y + yoffs);
-    std::copy_n(src_ptr, width, dst_ptr);
-    src_ptr += width;
-  }
-}
-
 void GPU_SW::CopyVRAM(u32 src_x, u32 src_y, u32 dst_x, u32 dst_y, u32 width, u32 height)
 {
   for (u32 yoffs = 0; yoffs < height; yoffs++)
diff --git a/src/core/gpu_sw.h b/src/core/gpu_sw.h
index 8a93495c8..da684eb56 100644
--- a/src/core/gpu_sw.h
+++ b/src/core/gpu_sw.h
@@ -36,7 +36,6 @@ protected:
 
   void ReadVRAM(u32 x, u32 y, u32 width, u32 height) override;
   void FillVRAM(u32 x, u32 y, u32 width, u32 height, u32 color) override;
-  void UpdateVRAM(u32 x, u32 y, u32 width, u32 height, const void* data) override;
   void CopyVRAM(u32 src_x, u32 src_y, u32 dst_x, u32 dst_y, u32 width, u32 height) override;
 
   //////////////////////////////////////////////////////////////////////////