diff --git a/src/Cafe/HW/Latte/Renderer/Metal/MetalBufferAllocator.h b/src/Cafe/HW/Latte/Renderer/Metal/MetalBufferAllocator.h
index 445fb823..702278ca 100644
--- a/src/Cafe/HW/Latte/Renderer/Metal/MetalBufferAllocator.h
+++ b/src/Cafe/HW/Latte/Renderer/Metal/MetalBufferAllocator.h
@@ -15,7 +15,13 @@ template<typename BufferT>
 class MetalBufferAllocator
 {
 public:
-    MetalBufferAllocator(class MetalRenderer* metalRenderer, MTL::ResourceOptions storageMode) : m_mtlr{metalRenderer}, m_storageMode{storageMode} {}
+    MetalBufferAllocator(class MetalRenderer* metalRenderer, MTL::ResourceOptions storageMode) : m_mtlr{metalRenderer} {
+        m_isCPUAccessible = (storageMode == MTL::ResourceStorageModeShared) || (storageMode == MTL::ResourceStorageModeManaged);
+
+        m_options = storageMode;
+        if (m_isCPUAccessible)
+            m_options |= MTL::ResourceCPUCacheModeWriteCombined;
+    }
 
     ~MetalBufferAllocator()
     {
@@ -54,7 +60,7 @@ public:
                 allocation.bufferIndex = range.bufferIndex;
                 allocation.offset = range.offset;
                 allocation.size = size;
-                allocation.data = (uint8*)buffer.m_buffer->contents() + range.offset;
+                allocation.data = (m_isCPUAccessible ? (uint8*)buffer.m_buffer->contents() + range.offset : nullptr);
 
                 range.offset += size;
                 range.size -= size;
@@ -70,7 +76,7 @@ public:
 
         // If no free range was found, allocate a new buffer
         m_allocationSize = std::max(m_allocationSize, size);
-        MTL::Buffer* buffer = m_mtlr->GetDevice()->newBuffer(m_allocationSize, m_storageMode | MTL::ResourceCPUCacheModeWriteCombined);
+        MTL::Buffer* buffer = m_mtlr->GetDevice()->newBuffer(m_allocationSize, m_options);
     #ifdef CEMU_DEBUG_ASSERT
         buffer->setLabel(GetLabel("Buffer from buffer allocator", buffer));
     #endif
@@ -79,7 +85,7 @@ public:
         allocation.bufferIndex = m_buffers.size();
         allocation.offset = 0;
         allocation.size = size;
-        allocation.data = buffer->contents();
+        allocation.data = (m_isCPUAccessible ? buffer->contents() : nullptr);
 
         m_buffers.push_back({buffer});
 
@@ -129,7 +135,8 @@ public:
 
 protected:
     class MetalRenderer* m_mtlr;
-    MTL::ResourceOptions m_storageMode;
+    bool m_isCPUAccessible;
+    MTL::ResourceOptions m_options;
 
     size_t m_allocationSize = 8 * 1024 * 1024;
 
diff --git a/src/Cafe/HW/Latte/Renderer/Metal/MetalRenderer.cpp b/src/Cafe/HW/Latte/Renderer/Metal/MetalRenderer.cpp
index 008cac99..44090a05 100644
--- a/src/Cafe/HW/Latte/Renderer/Metal/MetalRenderer.cpp
+++ b/src/Cafe/HW/Latte/Renderer/Metal/MetalRenderer.cpp
@@ -41,7 +41,7 @@ MetalRenderer::MetalRenderer()
     m_commandQueue = m_device->newCommandQueue();
 
     // Feature support
-    m_isAppleGPU = m_device->supportsFamily(MTL::GPUFamilyApple1);
+    m_isAppleGPU = false;//m_device->supportsFamily(MTL::GPUFamilyApple1);
     m_hasUnifiedMemory = m_device->hasUnifiedMemory();
     m_supportsMetal3 = m_device->supportsFamily(MTL::GPUFamilyMetal3);
     m_recommendedMaxVRAMUsage = m_device->recommendedMaxWorkingSetSize();
@@ -511,7 +511,7 @@ void MetalRenderer::texture_loadSlice(LatteTexture* hostTexture, sint32 width, s
     size_t bytesPerRow = GetMtlTextureBytesPerRow(textureMtl->GetFormat(), textureMtl->IsDepth(), width);
     // No need to set bytesPerImage for 3D textures, since we always load just one slice
     //size_t bytesPerImage = GetMtlTextureBytesPerImage(textureMtl->GetFormat(), textureMtl->IsDepth(), height, bytesPerRow);
-    if (IsAppleGPU())
+    if (m_isAppleGPU)
     {
         textureMtl->GetTexture()->replaceRegion(MTL::Region(0, 0, offsetZ, width, height, 1), mipIndex, sliceIndex, pixelData, bytesPerRow, 0);
     }
@@ -674,7 +674,7 @@ LatteTextureReadbackInfo* MetalRenderer::texture_createReadback(LatteTextureView
 
 void MetalRenderer::surfaceCopy_copySurfaceWithFormatConversion(LatteTexture* sourceTexture, sint32 srcMip, sint32 srcSlice, LatteTexture* destinationTexture, sint32 dstMip, sint32 dstSlice, sint32 width, sint32 height)
 {
-    GetCommandBuffer();
+    return;
 
     // scale copy size to effective size
 	sint32 effectiveCopyWidth = width;
@@ -907,10 +907,12 @@ void MetalRenderer::draw_execute(uint32 baseVertex, uint32 baseInstance, uint32
     bool usesGeometryShader = (geometryShader != nullptr || isPrimitiveRect);
 
 	// Depth stencil state
-	// TODO
+
 	// Disable depth write when there is no depth attachment
-	//if (!m_state.m_lastUsedFBO->depthBuffer.texture)
-	//    depthControl.set_Z_WRITE_ENABLE(false);
+	auto& depthControl = LatteGPUState.contextNew.DB_DEPTH_CONTROL;
+	bool depthWriteEnable = depthControl.get_Z_WRITE_ENABLE();
+	if (!m_state.m_lastUsedFBO->depthBuffer.texture)
+	    depthControl.set_Z_WRITE_ENABLE(false);
 
 	MTL::DepthStencilState* depthStencilState = m_depthStencilCache->GetDepthStencilState(LatteGPUState.contextNew);
 	if (depthStencilState != encoderState.m_depthStencilState)
@@ -919,6 +921,9 @@ void MetalRenderer::draw_execute(uint32 baseVertex, uint32 baseInstance, uint32
 		encoderState.m_depthStencilState = depthStencilState;
 	}
 
+	// Restore the original depth write state
+	depthControl.set_Z_WRITE_ENABLE(depthWriteEnable);
+
 	// Stencil reference
 	bool stencilEnable = LatteGPUState.contextNew.DB_DEPTH_CONTROL.get_STENCIL_ENABLE();
 	if (stencilEnable)