diff --git a/desmume/src/GPU.cpp b/desmume/src/GPU.cpp
index 18a2883f2..1b9224f1c 100644
--- a/desmume/src/GPU.cpp
+++ b/desmume/src/GPU.cpp
@@ -4142,7 +4142,8 @@ GPUSubsystem::GPUSubsystem()
 	_customVRAM = NULL;
 	_customVRAMBlank = NULL;
 	_customFramebuffer = (u16 *)malloc_alignedCacheLine(GPU_FRAMEBUFFER_NATIVE_WIDTH * GPU_FRAMEBUFFER_NATIVE_HEIGHT * sizeof(u16) * 2);
-	_nativeFramebuffer = (u16 *)malloc_alignedCacheLine(GPU_FRAMEBUFFER_NATIVE_WIDTH * GPU_FRAMEBUFFER_NATIVE_HEIGHT * sizeof(u16) * 2);
+	
+	ClearWithColor(0x8000);
 	
 	_displayInfo.isCustomSizeRequested = false;
 	_displayInfo.customWidth = GPU_FRAMEBUFFER_NATIVE_WIDTH;
@@ -4162,8 +4163,6 @@ GPUSubsystem::GPUSubsystem()
 	_displayInfo.renderedHeight[1] = GPU_FRAMEBUFFER_NATIVE_HEIGHT;
 	_displayInfo.renderedBuffer[0] = _displayInfo.nativeBuffer[0];
 	_displayInfo.renderedBuffer[1] = _displayInfo.nativeBuffer[1];
-
-	ClearWithColor(0x8000);
 }
 
 GPUSubsystem::~GPUSubsystem()
@@ -4171,7 +4170,6 @@ GPUSubsystem::~GPUSubsystem()
 	delete osd;
 	osd = NULL;
 	
-	free_aligned(this->_nativeFramebuffer);
 	free_aligned(this->_customFramebuffer);
 	free_aligned(this->_customVRAM);
 	
@@ -4186,6 +4184,17 @@ GPUSubsystem::~GPUSubsystem()
 	gfx3d_deinit();
 }
 
+GPUSubsystem* GPUSubsystem::Allocate()
+{
+	return new(malloc_aligned64(sizeof(GPUSubsystem))) GPUSubsystem();
+}
+
+void GPUSubsystem::FinalizeAndDeallocate()
+{
+	this->~GPUSubsystem();
+	free_aligned(this);
+}
+
 void GPUSubsystem::Reset()
 {
 	if (this->_customVRAM == NULL || this->_customVRAM == NULL || this->_customFramebuffer == NULL)
diff --git a/desmume/src/GPU.h b/desmume/src/GPU.h
index 8f36eb99a..6acf974ef 100644
--- a/desmume/src/GPU.h
+++ b/desmume/src/GPU.h
@@ -1324,16 +1324,17 @@ private:
 	u16 *_customVRAM;
 	u16 *_customVRAMBlank;
 	
-	//zero 13-sep-2015 - had to change this to a pointer instead of an array. as an array, we need the whole GPUSubsystem aligned, and that gets annoying. 
-	//If having the array instead of the pointer is faster, we can change it back and just deal with this type being allocated specially
-	u16 *_nativeFramebuffer;
+	CACHE_ALIGN u16 _nativeFramebuffer[GPU_FRAMEBUFFER_NATIVE_WIDTH * GPU_FRAMEBUFFER_NATIVE_HEIGHT * 2];
 	u16 *_customFramebuffer;
 	
 	NDSDisplayInfo _displayInfo;
 	
-public:
 	GPUSubsystem();
 	~GPUSubsystem();
+
+public:
+	static GPUSubsystem* Allocate();
+	void FinalizeAndDeallocate();
 	
 	void Reset();
 	VRAM3DUsageProperties& GetVRAM3DUsageProperties();
diff --git a/desmume/src/NDSSystem.cpp b/desmume/src/NDSSystem.cpp
index 1501c5bb9..9822a5af9 100644
--- a/desmume/src/NDSSystem.cpp
+++ b/desmume/src/NDSSystem.cpp
@@ -170,10 +170,10 @@ int NDS_Init()
 	
 	if (GPU != NULL)
 	{
-		delete GPU;
+		GPU->FinalizeAndDeallocate();
 	}
 	
-	GPU = new GPUSubsystem;
+	GPU = GPUSubsystem::Allocate();
 	
 	if (SPU_Init(SNDCORE_DUMMY, 740) != 0)
 		return -1;
@@ -191,7 +191,7 @@ void NDS_DeInit(void)
 	gameInfo.closeROM();
 	SPU_DeInit();
 	
-	delete GPU;
+	GPU->FinalizeAndDeallocate();
 	GPU = NULL;
 	
 	MMU_DeInit();