diff --git a/desmume/src/FIFO.cpp b/desmume/src/FIFO.cpp
index 4ae6db02b..7bff14f99 100644
--- a/desmume/src/FIFO.cpp
+++ b/desmume/src/FIFO.cpp
@@ -170,7 +170,11 @@ void GFX_FIFOclear()
 
 void GFX_FIFOsend(u8 cmd, u32 param)
 {
-	u32 gxstat = (T1ReadLong(MMU.MMU_MEM[ARMCPU_ARM9][0x40], 0x600) & 0xF000FFFF);
+	u32 gxstat = T1ReadLong(MMU.MMU_MEM[ARMCPU_ARM9][0x40], 0x600);
+
+	//INFO("gxFIFO: send 0x%02X = 0x%08X (size %03i/0x%02X) gxstat 0x%08X\n", cmd, param, gxFIFO.size, gxFIFO.size, gxstat);
+
+	gxstat &= 0xF000FFFF;
 
 	if (gxFIFO.size == 0)				// FIFO empty
 	{
@@ -192,11 +196,11 @@ void GFX_FIFOsend(u8 cmd, u32 param)
 		}
 	}
 
-	//INFO("GFX FIFO: Send GFX 3D cmd 0x%02X to FIFO - 0x%08X (%03i/%02X)\n", cmd, param, gxFIFO.tail, gxFIFO.tail);
 	if (gxFIFO.size > 255)
 	{
 #ifdef USE_GEOMETRY_FIFO_EMULATION
-		gxstat |= 0x01000000;
+		gxstat |= 0x01000000;			// full
+		gxstat |= 0x08000000;			// busy
 #else
 		gxstat |= 0x02000000;				// this is hack (must be removed later)
 #endif
@@ -232,7 +236,7 @@ void GFX_FIFOsend(u8 cmd, u32 param)
 	NDS_RescheduleGXFIFO();
 }
 
-// this function used ONLY in new gxFIFO
+// this function used ONLY in gxFIFO
 BOOL GFX_PIPErecv(u8 *cmd, u32 *param)
 {
 	u32 gxstat = T1ReadLong(MMU.MMU_MEM[ARMCPU_ARM9][0x40], 0x600);
@@ -240,7 +244,6 @@ BOOL GFX_PIPErecv(u8 *cmd, u32 *param)
 
 	if (gxPIPE.size > 0)
 	{
-
 		*cmd = gxPIPE.cmd[gxPIPE.head];
 		*param = gxPIPE.param[gxPIPE.head];
 		gxPIPE.head++;
@@ -282,17 +285,25 @@ BOOL GFX_PIPErecv(u8 *cmd, u32 *param)
 				{
 					gxstat |= 0x02000000;
 					if (gxstat & 0x40000000)	// IRQ: less half
+					{
 						setIF(0, (1<<21));
-					execHardware_doAllDma(EDMAMode_GXFifo);
+					}
+						execHardware_doAllDma(EDMAMode_GXFifo);
 				}
 
 				if (gxFIFO.size == 0)		// empty
+				{
 					gxstat |= 0x04000000;
+					if (gxstat & 0x80000000)	// IRQ: empty
+						setIF(0, (1<<21));
+				}
 			}
 			else		// FIFO empty
 			{
 				gxstat &= 0xF000FFFF;
 				gxstat |= 0x06000000;
+				if (gxstat & 0x80000000)	// IRQ: empty
+						setIF(0, (1<<21));
 			}
 		}
 
@@ -308,19 +319,32 @@ BOOL GFX_PIPErecv(u8 *cmd, u32 *param)
 	return FALSE;
 }
 
+extern void gfx3d_ClearStack();
 void GFX_FIFOcnt(u32 val)
 {
 	u32 gxstat = T1ReadLong(MMU.MMU_MEM[ARMCPU_ARM9][0x40], 0x600);
-	//INFO("GFX FIFO: write context 0x%08X (prev 0x%08X) tail %i\n", val, gxstat, gxFIFO.tail);
-	if (val & (1<<29))		// clear? (homebrew)
+	//INFO("gxFIFO: write cnt 0x%08X (prev 0x%08X) FIFO size %03i PIPE size %03i\n", val, gxstat, gxFIFO.size, gxPIPE.size);
+
+	if (val & (1<<29))		// clear? (only in homebrew?)
 	{
 		GFX_PIPEclear();
 		GFX_FIFOclear();
 		return;
 	}
-	T1WriteLong(MMU.MMU_MEM[ARMCPU_ARM9][0x40], 0x600, gxstat);
 
-	NDS_RescheduleGXFIFO();
+	if (val & (1<<15))		// projection stack pointer reset
+	{
+		gfx3d_ClearStack();
+		val &= 0xFFFF5FFF;		// clear reset (bit15) & stack level (bit13)
+	}
+
+	T1WriteLong(MMU.MMU_MEM[ARMCPU_ARM9][0x40], 0x600, val);
+
+	if (gxFIFO.size == 0)		// empty
+	{
+		if (val & 0x80000000)	// IRQ: empty
+			setIF(0, (1<<21));
+	}
 }
 
 // ========================================================= DISP FIFO
diff --git a/desmume/src/MMU.cpp b/desmume/src/MMU.cpp
index 2a711a32b..5ce69b56a 100644
--- a/desmume/src/MMU.cpp
+++ b/desmume/src/MMU.cpp
@@ -2480,9 +2480,9 @@ void FASTCALL _MMU_ARM9_write32(u32 adr, u32 val)
 				return;
 
 			case 0x04000600:
-				((u32 *)(MMU.MMU_MEM[ARMCPU_ARM9][0x40]))[0x600>>2] = val;
 				GFX_FIFOcnt(val);
 				return;
+
 			// Alpha test reference value - Parameters:1
 			case 0x04000340:
 			{
diff --git a/desmume/src/gfx3d.cpp b/desmume/src/gfx3d.cpp
index 89c311339..8f1f9a0cc 100644
--- a/desmume/src/gfx3d.cpp
+++ b/desmume/src/gfx3d.cpp
@@ -1335,6 +1335,12 @@ s32 gfx3d_GetDirectionalMatrix (unsigned int index)
 	return (s32)(mtxCurrent[2][_index]*(1<<12));
 }
 
+void gfx3d_ClearStack()
+{
+	MatrixStackSetStackPosition(&mtxStack[0], -5);
+	MatrixStackSetStackPosition(&mtxStack[3], -5);
+}
+
 void gfx3d_glAlphaFunc(u32 v)
 {
 	gfx3d.alphaTestRef = v&31;
@@ -1669,12 +1675,11 @@ void gfx3d_VBlankEndSignal(bool skipFrame)
 
 	if (!drawPending) return;
 	drawPending = FALSE;
-	if(skipFrame) 
-	{
-		GFX_DELAY(392);
-		NDS_RescheduleGXFIFO();
-		return;
-	}
+
+	GFX_DELAY(392);
+	NDS_RescheduleGXFIFO();
+
+	if(skipFrame) return;
 	//if the null 3d core is chosen, then we need to clear out the 3d buffers to keep old data from being rendered
 	if(gpu3D == &gpu3DNull || !CommonSettings.showGpu.main)
 	{
@@ -1683,8 +1688,6 @@ void gfx3d_VBlankEndSignal(bool skipFrame)
 	}
 
 	gpu3D->NDS_3D_Render();
-	GFX_DELAY(392);
-	NDS_RescheduleGXFIFO();
 #else
 	//if we are skipping 3d frames then the 3d rendering will get held up here.
 	//but, as soon as we quit skipping frames, the held-up 3d frame will render
@@ -1707,6 +1710,27 @@ void gfx3d_VBlankEndSignal(bool skipFrame)
 #ifdef USE_GEOMETRY_FIFO_EMULATION
 //#define _3D_LOG
 
+// http://nocash.emubase.de/gbatek.htm#ds3dvideo
+// DS 3D Geometry Commands
+// Sending Commands by Ports 4000440h..40005FFh
+//
+// If the FIFO is full, then a wait is generated until data is removed from the FIFO, 
+// ie. the STR opcode gets freezed, during the wait, the bus cannot be used even by DMA,
+// interrupts, or by the NDS7 CPU.
+
+// this is a hack
+#if 1
+#define CHECKFULL() if (gxFIFO.size > 255) \
+					{\
+						gfx3d_execute3D();\
+						gfx3d_execute3D();\
+						gfx3d_execute3D();\
+						gfx3d_execute3D();\
+					}
+#else
+#define CHECKFULL() ;
+#endif
+
 static void NOPARAMS()
 {
 	for (;;)
@@ -1723,13 +1747,7 @@ static void NOPARAMS()
 			case 0x15:
 			case 0x41:
 				{
-					if (gxFIFO.size > 255) 
-					{
-						gfx3d_execute3D();
-						gfx3d_execute3D();
-						gfx3d_execute3D();
-						gfx3d_execute3D();
-					}
+					CHECKFULL();
 					GFX_FIFOsend(clCmd & 0xFF, 0);
 					clCmd >>= 8;
 					continue;
@@ -1751,13 +1769,8 @@ void gfx3d_sendCommandToFIFO(u32 val)
 #ifdef _3D_LOG
 	INFO("gxFIFO: send 0x%02X: val=0x%08X, pipe %02i, fifo %03i\n", clCmd & 0xFF, val, gxPIPE.tail, gxFIFO.tail);
 #endif
-	if (gxFIFO.size > 255) 
-	{
-		gfx3d_execute3D();
-		gfx3d_execute3D();
-		gfx3d_execute3D();
-		gfx3d_execute3D();
-	}
+
+	CHECKFULL();
 
 	switch (clCmd & 0xFF)
 	{
@@ -1863,13 +1876,8 @@ void gfx3d_sendCommand(u32 cmd, u32 param)
 #ifdef _3D_LOG
 	INFO("gxFIFO: send 0x%02X: val=0x%08X, pipe %02i, fifo %03i (direct)\n", cmd, param, gxPIPE.tail, gxFIFO.tail);
 #endif
-	if (gxFIFO.size > 255) 
-	{
-		gfx3d_execute3D();
-		gfx3d_execute3D();
-		gfx3d_execute3D();
-		gfx3d_execute3D();
-	}
+
+	CHECKFULL();
 
 	switch (cmd)
 	{
diff --git a/desmume/src/wifi.cpp b/desmume/src/wifi.cpp
index be838f524..4402bcdc5 100644
--- a/desmume/src/wifi.cpp
+++ b/desmume/src/wifi.cpp
@@ -225,7 +225,7 @@ WifiComInterface* wifiCom;
 // 3: medium logging, for debugging, shows lots of stuff
 // 4: high logging, for debugging, shows almost everything, may slow down
 // 5: highest logging, for debugging, shows everything, may slow down a lot
-#define WIFI_LOGGING_LEVEL 3
+#define WIFI_LOGGING_LEVEL 0
 
 #define WIFI_LOG_USE_LOGC 0