diff --git a/Core/HLE/HLETables.cpp b/Core/HLE/HLETables.cpp
index 5fe9411f67..c0877ea20e 100644
--- a/Core/HLE/HLETables.cpp
+++ b/Core/HLE/HLETables.cpp
@@ -97,7 +97,7 @@ const HLEFunction FakeSysCalls[] = {
 	{NID_EXTENDRETURN, __KernelReturnFromExtendStack, "__KernelReturnFromExtendStack", 'x', ""},
 	{NID_MODULERETURN, __KernelReturnFromModuleFunc, "__KernelReturnFromModuleFunc", 'x', ""},
 	{NID_IDLE, __KernelIdle, "_sceKernelIdle", 'x', ""},
-	{NID_GPUREPLAY, __KernelGPUReplay, "__KernelGPUReplay", 'x', ""},
+	{NID_GPUREPLAY, &WrapI_V<__KernelGPUReplay>, "__KernelGPUReplay", 'x', ""},
 	{NID_HLECALLRETURN, HLEReturnFromMipsCall, "HLEReturnFromMipsCall", 'x', ""},
 };
 
diff --git a/Core/HLE/sceKernelModule.cpp b/Core/HLE/sceKernelModule.cpp
index 40fb6a26df..dd510c79fb 100644
--- a/Core/HLE/sceKernelModule.cpp
+++ b/Core/HLE/sceKernelModule.cpp
@@ -1984,21 +1984,26 @@ bool __KernelLoadExec(const char *filename, u32 paramPtr, std::string *error_str
 bool __KernelLoadGEDump(const std::string &base_filename, std::string *error_string) {
 	__KernelLoadReset();
 
-	mipsr4k.pc = PSP_GetUserMemoryBase();
+	constexpr u32 codeStart = PSP_GetUserMemoryBase();
+	mipsr4k.pc = codeStart;
 
 	const static u32_le runDumpCode[] = {
 		// Save the filename.
 		MIPS_MAKE_ORI(MIPS_REG_S0, MIPS_REG_A0, 0),
 		MIPS_MAKE_ORI(MIPS_REG_S1, MIPS_REG_A1, 0),
-		// Call the actual render.
+		// Call the actual render. Jump here to start over.
 		MIPS_MAKE_SYSCALL("FakeSysCalls", "__KernelGPUReplay"),
+		MIPS_MAKE_NOP(),
+		// Re-run immediately if requested by the return value from __KernelGPUReplay
+		MIPS_MAKE_BNEZ(codeStart + 4 * 4, codeStart + 8, MIPS_REG_V0),
+		MIPS_MAKE_NOP(),
 		// Make sure we don't get out of sync.
 		MIPS_MAKE_LUI(MIPS_REG_A0, 0),
 		MIPS_MAKE_SYSCALL("sceGe_user", "sceGeDrawSync"),
 		// Wait for the next vblank to render again.
-		MIPS_MAKE_J(mipsr4k.pc + 8),
+		MIPS_MAKE_J(codeStart + 8),
 		MIPS_MAKE_SYSCALL("sceDisplay", "sceDisplayWaitVblankStart"),
-		// This never gets reached, just here to be safe.
+		// This never gets reached, just here to be "safe".
 		MIPS_MAKE_BREAK(0),
 	};
 
@@ -2021,13 +2026,13 @@ bool __KernelLoadGEDump(const std::string &base_filename, std::string *error_str
 	return true;
 }
 
-void __KernelGPUReplay() {
+int __KernelGPUReplay() {
 	// Special ABI: s0 and s1 are the "args".  Not null terminated.
 	const char *filenamep = Memory::GetCharPointer(currentMIPS->r[MIPS_REG_S1]);
 	if (!filenamep) {
 		ERROR_LOG(Log::G3D, "Failed to load dump filename");
 		Core_Stop();
-		return;
+		return 0;
 	}
 
 	std::string filename(filenamep, currentMIPS->r[MIPS_REG_S0]);
@@ -2042,6 +2047,7 @@ void __KernelGPUReplay() {
 		System_SendDebugScreenshot(std::string((const char *)&topaddr[0], linesize * 272), 272);
 		Core_Stop();
 	}
+	return 0;
 }
 
 int sceKernelLoadExec(const char *filename, u32 paramPtr)
diff --git a/Core/HLE/sceKernelModule.h b/Core/HLE/sceKernelModule.h
index a227153477..4e76a3f008 100644
--- a/Core/HLE/sceKernelModule.h
+++ b/Core/HLE/sceKernelModule.h
@@ -43,7 +43,7 @@ u32 __KernelGetModuleGP(SceUID module);
 bool KernelModuleIsKernelMode(SceUID module);
 bool __KernelLoadGEDump(const std::string &base_filename, std::string *error_string);
 bool __KernelLoadExec(const char *filename, u32 paramPtr, std::string *error_string);
-void __KernelGPUReplay();
+int __KernelGPUReplay();
 void __KernelReturnFromModuleFunc();
 SceUID KernelLoadModule(const std::string &filename, std::string *error_string);
 int KernelStartModule(SceUID moduleId, u32 argsize, u32 argAddr, u32 returnValueAddr, SceKernelSMOption *smoption, bool *needsWait);
diff --git a/Core/MIPS/MIPSCodeUtils.h b/Core/MIPS/MIPSCodeUtils.h
index 17789f72c7..9665f49886 100644
--- a/Core/MIPS/MIPSCodeUtils.h
+++ b/Core/MIPS/MIPSCodeUtils.h
@@ -29,6 +29,7 @@
 #define MIPS_MAKE_JAL(addr) (0x0C000000 | ((addr)>>2))
 #define MIPS_MAKE_JR_RA()   (0x03e00008)
 #define MIPS_MAKE_NOP()     (0)
+#define MIPS_MAKE_BNEZ(pc, addr, rs) (0x14000000 | (rs << 21) | (((int)(addr - (pc + 4)) >> 2) & 0xFFFF))
 
 #define MIPS_MAKE_ADDIU(dreg, sreg, immval) ((9 << 26) | ((dreg) << 16) | ((sreg) << 21) | (immval))
 #define MIPS_MAKE_LUI(reg, immval) (0x3c000000 | ((reg) << 16) | (immval))