diff --git a/pcsx2/MTVU.cpp b/pcsx2/MTVU.cpp
index 72a70cd1fc..bebd5f3a96 100644
--- a/pcsx2/MTVU.cpp
+++ b/pcsx2/MTVU.cpp
@@ -113,6 +113,7 @@ void VU_Thread::ExecuteRingBuffer()
 					vifRegs.itop = Read();
 
 					if (addr != -1) vuRegs.VI[REG_TPC].UL = addr;
+					vuCPU->SetStartPC(vuRegs.VI[REG_TPC].UL << 3);
 					vuCPU->Execute(vu1RunCycles);
 					gifUnit.gifPath[GIF_PATH_1].FinishGSPacketMTVU();
 					semaXGkick.Post(); // Tell MTGS a path1 packet is complete
diff --git a/pcsx2/SaveState.h b/pcsx2/SaveState.h
index d817ff642d..8d993d0ac6 100644
--- a/pcsx2/SaveState.h
+++ b/pcsx2/SaveState.h
@@ -24,7 +24,7 @@
 //  the lower 16 bit value.  IF the change is breaking of all compatibility with old
 //  states, increment the upper 16 bit value, and clear the lower 16 bits to 0.
 
-static const u32 g_SaveVersion = (0x9A11 << 16) | 0x0000;
+static const u32 g_SaveVersion = (0x9A12 << 16) | 0x0000;
 
 // this function is meant to be used in the place of GSfreeze, and provides a safe layer
 // between the GS saving function and the MTGS's needs. :)
diff --git a/pcsx2/VU.h b/pcsx2/VU.h
index 117089e75e..dd67005d36 100644
--- a/pcsx2/VU.h
+++ b/pcsx2/VU.h
@@ -134,6 +134,7 @@ struct __aligned16 VURegs {
 	// Current opcode being interpreted or recompiled (this var is used by Interps
 	// but not microVU.  Would like to have it local to their respective classes... someday)
 	u32 code;
+	u32 start_pc;
 
 	// branch/branchpc are used by interpreter only, but making them local to the interpreter
 	// classes requires considerable code refactoring.  Maybe later. >_<
diff --git a/pcsx2/VU0micro.cpp b/pcsx2/VU0micro.cpp
index 3be6664d0b..55de32c6f1 100644
--- a/pcsx2/VU0micro.cpp
+++ b/pcsx2/VU0micro.cpp
@@ -48,6 +48,8 @@ void __fastcall vu0ExecMicro(u32 addr) {
 	VU0.VI[REG_VPU_STAT].UL |=  0x01;
 	VU0.cycle = cpuRegs.cycle;
 	if ((s32)addr != -1) VU0.VI[REG_TPC].UL = addr;
+
+	CpuVU0->SetStartPC(VU0.VI[REG_TPC].UL << 3);
 	_vuExecMicroDebug(VU0);
 	CpuVU0->ExecuteBlock(1);
 }
diff --git a/pcsx2/VU0microInterp.cpp b/pcsx2/VU0microInterp.cpp
index acd0084adf..77ea99e13f 100644
--- a/pcsx2/VU0microInterp.cpp
+++ b/pcsx2/VU0microInterp.cpp
@@ -198,6 +198,11 @@ InterpVU0::InterpVU0()
 	IsInterpreter = true;
 }
 
+void InterpVU0::SetStartPC(u32 startPC)
+{
+	VU0.start_pc = startPC;
+}
+
 void InterpVU0::Step()
 {
 	vu0Exec( &VU0 );
diff --git a/pcsx2/VU1micro.cpp b/pcsx2/VU1micro.cpp
index 9a67898a75..f987269c67 100644
--- a/pcsx2/VU1micro.cpp
+++ b/pcsx2/VU1micro.cpp
@@ -61,7 +61,8 @@ void __fastcall vu1ExecMicro(u32 addr)
 	VU0.VI[REG_VPU_STAT].UL &= ~0xFF00;
 	VU0.VI[REG_VPU_STAT].UL |=  0x0100;
 	if ((s32)addr != -1) VU1.VI[REG_TPC].UL = addr;
-	_vuExecMicroDebug(VU1);
 
+	CpuVU1->SetStartPC(VU1.VI[REG_TPC].UL << 3);
+	_vuExecMicroDebug(VU1);
 	CpuVU1->Execute(vu1RunCycles);
 }
diff --git a/pcsx2/VU1microInterp.cpp b/pcsx2/VU1microInterp.cpp
index 9f158e6af4..efc731befa 100644
--- a/pcsx2/VU1microInterp.cpp
+++ b/pcsx2/VU1microInterp.cpp
@@ -202,6 +202,11 @@ void InterpVU1::Shutdown() noexcept {
 	vu1Thread.WaitVU();
 }
 
+void InterpVU1::SetStartPC(u32 startPC)
+{
+	VU1.start_pc = startPC;
+}
+
 void InterpVU1::Step()
 {
 	VU1.VI[REG_TPC].UL &= VU1_PROGMASK;
diff --git a/pcsx2/VUmicro.h b/pcsx2/VUmicro.h
index 86eb2acbb3..b0f6a8dd1d 100644
--- a/pcsx2/VUmicro.h
+++ b/pcsx2/VUmicro.h
@@ -85,6 +85,7 @@ public:
 	virtual void Reserve()=0;
 	virtual void Shutdown()=0;
 	virtual void Reset()=0;
+	virtual void SetStartPC(u32 startPC)=0;
 	virtual void Execute(u32 cycles)=0;
 	virtual void ExecuteBlock(bool startUp)=0;
 
@@ -172,6 +173,7 @@ public:
 	void Reset() { }
 
 	void Step();
+	void SetStartPC(u32 startPC);
 	void Execute(u32 cycles);
 	void Clear(u32 addr, u32 size) {}
 
@@ -192,6 +194,7 @@ public:
 	void Shutdown() noexcept;
 	void Reset();
 
+	void SetStartPC(u32 startPC);
 	void Step();
 	void Execute(u32 cycles);
 	void Clear(u32 addr, u32 size) {}
@@ -217,6 +220,7 @@ public:
 	void Shutdown() noexcept;
 
 	void Reset();
+	void SetStartPC(u32 startPC);
 	void Execute(u32 cycles);
 	void Clear(u32 addr, u32 size);
 	void Vsync() noexcept;
@@ -237,6 +241,7 @@ public:
 	void Reserve();
 	void Shutdown() noexcept;
 	void Reset();
+	void SetStartPC(u32 startPC);
 	void Execute(u32 cycles);
 	void Clear(u32 addr, u32 size);
 	void Vsync() noexcept;
diff --git a/pcsx2/x86/microVU.cpp b/pcsx2/x86/microVU.cpp
index 60e02baecf..d8ca0759f4 100644
--- a/pcsx2/x86/microVU.cpp
+++ b/pcsx2/x86/microVU.cpp
@@ -256,8 +256,9 @@ __fi bool mVUcmpProg(microVU& mVU, microProgram& prog, const bool cmpWholeProg)
 // Searches for Cached Micro Program and sets prog.cur to it (returns entry-point to program)
 _mVUt __fi void* mVUsearchProg(u32 startPC, uptr pState) {
 	microVU& mVU = mVUx;
-	microProgramQuick& quick = mVU.prog.quick[startPC/8];
-	microProgramList*  list  = mVU.prog.prog [startPC/8];
+	microProgramQuick& quick = mVU.prog.quick[mVU.regs().start_pc/8];
+	microProgramList*  list  = mVU.prog.prog [mVU.regs().start_pc/8];
+
 	if(!quick.prog) { // If null, we need to search for new program
 		std::deque<microProgram*>::iterator it(list->begin());
 		for ( ; it != list->end(); ++it) {
@@ -293,7 +294,7 @@ _mVUt __fi void* mVUsearchProg(u32 startPC, uptr pState) {
 		// If cleared and program not found, make a new program instance
 		mVU.prog.cleared	= 0;
 		mVU.prog.isSame		= 1;
-		mVU.prog.cur		= mVUcreateProg(mVU,  startPC/8);
+		mVU.prog.cur		= mVUcreateProg(mVU, mVU.regs().start_pc/8);
 		void* entryPoint	= mVUblockFetch(mVU,  startPC, pState);
 		quick.block			= mVU.prog.cur->block[startPC/8];
 		quick.prog			= mVU.prog.cur;
@@ -301,9 +302,13 @@ _mVUt __fi void* mVUsearchProg(u32 startPC, uptr pState) {
 		//mVUprintUniqueRatio(mVU);
 		return entryPoint;
 	}
+
 	// If list.quick, then we've already found and recompiled the program ;)
-	mVU.prog.isSame	= -1;
-	mVU.prog.cur	=  quick.prog;
+	mVU.prog.isSame = -1;
+	mVU.prog.cur = quick.prog;
+	// Because the VU's can now run in sections and not whole programs at once
+	// we need to set the current block so it gets the right program back
+	quick.block = mVU.prog.cur->block[startPC / 8];
 	return mVUentryGet(mVU, quick.block, startPC, pState);
 }
 
@@ -347,6 +352,11 @@ void recMicroVU1::Reset() {
 	mVUreset(microVU1, true);
 }
 
+void recMicroVU0::SetStartPC(u32 startPC)
+{
+	VU0.start_pc = startPC;
+}
+
 void recMicroVU0::Execute(u32 cycles) {
 	pxAssert(m_Reserved); // please allocate me first! :|
 
@@ -366,6 +376,12 @@ void recMicroVU0::Execute(u32 cycles) {
 		hwIntcIrq(6);
 	}
 }
+
+void recMicroVU1::SetStartPC(u32 startPC)
+{
+	VU1.start_pc = startPC;
+}
+
 void recMicroVU1::Execute(u32 cycles) {
 	pxAssert(m_Reserved); // please allocate me first! :|
 
diff --git a/pcsx2/x86/microVU_Branch.inl b/pcsx2/x86/microVU_Branch.inl
index 58beb29f6a..83b4e6f064 100644
--- a/pcsx2/x86/microVU_Branch.inl
+++ b/pcsx2/x86/microVU_Branch.inl
@@ -118,19 +118,19 @@ void mVUDTendProgram(mV, microFlagCycles* mFC, int isEbit) {
 
 void mVUendProgram(mV, microFlagCycles* mFC, int isEbit) {
 
-	int fStatus = getLastFlagInst(mVUpBlock->pState, mFC->xStatus, 0, isEbit);
-	int fMac	= getLastFlagInst(mVUpBlock->pState, mFC->xMac,    1, isEbit);
-	int fClip	= getLastFlagInst(mVUpBlock->pState, mFC->xClip,   2, isEbit);
+	int fStatus = getLastFlagInst(mVUpBlock->pState, mFC->xStatus, 0, isEbit && isEbit != 3);
+	int fMac	= getLastFlagInst(mVUpBlock->pState, mFC->xMac,    1, isEbit && isEbit != 3);
+	int fClip	= getLastFlagInst(mVUpBlock->pState, mFC->xClip,   2, isEbit && isEbit != 3);
 	int qInst	= 0;
 	int pInst	= 0;
 	microBlock stateBackup;
 	memcpy(&stateBackup, &mVUregs, sizeof(mVUregs)); //backup the state, it's about to get screwed with.
-	if(!isEbit)
+	if(!isEbit || isEbit == 3)
 		mVU.regAlloc->TDwritebackAll(); //Writing back ok, invalidating early kills the rec, so don't do it :P
 	else
 		mVU.regAlloc->flushAll();
 
-	if (isEbit) {
+	if (isEbit && isEbit != 3) {
 		memzero(mVUinfo);
 		memzero(mVUregsTemp);
 		mVUincCycles(mVU, 100); // Ensures Valid P/Q instances (And sets all cycle data to 0)
@@ -178,7 +178,7 @@ void mVUendProgram(mV, microFlagCycles* mFC, int isEbit) {
 	xMOV(ptr32[&mVU.regs().VI[REG_MAC_FLAG].UL],	gprT1);
 	xMOV(ptr32[&mVU.regs().VI[REG_CLIP_FLAG].UL],	gprT2);
 
-	if (!isEbit) { // Backup flag instances
+	if (!isEbit || isEbit == 3) { // Backup flag instances
 		xMOVAPS(xmmT1, ptr128[mVU.macFlag]);
 		xMOVAPS(ptr128[&mVU.regs().micro_macflags], xmmT1);
 		xMOVAPS(xmmT1, ptr128[mVU.clipFlag]);
@@ -204,14 +204,14 @@ void mVUendProgram(mV, microFlagCycles* mFC, int isEbit) {
 	}
 
 
-	if (isEbit || isVU1) { // Clear 'is busy' Flags
+	if ((isEbit && isEbit != 3) || isVU1) { // Clear 'is busy' Flags
 		if (!mVU.index || !THREAD_VU1) {
 			xAND(ptr32[&VU0.VI[REG_VPU_STAT].UL], (isVU1 ? ~0x100 : ~0x001)); // VBS0/VBS1 flag
 			//xAND(ptr32[&mVU.getVifRegs().stat], ~VIF1_STAT_VEW); // Clear VU 'is busy' signal for vif
 		}
 	}
 
-	if (isEbit != 2) { // Save PC, and Jump to Exit Point
+	if (isEbit != 2 && isEbit != 3) { // Save PC, and Jump to Exit Point
 		xMOV(ptr32[&mVU.regs().VI[REG_TPC].UL], xPC);
 		xJMP(mVU.exitFunct);
 	}
@@ -293,6 +293,21 @@ void normBranch(mV, microFlagCycles& mFC) {
 		eJMP.SetTarget();
 		iPC = tempPC;	
 	}
+	if (mVUup.mBit)
+	{
+		DevCon.Warning("M-Bit on normal branch, report if broken");
+		u32 tempPC = iPC;
+		u32* cpS = (u32*)&mVUregs;
+		u32* lpS = (u32*)&mVU.prog.lpState;
+		for (size_t i = 0; i < (sizeof(microRegInfo) - 4) / 4; i++, lpS++, cpS++) {
+			xMOV(ptr32[lpS], cpS[0]);
+		}
+		mVUendProgram(mVU, &mFC, 3);
+		iPC = branchAddr(mVU) / 4;
+		xMOV(ptr32[&mVU.regs().VI[REG_TPC].UL], xPC);
+		xJMP(mVU.exitFunct);
+		iPC = tempPC;
+	}
 	if (mVUup.eBit) { 
 		if(mVUlow.badBranch) 
 			DevCon.Warning("End on evil Unconditional branch! - Not implemented! - If game broken report to PCSX2 Team"); 
@@ -408,7 +423,27 @@ void condBranch(mV, microFlagCycles& mFC, int JMPcc) {
 		eJMP.SetTarget();
 		iPC = tempPC;
 	}
-
+	if (mVUup.mBit)
+	{
+		u32 tempPC = iPC;
+		u32* cpS = (u32*)&mVUregs;
+		u32* lpS = (u32*)&mVU.prog.lpState;
+		for (size_t i = 0; i < (sizeof(microRegInfo) - 4) / 4; i++, lpS++, cpS++) {
+			xMOV(ptr32[lpS], cpS[0]);
+		}
+		mVUendProgram(mVU, &mFC, 3);
+		xCMP(ptr16[&mVU.branch], 0);
+		xForwardJump32 dJMP((JccComparisonType)JMPcc);
+		incPC(4); // Set PC to First instruction of Non-Taken Side
+		xMOV(ptr32[&mVU.regs().VI[REG_TPC].UL], xPC);
+		xJMP(mVU.exitFunct);
+		dJMP.SetTarget();
+		incPC(-4); // Go Back to Branch Opcode to get branchAddr
+		iPC = branchAddr(mVU) / 4;
+		xMOV(ptr32[&mVU.regs().VI[REG_TPC].UL], xPC);
+		xJMP(mVU.exitFunct);
+		iPC = tempPC;
+	}
 	if (mVUup.eBit) { // Conditional Branch With E-Bit Set
 		if(mVUlow.evilBranch) 
 			DevCon.Warning("End on evil branch! - Not implemented! - If game broken report to PCSX2 Team");
@@ -471,6 +506,10 @@ void condBranch(mV, microFlagCycles& mFC, int JMPcc) {
 }
 
 void normJump(mV, microFlagCycles& mFC) {
+	if (mVUup.mBit)
+	{
+		DevCon.Warning("M-Bit on Jump! Please report if broken");
+	}
 	if (mVUlow.constJump.isValid) { // Jump Address is Constant
 		if (mVUup.eBit) { // E-bit Jump
 			iPC = (mVUlow.constJump.regValue*2) & (mVU.progMemMask);
diff --git a/pcsx2/x86/microVU_Compile.inl b/pcsx2/x86/microVU_Compile.inl
index a3b1195e90..a0617cac57 100644
--- a/pcsx2/x86/microVU_Compile.inl
+++ b/pcsx2/x86/microVU_Compile.inl
@@ -567,13 +567,18 @@ void* mVUcompile(microVU& mVU, u32 startPC, uptr pState)
 		}
 
 		if ((curI & _Mbit_) && isVU0) {
-			incPC(-2);
-			if (!(curI & _Mbit_)) { //If the last instruction was also M-Bit we don't need to sync again
-				incPC(2);
-				mVUup.mBit = true;
+			if (xPC > 0)
+			{
+				incPC(-2);
+				if (!(curI & _Mbit_)) { //If the last instruction was also M-Bit we don't need to sync again
+					incPC(2);
+					mVUup.mBit = true;
+				}
+				else
+					incPC(2);
 			}
 			else
-				incPC(2);
+				mVUup.mBit = true;
 		}
 
 		if (curI & _Ibit_) {
@@ -677,6 +682,7 @@ void* mVUcompile(microVU& mVU, u32 startPC, uptr pState)
 				mVUsetupRange(mVU, xPC, false);
 				incPC(2);
 				mVUendProgram(mVU, &mFC, 0);
+				normBranchCompile(mVU, xPC);
 				incPC(-2);
 				goto perf_and_return;
 			}
@@ -773,6 +779,7 @@ _mVUt void* __fastcall mVUcompileJIT(u32 startPC, uptr ptr) {
 	if (doJumpAsSameProgram) { // Treat jump as part of same microProgram
 		return mVUblockFetch(mVUx, startPC, ptr);
 	}
+	mVUx.regs().start_pc = startPC;
 	if (doJumpCaching) { // When doJumpCaching, ptr is a microBlock pointer
 		microVU& mVU = mVUx;
 		microBlock* pBlock = (microBlock*)ptr;