diff --git a/plugins/GSdx/GSRasterizer.h b/plugins/GSdx/GSRasterizer.h
index 035179ccbe..e90b499ec6 100644
--- a/plugins/GSdx/GSRasterizer.h
+++ b/plugins/GSdx/GSRasterizer.h
@@ -42,10 +42,14 @@ public:
 	void* param;
 
 	GSRasterizerData() 
-		: vertices(NULL)
+		: scissor(GSVector4i::zero())
+		, bbox(GSVector4i::zero())
+		, primclass(GS_INVALID_CLASS)
+		, vertices(NULL)
 		, count(0)
 		, solidrect(false)
 		, syncpoint(false)
+		, frame(0)
 		, param(NULL)
 	{
 	}
diff --git a/plugins/GSdx/GSRendererSW.cpp b/plugins/GSdx/GSRendererSW.cpp
index f30d927534..ce579526fb 100644
--- a/plugins/GSdx/GSRendererSW.cpp
+++ b/plugins/GSdx/GSRendererSW.cpp
@@ -185,6 +185,25 @@ void GSRendererSW::Draw()
 	data->solidrect = gd->sel.IsSolidRect();
 	data->frame = m_perfmon.GetFrame();
 
+	GSVector4i r = data->bbox.rintersect(data->scissor);
+
+	list<uint32>* fb_pages = m_context->offset.fb->GetPages(r);
+	list<uint32>* zb_pages = m_context->offset.zb->GetPages(r);
+
+	//
+
+	if(gd->sel.fwrite)
+	{
+		m_tc->InvalidatePages(fb_pages, m_context->offset.fb->psm);
+	}
+
+	if(gd->sel.zwrite)
+	{
+		m_tc->InvalidatePages(zb_pages, m_context->offset.zb->psm);
+	}
+
+	//
+
 	if(m_fzb != m_context->offset.fzb)
 	{
 		m_fzb = m_context->offset.fzb;
@@ -192,35 +211,39 @@ void GSRendererSW::Draw()
 		data->syncpoint = true;
 	}
 
-	if(m_context->FRAME.Block() == m_context->ZBUF.Block())
-	{
-		// Writing the same address in a different format is incompatible with our screen splitting technique,
-		// it must not necessarily be done by the same batch, enough if there are two in the queue one after eachother, 
-		// first it writes frame buffer at address X, then as z buffer also at address X, due to format differences the 
-		// block layout in memory is not the same.
-		//
-		// Most of these situations are detected by the previous m_fzb != m_context->offset.fzb check, 
-		// but when FRAME.Block() == ZBUF.Block() and f/z writes are switched on/off mutually then offset.fzb stays the same.
-		//
-		// Bully: FBP/ZBP = 0x2300
+	// - chross-check frame and z-buffer pages, they cannot overlap with eachother and with previous batches in queue
+	// - m_fzb filters out most of these cases, only have to be careful when the addresses stay the same and the output is mutually enabled/disabled and alternating (Bully FBP/ZBP = 0x2300)
 
-		data->syncpoint = true;
+	if(!data->syncpoint)
+	{
+		if(gd->sel.fwrite)
+		{
+			for(list<uint32>::iterator i = fb_pages->begin(); i != fb_pages->end(); i++)
+			{
+				if(m_fzb_pages[*i] & 0xffff0000) data->syncpoint = true; // already used as a z-buffer
+			}
+		}
 	}
 
-	GSVector4i r = data->bbox.rintersect(data->scissor);
+	if(!data->syncpoint)
+	{
+		if(gd->sel.zwrite)
+		{
+			for(list<uint32>::iterator i = zb_pages->begin(); i != zb_pages->end(); i++)
+			{
+				if(m_fzb_pages[*i] & 0x0000ffff) data->syncpoint = true; // already used as a frame buffer
+			}
+		}
+	}
 
 	if(gd->sel.fwrite)
 	{
-		m_tc->InvalidateVideoMem(m_context->offset.fb, r);
-
-		UseTargetPages(m_context->offset.fb, r);
+		UseTargetPages(fb_pages, 0);
 	}
 
 	if(gd->sel.zwrite)
 	{
-		m_tc->InvalidateVideoMem(m_context->offset.zb, r);
-
-		UseTargetPages(m_context->offset.zb, r);
+		UseTargetPages(zb_pages, 1);
 	}
 
 	if(s_dump)
@@ -322,13 +345,13 @@ void GSRendererSW::InvalidateVideoMem(const GIFRegBITBLTBUF& BITBLTBUF, const GS
 {
 	GSOffset* o = m_mem.GetOffset(BITBLTBUF.DBP, BITBLTBUF.DBW, BITBLTBUF.DPSM);
 
-	m_tc->InvalidateVideoMem(o, r);
+	list<uint32>* pages = o->GetPages(r);
+
+	m_tc->InvalidatePages(pages, o->psm);
 
 	// check if the changing pages either used as a texture or a target
 
-	list<uint32>* pages = o->GetPages(r);
-
-	for(list<uint32>::iterator i = pages->begin(); i != pages->end(); i++)
+	for(list<uint32>::const_iterator i = pages->begin(); i != pages->end(); i++)
 	{
 		uint32 page = *i;
 
@@ -347,11 +370,9 @@ void GSRendererSW::InvalidateLocalMem(const GIFRegBITBLTBUF& BITBLTBUF, const GS
 
 	list<uint32>* pages = o->GetPages(r);
 
-	for(list<uint32>::iterator i = pages->begin(); i != pages->end(); i++)
+	for(list<uint32>::const_iterator i = pages->begin(); i != pages->end(); i++)
 	{
-		uint32 page = *i;
-
-		if(m_fzb_pages[page])
+		if(m_fzb_pages[*i])
 		{
 			Sync(6);
 
@@ -360,25 +381,23 @@ void GSRendererSW::InvalidateLocalMem(const GIFRegBITBLTBUF& BITBLTBUF, const GS
 	}
 }
 
-void GSRendererSW::UseTargetPages(GSOffset* o, const GSVector4i& rect)
+void GSRendererSW::UseTargetPages(const list<uint32>* pages, int offset)
 {
-	list<uint32>* pages = o->GetPages(rect);
-
-	for(list<uint32>::iterator i = pages->begin(); i != pages->end(); i++)
+	for(list<uint32>::const_iterator i = pages->begin(); i != pages->end(); i++)
 	{
-		_InterlockedIncrement(&m_fzb_pages[*i]);
+		ASSERT(((short*)&m_fzb_pages[*i])[offset] < SHRT_MAX);
+
+		_InterlockedIncrement16((short*)&m_fzb_pages[*i] + offset);
 	}
 }
 
-void GSRendererSW::ReleaseTargetPages(GSOffset* o, const GSVector4i& rect)
+void GSRendererSW::ReleaseTargetPages(const list<uint32>* pages, int offset)
 {
-	list<uint32>* pages = o->GetPages(rect);
-
-	for(list<uint32>::iterator i = pages->begin(); i != pages->end(); i++)
+	for(list<uint32>::const_iterator i = pages->begin(); i != pages->end(); i++)
 	{
-		ASSERT(m_fzb_pages[*i] > 0);
+		ASSERT(((short*)&m_fzb_pages[*i])[offset] > 0);
 
-		_InterlockedDecrement(&m_fzb_pages[*i]);
+		_InterlockedDecrement16((short*)&m_fzb_pages[*i] + offset);
 	}
 }
 
diff --git a/plugins/GSdx/GSRendererSW.h b/plugins/GSdx/GSRendererSW.h
index 545349ad19..af249121bd 100644
--- a/plugins/GSdx/GSRendererSW.h
+++ b/plugins/GSdx/GSRendererSW.h
@@ -41,6 +41,8 @@ class GSRendererSW : public GSRendererT<GSVertexSW>
 		{
 			GSScanlineGlobalData* gd = (GSScanlineGlobalData*)_aligned_malloc(sizeof(GSScanlineGlobalData), 32);
 
+			gd->sel.key = 0;
+
 			gd->clut = NULL;
 			gd->dimx = NULL;
 
@@ -55,12 +57,12 @@ class GSRendererSW : public GSRendererT<GSVertexSW>
 				
 			if(gd->sel.fwrite)
 			{
-				m_parent->ReleaseTargetPages(m_fb, r);
+				m_parent->ReleaseTargetPages(m_fb->GetPages(r), 0);
 			}
 				
 			if(gd->sel.zwrite)
 			{
-				m_parent->ReleaseTargetPages(m_zb, r);
+				m_parent->ReleaseTargetPages(m_zb->GetPages(r), 1);
 			}
 
 			if(gd->clut) _aligned_free(gd->clut);
@@ -77,7 +79,7 @@ protected:
 	uint8* m_output;
 	bool m_reset;
 	GSPixelOffset4* m_fzb;
-	long m_fzb_pages[512];
+	uint32 m_fzb_pages[512]; // uint16 frame/zbuf pages interleaved
 	uint32 m_tex_pages[16];
 
 	void Reset();
@@ -90,8 +92,8 @@ protected:
 	void InvalidateVideoMem(const GIFRegBITBLTBUF& BITBLTBUF, const GSVector4i& r);
 	void InvalidateLocalMem(const GIFRegBITBLTBUF& BITBLTBUF, const GSVector4i& r, bool clut = false);
 
-	void UseTargetPages(GSOffset* o, const GSVector4i& rect);
-	void ReleaseTargetPages(GSOffset* o, const GSVector4i& rect);
+	void UseTargetPages(const list<uint32>* pages, int offset);
+	void ReleaseTargetPages(const list<uint32>* pages, int offset);
 	void UseSourcePages(const GSTextureCacheSW::Texture* t);
 
 	bool GetScanlineGlobalData(GSScanlineGlobalData& gd);
diff --git a/plugins/GSdx/GSTextureCacheSW.cpp b/plugins/GSdx/GSTextureCacheSW.cpp
index 3ac54470f3..2f45a9bcde 100644
--- a/plugins/GSdx/GSTextureCacheSW.cpp
+++ b/plugins/GSdx/GSTextureCacheSW.cpp
@@ -83,11 +83,9 @@ GSTextureCacheSW::Texture* GSTextureCacheSW::Lookup(const GIFRegTEX0& TEX0, cons
 	return t;
 }
 
-void GSTextureCacheSW::InvalidateVideoMem(GSOffset* o, const GSVector4i& rect)
+void GSTextureCacheSW::InvalidatePages(const list<uint32>* pages, uint32 psm)
 {
-	list<uint32>* pages = o->GetPages(rect);
-
-	for(list<uint32>::iterator p = pages->begin(); p != pages->end(); p++)
+	for(list<uint32>::const_iterator p = pages->begin(); p != pages->end(); p++)
 	{
 		uint32 page = *p;
 
@@ -97,7 +95,7 @@ void GSTextureCacheSW::InvalidateVideoMem(GSOffset* o, const GSVector4i& rect)
 		{
 			Texture* t = *i;
 
-			if(GSUtil::HasSharedBits(o->psm, t->m_TEX0.PSM))
+			if(GSUtil::HasSharedBits(psm, t->m_TEX0.PSM))
 			{
 				if(t->m_repeating)
 				{
@@ -185,7 +183,7 @@ GSTextureCacheSW::Texture::Texture(GSState* state, uint32 tw0, const GIFRegTEX0&
 
 	list<uint32>* pages = m_offset->GetPages(GSVector4i(0, 0, 1 << TEX0.TW, 1 << TEX0.TH));
 
-	for(list<uint32>::iterator i = pages->begin(); i != pages->end(); i++)
+	for(list<uint32>::const_iterator i = pages->begin(); i != pages->end(); i++)
 	{
 		uint32 page = *i;
 
diff --git a/plugins/GSdx/GSTextureCacheSW.h b/plugins/GSdx/GSTextureCacheSW.h
index 346bc1e196..f65bcf42ba 100644
--- a/plugins/GSdx/GSTextureCacheSW.h
+++ b/plugins/GSdx/GSTextureCacheSW.h
@@ -64,7 +64,7 @@ public:
 
 	Texture* Lookup(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, uint32 tw0 = 0);
 
-	void InvalidateVideoMem(GSOffset* o, const GSVector4i& r);
+	void InvalidatePages(const list<uint32>* pages, uint32 psm);
 
 	void RemoveAll();
 	void RemoveAt(Texture* t);
diff --git a/plugins/GSdx/GSThread.h b/plugins/GSdx/GSThread.h
index 026d1358af..bd6d403c01 100644
--- a/plugins/GSdx/GSThread.h
+++ b/plugins/GSdx/GSThread.h
@@ -204,7 +204,7 @@ protected:
 				}
 
 				{
-					// NOTE: this is scoped because we must make sure the last "item" is no longer around when Wait detects an empty queue
+					// NOTE: this is scoped because we must make sure the last item is no longer around when Wait detects an empty queue
 
 					T item = m_queue.front();
 
diff --git a/plugins/GSdx/stdafx.h b/plugins/GSdx/stdafx.h
index 9e256e0378..c4ff5070ea 100644
--- a/plugins/GSdx/stdafx.h
+++ b/plugins/GSdx/stdafx.h
@@ -331,6 +331,13 @@ typedef signed long long int64;
 		__asm__("lock; xaddl %[retval], %[Addend]" : [retval] "+r" (retval) : [Addend] "m" (*Addend) : "memory");
 		return retval;
 	}
+	
+	__forceinline long _InterlockedExchangeAdd16(volatile short* const Addend, const short Value)
+	{
+		long retval = Value;
+		__asm__("lock; xaddw %[retval], %[Addend]" : [retval] "+r" (retval) : [Addend] "m" (*Addend) : "memory");
+		return retval;
+	}
 
 	__forceinline long _InterlockedDecrement(volatile long* const lpAddend)
 	{
@@ -341,6 +348,16 @@ typedef signed long long int64;
 	{
 		return _InterlockedExchangeAdd(lpAddend, 1) + 1;
 	}
+	
+	__forceinline short _InterlockedDecrement16(volatile short* const lpAddend)
+	{
+		return _InterlockedExchangeAdd16(lpAddend, -1) - 1;
+	}
+	
+	__forceinline short _InterlockedIncrement16(volatile short* const lpAddend)
+	{
+		return _InterlockedExchangeAdd16(lpAddend, 1) + 1;
+	}
 
 	#ifdef __GNUC__