diff --git a/GPU/Common/DrawEngineCommon.cpp b/GPU/Common/DrawEngineCommon.cpp
index 1a72aaf52f57..cbc231608ef0 100644
--- a/GPU/Common/DrawEngineCommon.cpp
+++ b/GPU/Common/DrawEngineCommon.cpp
@@ -184,7 +184,9 @@ void DrawEngineCommon::DispatchSubmitImm(GEPrimitiveType prim, TransformedVertex
 
 	int bytesRead;
 	uint32_t vertTypeID = GetVertTypeID(vtype, 0, decOptions_.applySkinInDecode);
-	SubmitPrim(&temp[0], nullptr, prim, vertexCount, vertTypeID, cullMode, &bytesRead);
+
+	bool clockwise = !gstate.isCullEnabled() || gstate.getCullMode() == cullMode;
+	SubmitPrim(&temp[0], nullptr, prim, vertexCount, vertTypeID, clockwise, &bytesRead);
 	DispatchFlush();
 
 	if (!prevThrough) {
@@ -681,7 +683,7 @@ uint64_t DrawEngineCommon::ComputeHash() {
 	return fullhash;
 }
 
-int DrawEngineCommon::ExtendNonIndexedPrim(const uint32_t *cmd, const uint32_t *stall, u32 vertTypeID, int cullMode, int *bytesRead, bool isTriangle) {
+int DrawEngineCommon::ExtendNonIndexedPrim(const uint32_t *cmd, const uint32_t *stall, u32 vertTypeID, bool clockwise, int *bytesRead, bool isTriangle) {
 	const uint32_t *start = cmd;
 	int prevDrawVerts = numDrawVerts_ - 1;
 	DeferredVerts &dv = drawVerts_[prevDrawVerts];
@@ -705,7 +707,7 @@ int DrawEngineCommon::ExtendNonIndexedPrim(const uint32_t *cmd, const uint32_t *
 		DeferredInds &di = drawInds_[numDrawInds_++];
 		di.indexType = 0;
 		di.prim = newPrim;
-		di.cullMode = cullMode;
+		di.clockwise = clockwise;
 		di.vertexCount = vertexCount;
 		di.vertDecodeIndex = prevDrawVerts;
 		di.offset = offset;
@@ -722,7 +724,7 @@ int DrawEngineCommon::ExtendNonIndexedPrim(const uint32_t *cmd, const uint32_t *
 }
 
 // vertTypeID is the vertex type but with the UVGen mode smashed into the top bits.
-bool DrawEngineCommon::SubmitPrim(const void *verts, const void *inds, GEPrimitiveType prim, int vertexCount, u32 vertTypeID, int cullMode, int *bytesRead) {
+bool DrawEngineCommon::SubmitPrim(const void *verts, const void *inds, GEPrimitiveType prim, int vertexCount, u32 vertTypeID, bool clockwise, int *bytesRead) {
 	if (!indexGen.PrimCompatible(prevPrim_, prim) || numDrawVerts_ >= MAX_DEFERRED_DRAW_VERTS || numDrawInds_ >= MAX_DEFERRED_DRAW_INDS || vertexCountInDrawCalls_ + vertexCount > VERTEX_BUFFER_MAX) {
 		DispatchFlush();
 	}
@@ -758,7 +760,7 @@ bool DrawEngineCommon::SubmitPrim(const void *verts, const void *inds, GEPrimiti
 	di.inds = inds;
 	di.indexType = (vertTypeID & GE_VTYPE_IDX_MASK) >> GE_VTYPE_IDX_SHIFT;
 	di.prim = prim;
-	di.cullMode = cullMode;
+	di.clockwise = clockwise;
 	di.vertexCount = vertexCount;
 	di.vertDecodeIndex = numDrawVerts_;
 	di.offset = 0;
@@ -824,10 +826,7 @@ void DrawEngineCommon::DecodeInds() {
 		const DeferredInds &di = drawInds_[i];
 
 		int indexOffset = drawVertexOffsets_[di.vertDecodeIndex] + di.offset;
-		bool clockwise = true;
-		if (gstate.isCullEnabled() && gstate.getCullMode() != di.cullMode) {
-			clockwise = false;
-		}
+		bool clockwise = di.clockwise;
 		// We've already collapsed subsequent draws with the same vertex pointer, so no tricky logic here anymore.
 		// 2. Loop through the drawcalls, translating indices as we go.
 		switch (di.indexType) {
diff --git a/GPU/Common/DrawEngineCommon.h b/GPU/Common/DrawEngineCommon.h
index a10e9f1cdef5..5827c958f439 100644
--- a/GPU/Common/DrawEngineCommon.h
+++ b/GPU/Common/DrawEngineCommon.h
@@ -96,8 +96,8 @@ class DrawEngineCommon {
 	// This would seem to be unnecessary now, but is still required for splines/beziers to work in the software backend since SubmitPrim
 	// is different. Should probably refactor that.
 	// Note that vertTypeID should be computed using GetVertTypeID().
-	virtual void DispatchSubmitPrim(const void *verts, const void *inds, GEPrimitiveType prim, int vertexCount, u32 vertTypeID, int cullMode, int *bytesRead) {
-		SubmitPrim(verts, inds, prim, vertexCount, vertTypeID, cullMode, bytesRead);
+	virtual void DispatchSubmitPrim(const void *verts, const void *inds, GEPrimitiveType prim, int vertexCount, u32 vertTypeID, bool clockwise, int *bytesRead) {
+		SubmitPrim(verts, inds, prim, vertexCount, vertTypeID, clockwise, bytesRead);
 	}
 
 	virtual void DispatchSubmitImm(GEPrimitiveType prim, TransformedVertex *buffer, int vertexCount, int cullMode, bool continuation);
@@ -111,8 +111,8 @@ class DrawEngineCommon {
 		}
 	}
 
-	int ExtendNonIndexedPrim(const uint32_t *cmd, const uint32_t *stall, u32 vertTypeID, int cullMode, int *bytesRead, bool isTriangle);
-	bool SubmitPrim(const void *verts, const void *inds, GEPrimitiveType prim, int vertexCount, u32 vertTypeID, int cullMode, int *bytesRead);
+	int ExtendNonIndexedPrim(const uint32_t *cmd, const uint32_t *stall, u32 vertTypeID, bool clockwise, int *bytesRead, bool isTriangle);
+	bool SubmitPrim(const void *verts, const void *inds, GEPrimitiveType prim, int vertexCount, u32 vertTypeID, bool clockwise, int *bytesRead);
 	template<class Surface>
 	void SubmitCurve(const void *control_points, const void *indices, Surface &surface, u32 vertType, int *bytesRead, const char *scope);
 	void ClearSplineBezierWeights();
@@ -257,7 +257,7 @@ class DrawEngineCommon {
 		u8 vertDecodeIndex;  // index into the drawVerts_ array to look up the vertexOffset.
 		u8 indexType;
 		s8 prim;
-		u8 cullMode;
+		bool clockwise;
 		u16 offset;
 	};
 
diff --git a/GPU/Common/SplineCommon.cpp b/GPU/Common/SplineCommon.cpp
index eeb2c73367c0..69c80d1c02cc 100644
--- a/GPU/Common/SplineCommon.cpp
+++ b/GPU/Common/SplineCommon.cpp
@@ -577,7 +577,7 @@ void DrawEngineCommon::SubmitCurve(const void *control_points, const void *indic
 	uint32_t vertTypeID = GetVertTypeID(vertTypeWithIndex16, gstate.getUVGenMode(), decOptions_.applySkinInDecode);
 	int generatedBytesRead;
 	if (output.count)
-		DispatchSubmitPrim(output.vertices, output.indices, PatchPrimToPrim(surface.primType), output.count, vertTypeID, gstate.getCullMode(), &generatedBytesRead);
+		DispatchSubmitPrim(output.vertices, output.indices, PatchPrimToPrim(surface.primType), output.count, vertTypeID, true, &generatedBytesRead);
 
 	if (flushOnParams_)
 		DispatchFlush();
diff --git a/GPU/GPUCommonHW.cpp b/GPU/GPUCommonHW.cpp
index 5f3709783f0d..b08839ac7795 100644
--- a/GPU/GPUCommonHW.cpp
+++ b/GPU/GPUCommonHW.cpp
@@ -989,7 +989,7 @@ void GPUCommonHW::Execute_Prim(u32 op, u32 diff) {
 	int cullMode = gstate.getCullMode();
 
 	uint32_t vertTypeID = GetVertTypeID(vertexType, gstate.getUVGenMode(), g_Config.bSoftwareSkinning);
-	if (!drawEngineCommon_->SubmitPrim(verts, inds, prim, count, vertTypeID, cullMode, &bytesRead)) {
+	if (!drawEngineCommon_->SubmitPrim(verts, inds, prim, count, vertTypeID, true, &bytesRead)) {
 		canExtend = false;
 	}
 	// After drawing, we advance the vertexAddr (when non indexed) or indexAddr (when indexed).
@@ -1024,11 +1024,12 @@ void GPUCommonHW::Execute_Prim(u32 op, u32 diff) {
 			// TODO: more efficient updating of verts/inds
 
 			u32 count = data & 0xFFFF;
+			bool clockwise = !gstate.isCullEnabled() || gstate.getCullMode() == cullMode;
 			if (canExtend) {
 				// Non-indexed draws can be cheaply merged if vertexAddr hasn't changed, that means the vertices
 				// are consecutive in memory.
 				_dbg_assert_((vertexType & GE_VTYPE_IDX_MASK) == GE_VTYPE_IDX_NONE);
-				int commandsExecuted = drawEngineCommon_->ExtendNonIndexedPrim(src, stall, vertTypeID, cullMode, &bytesRead, isTriangle);
+				int commandsExecuted = drawEngineCommon_->ExtendNonIndexedPrim(src, stall, vertTypeID, clockwise, &bytesRead, isTriangle);
 				if (!commandsExecuted) {
 					goto bail;
 				}
@@ -1046,7 +1047,7 @@ void GPUCommonHW::Execute_Prim(u32 op, u32 diff) {
 				// We can extend again after submitting a normal draw.
 				canExtend = isTriangle;
 			}
-			if (!drawEngineCommon_->SubmitPrim(verts, inds, newPrim, count, vertTypeID, cullMode, &bytesRead)) {
+			if (!drawEngineCommon_->SubmitPrim(verts, inds, newPrim, count, vertTypeID, clockwise, &bytesRead)) {
 				canExtend = false;
 			}
 			AdvanceVerts(vertexType, count, bytesRead);
diff --git a/GPU/Software/TransformUnit.cpp b/GPU/Software/TransformUnit.cpp
index eacdb481ae95..234ef5aca3ed 100644
--- a/GPU/Software/TransformUnit.cpp
+++ b/GPU/Software/TransformUnit.cpp
@@ -71,8 +71,8 @@ void SoftwareDrawEngine::DispatchFlush() {
 	transformUnit.Flush("debug");
 }
 
-void SoftwareDrawEngine::DispatchSubmitPrim(const void *verts, const void *inds, GEPrimitiveType prim, int vertexCount, u32 vertTypeID, int cullMode, int *bytesRead) {
-	_assert_msg_(cullMode == gstate.getCullMode(), "Mixed cull mode not supported.");
+void SoftwareDrawEngine::DispatchSubmitPrim(const void *verts, const void *inds, GEPrimitiveType prim, int vertexCount, u32 vertTypeID, bool clockwise, int *bytesRead) {
+	_assert_msg_(clockwise, "Mixed cull mode not supported.");
 	transformUnit.SubmitPrimitive(verts, inds, prim, vertexCount, vertTypeID, bytesRead, this);
 }
 
diff --git a/GPU/Software/TransformUnit.h b/GPU/Software/TransformUnit.h
index f2a2a9347234..295953a8e0ee 100644
--- a/GPU/Software/TransformUnit.h
+++ b/GPU/Software/TransformUnit.h
@@ -178,7 +178,7 @@ class SoftwareDrawEngine : public DrawEngineCommon {
 
 	void NotifyConfigChanged() override;
 	void DispatchFlush() override;
-	void DispatchSubmitPrim(const void *verts, const void *inds, GEPrimitiveType prim, int vertexCount, u32 vertType, int cullMode, int *bytesRead) override;
+	void DispatchSubmitPrim(const void *verts, const void *inds, GEPrimitiveType prim, int vertexCount, u32 vertType, bool clockwise, int *bytesRead) override;
 	void DispatchSubmitImm(GEPrimitiveType prim, TransformedVertex *buffer, int vertexCount, int cullMode, bool continuation) override;
 
 	VertexDecoder *FindVertexDecoder(u32 vtype);