diff --git a/GPU/Common/DrawEngineCommon.cpp b/GPU/Common/DrawEngineCommon.cpp
index ed58732a7d..0ea6f18a60 100644
--- a/GPU/Common/DrawEngineCommon.cpp
+++ b/GPU/Common/DrawEngineCommon.cpp
@@ -168,7 +168,8 @@ void DrawEngineCommon::DispatchSubmitImm(GEPrimitiveType prim, TransformedVertex
 	uint32_t vertTypeID = GetVertTypeID(vtype, 0, applySkinInDecode_);
 
 	bool clockwise = !gstate.isCullEnabled() || gstate.getCullMode() == cullMode;
-	SubmitPrim(&temp[0], nullptr, prim, vertexCount, vertTypeID, clockwise, &bytesRead);
+	VertexDecoder *dec = GetVertexDecoder(vertTypeID);
+	SubmitPrim(&temp[0], nullptr, prim, vertexCount, dec, vertTypeID, clockwise, &bytesRead);
 	Flush();
 
 	if (!prevThrough) {
@@ -826,7 +827,7 @@ void DrawEngineCommon::SkipPrim(GEPrimitiveType prim, int vertexCount, u32 vertT
 }
 
 // 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, bool clockwise, int *bytesRead) {
+bool DrawEngineCommon::SubmitPrim(const void *verts, const void *inds, GEPrimitiveType prim, int vertexCount, VertexDecoder *dec, 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) {
 		Flush();
 	}
@@ -846,8 +847,11 @@ bool DrawEngineCommon::SubmitPrim(const void *verts, const void *inds, GEPrimiti
 
 	// If vtype has changed, setup the vertex decoder. Don't need to nullcheck dec_ since we set lastVType_ to an invalid value whenever we null it.
 	if (vertTypeID != lastVType_) {
-		dec_ = GetVertexDecoder(vertTypeID);
+		dec_ = dec;
+		_dbg_assert_(dec->VertexType() == vertTypeID);
 		lastVType_ = vertTypeID;
+	} else {
+		_dbg_assert_(dec_->VertexType() == lastVType_);
 	}
 
 	*bytesRead = vertexCount * dec_->VertexSize();
diff --git a/GPU/Common/DrawEngineCommon.h b/GPU/Common/DrawEngineCommon.h
index 38f41eca15..fd538c04ed 100644
--- a/GPU/Common/DrawEngineCommon.h
+++ b/GPU/Common/DrawEngineCommon.h
@@ -95,7 +95,8 @@ public:
 	// 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, bool clockwise, int *bytesRead) {
-		SubmitPrim(verts, inds, prim, vertexCount, vertTypeID, clockwise, bytesRead);
+		VertexDecoder *dec = GetVertexDecoder(vertTypeID);
+		SubmitPrim(verts, inds, prim, vertexCount, dec, vertTypeID, clockwise, bytesRead);
 	}
 
 	virtual void DispatchSubmitImm(GEPrimitiveType prim, TransformedVertex *buffer, int vertexCount, int cullMode, bool continuation);
@@ -118,7 +119,7 @@ public:
 	}
 
 	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);
+	bool SubmitPrim(const void *verts, const void *inds, GEPrimitiveType prim, int vertexCount, VertexDecoder *dec, u32 vertTypeID, bool clockwise, int *bytesRead);
 	void SkipPrim(GEPrimitiveType prim, int vertexCount, u32 vertTypeID, int *bytesRead);
 
 	template<class Surface>
@@ -145,14 +146,14 @@ public:
 		return numDrawVerts_;
 	}
 
-	VertexDecoder *GetVertexDecoder(u32 vtype) {
+	VertexDecoder *GetVertexDecoder(u32 vertTypeID) {
 		VertexDecoder *dec;
-		if (decoderMap_.Get(vtype, &dec))
+		if (decoderMap_.Get(vertTypeID, &dec))
 			return dec;
 		dec = new VertexDecoder();
 		_assert_(dec);
-		dec->SetVertexType(vtype, decOptions_, decJitCache_);
-		decoderMap_.Insert(vtype, dec);
+		dec->SetVertexType(vertTypeID, decOptions_, decJitCache_);
+		decoderMap_.Insert(vertTypeID, dec);
 		return dec;
 	}
 
diff --git a/GPU/GPUCommonHW.cpp b/GPU/GPUCommonHW.cpp
index b4a8c97a7d..d4f11518b0 100644
--- a/GPU/GPUCommonHW.cpp
+++ b/GPU/GPUCommonHW.cpp
@@ -997,6 +997,7 @@ void GPUCommonHW::Execute_Prim(u32 op, u32 diff) {
 	int cullMode = gstate.getCullMode();
 
 	uint32_t vertTypeID = GetVertTypeID(vertexType, gstate.getUVGenMode(), g_Config.bSoftwareSkinning);
+	VertexDecoder *decoder = drawEngineCommon_->GetVertexDecoder(vertTypeID);
 
 	// Through mode early-out for simple float 2D draws, like in Fate Extra CCC (very beneficial there due to avoiding texture loads)
 	if ((vertexType & (GE_VTYPE_THROUGH_MASK | GE_VTYPE_POS_MASK | GE_VTYPE_IDX_MASK)) == (GE_VTYPE_THROUGH_MASK | GE_VTYPE_POS_FLOAT | GE_VTYPE_IDX_NONE)) {
@@ -1037,7 +1038,7 @@ void GPUCommonHW::Execute_Prim(u32 op, u32 diff) {
 	// Cuts down on checking, while not losing that much efficiency.
 	bool onePassed = false;
 	if (passCulling) {
-		if (!drawEngineCommon_->SubmitPrim(verts, inds, prim, count, vertTypeID, true, &bytesRead)) {
+		if (!drawEngineCommon_->SubmitPrim(verts, inds, prim, count, decoder, vertTypeID, true, &bytesRead)) {
 			canExtend = false;
 		}
 		onePassed = true;
@@ -1115,7 +1116,7 @@ void GPUCommonHW::Execute_Prim(u32 op, u32 diff) {
 				}
 			}
 			if (passCulling) {
-				if (!drawEngineCommon_->SubmitPrim(verts, inds, newPrim, count, vertTypeID, clockwise, &bytesRead)) {
+				if (!drawEngineCommon_->SubmitPrim(verts, inds, newPrim, count, decoder, vertTypeID, clockwise, &bytesRead)) {
 					canExtend = false;
 				}
 				// As soon as one passes, assume we don't need to check the rest of this batch.
@@ -1140,6 +1141,7 @@ void GPUCommonHW::Execute_Prim(u32 op, u32 diff) {
 				canExtend = false;  // TODO: Might support extending between some vertex types in the future.
 				vertexType = data;
 				vertTypeID = GetVertTypeID(vertexType, gstate.getUVGenMode(), g_Config.bSoftwareSkinning);
+				decoder = drawEngineCommon_->GetVertexDecoder(vertTypeID);
 			}
 			break;
 		}