diff --git a/GPU/Directx9/ShaderManagerDX9.cpp b/GPU/Directx9/ShaderManagerDX9.cpp
index b600524db6..b58e9687d3 100644
--- a/GPU/Directx9/ShaderManagerDX9.cpp
+++ b/GPU/Directx9/ShaderManagerDX9.cpp
@@ -222,13 +222,27 @@ void ShaderManagerDX9::VSSetMatrix(int creg, const float* pMatrix) {
 }
 
 // Depth in ogl is between -1;1 we need between 0;1 and optionally reverse it
-static void ConvertProjMatrixToD3D(Matrix4x4 & in, bool invertedX, bool invertedY, bool invertedZ) {
+static void ConvertProjMatrixToD3D(Matrix4x4 &in, bool invertedX, bool invertedY, bool invertedZ) {
 	Matrix4x4 s;
 	Matrix4x4 t;
-	s.setScaling(Vec3(1, 1, invertedZ ? -0.5 : 0.5f));
+
+	s.setScaling(Vec3(gstate_c.vpWidthScale, gstate_c.vpHeightScale, invertedZ ? -0.5 : 0.5f));
 	float xoff = 0.5f / gstate_c.curRTRenderWidth;
+	xoff = gstate_c.vpXOffset + (invertedX ? xoff : -xoff);
 	float yoff = 0.5f / gstate_c.curRTRenderHeight;
-	t.setTranslation(Vec3(invertedX ? xoff : -xoff, invertedY ? -yoff : yoff, 0.5f));
+	yoff = gstate_c.vpYOffset + (invertedY ? yoff : -yoff);
+	t.setTranslation(Vec3(xoff, yoff, 0.5f));
+	in = in * s * t;
+}
+
+static void ConvertProjMatrixToD3DThrough(Matrix4x4 &in) {
+	Matrix4x4 s;
+	Matrix4x4 t;
+
+	s.setScaling(Vec3(1.0f, 1.0f, 0.5f));
+	float xoff = -0.5f / gstate_c.curRTRenderWidth;
+	float yoff = -0.5f / gstate_c.curRTRenderHeight;
+	t.setTranslation(Vec3(xoff, yoff, 0.5f));
 	in = in * s * t;
 }
 
@@ -310,7 +324,7 @@ void ShaderManagerDX9::VSUpdateUniforms(int dirtyUniforms) {
 		Matrix4x4 proj_through;
 		proj_through.setOrtho(0.0f, gstate_c.curRTWidth, gstate_c.curRTHeight, 0, 0, 1);
 
-		ConvertProjMatrixToD3D(proj_through, false, false, false);
+		ConvertProjMatrixToD3DThrough(proj_through);
 
 		VSSetMatrix(CONST_VS_PROJ_THROUGH, proj_through.getReadPtr());
 	}
diff --git a/GPU/Directx9/StateMappingDX9.cpp b/GPU/Directx9/StateMappingDX9.cpp
index 28211c8a73..21e5a1c640 100644
--- a/GPU/Directx9/StateMappingDX9.cpp
+++ b/GPU/Directx9/StateMappingDX9.cpp
@@ -713,7 +713,6 @@ void TransformDrawEngineDX9::ApplyDrawState(int prim) {
 			(regionY2 - regionY1) * renderHeightFactor,
 			0.f, 1.f);
 	} else {
-		// These we can turn into a glViewport call, offset by offsetX and offsetY. Math after.
 		float vpXScale = getFloat24(gstate.viewportx1);
 		float vpXCenter = getFloat24(gstate.viewportx2);
 		float vpYScale = getFloat24(gstate.viewporty1);
@@ -737,8 +736,6 @@ void TransformDrawEngineDX9::ApplyDrawState(int prim) {
 		vpY0 *= renderHeightFactor;
 		vpWidth *= renderWidthFactor;
 		vpHeight *= renderHeightFactor;
-		
-		// shaderManager_->DirtyUniform(DIRTY_PROJMATRIX);
 
 		float zScale = getFloat24(gstate.viewportz1) / 65535.0f;
 		float zOff = getFloat24(gstate.viewportz2) / 65535.0f;
@@ -748,15 +745,59 @@ void TransformDrawEngineDX9::ApplyDrawState(int prim) {
 
 		gstate_c.vpDepth = zScale * 2;
 
-		// D3D doesn't like viewports partially outside the target. Clamp the viewport for now. Should also adjust
-		// the projection matrix to compensate, really.
-		float left = std::max(0.0f, vpX0 + renderX);
-		float top = std::max(0.0f, vpY0 + renderY);
-		float right = std::min(left + vpWidth, renderWidth);
-		float bottom = std::min(top + vpHeight, renderHeight);
+		// D3D doesn't like viewports partially outside the target, so we
+		// apply the viewport partially in the shader.
+		float left = renderX + vpX0;
+		float top = renderY + vpY0;
+		float right = left + vpWidth;
+		float bottom = top + vpHeight;
+
+		float wScale = 1.0f;
+		float xOffset = 0.0f;
+		float hScale = 1.0f;
+		float yOffset = 0.0f;
+
+		// If we're within the bounds, we want clipping the viewport way.  So leave it be.
+		if (left < 0.0f || right > renderWidth) {
+			float overageLeft = std::max(-left, 0.0f);
+			float overageRight = std::max(right - renderWidth, 0.0f);
+			// Our center drifted by the difference in overages.
+			float drift = overageRight - overageLeft;
+
+			left += overageLeft;
+			right -= overageRight;
+
+			wScale = vpWidth / (right - left);
+			xOffset = drift / (right - left);
+		}
+
+		if (top < 0.0f || bottom > renderHeight) {
+			float overageTop = std::max(-top, 0.0f);
+			float overageBottom = std::max(bottom - renderHeight, 0.0f);
+			// Our center drifted by the difference in overages.
+			float drift = overageBottom - overageTop;
+
+			top += overageTop;
+			bottom -= overageBottom;
+
+			hScale = vpHeight / (bottom - top);
+			yOffset = -drift / (bottom - top);
+		}
+
 		depthRangeMin = std::max(0.0f, depthRangeMin);
 		depthRangeMax = std::min(1.0f, depthRangeMax);
 
+		bool scaleChanged = gstate_c.vpWidthScale != wScale || gstate_c.vpHeightScale != hScale;
+		bool offsetChanged = gstate_c.vpXOffset != xOffset || gstate_c.vpYOffset != yOffset;
+		if (scaleChanged || offsetChanged)
+		{
+			gstate_c.vpWidthScale = wScale;
+			gstate_c.vpHeightScale = hScale;
+			gstate_c.vpXOffset = xOffset;
+			gstate_c.vpYOffset = yOffset;
+			shaderManager_->DirtyUniform(DIRTY_PROJMATRIX);
+		}
+
 		dxstate.viewport.set(left, top, right - left, bottom - top, depthRangeMin, depthRangeMax);
 	}
 }
diff --git a/GPU/GPUState.h b/GPU/GPUState.h
index 55bb87bb57..b14a94f9b0 100644
--- a/GPU/GPUState.h
+++ b/GPU/GPUState.h
@@ -476,6 +476,11 @@ struct GPUStateCache
 	float vpWidth;
 	float vpHeight;
 	float vpDepth;
+	// Only used by Direct3D, not saved.
+	float vpXOffset;
+	float vpYOffset;
+	float vpWidthScale;
+	float vpHeightScale;
 
 	u32 curRTWidth;
 	u32 curRTHeight;