[markrankin]

 Are you using DXVA2 (copy-back) or DXVA2 (native)? If you are using DXVA2 Native, that could be the cause your difference in rendering stats. madshi advises against using DXVA2 Native with madVR. There are a few image quality issues with this setting including a misalignment of the chroma information. You may be missing some processing, which is pushing your rendering stats down.

madVR is actually coded to run fastest in D3D11 Native. At this point, Error Diffusion dithering and highlight recovery strength under pixel shader tone mapping should be always be fastest with D3D11 Native, sometimes significantly faster. In reality, with proper working hardware, you shouldn't see ANY difference in rendering times in most circumstances with the different hardware decoders. So something seems wrong. It certainly shouldn't be possible for D3D11 to be any slower than any of the other decoders because madVR is coded to use this hardware decoder. 

I think using NGU Sharp is the way to go. Seeing a difference in upscaling is a factor of the quality of the source and the size of the pixel grid on your display. 4K upscaling will look best on 65' screens and larger.  

I'm sorry but I made a mistake writting. What I'm using exactly is:

LAV VIDEO Filter
1. DXVA2 (copy-back)
2. Hardware Device to Use: Automatic
3. Dithering Mode: Ordered Dithering

madVR:
Rendering/General Settings: Use Direct3D 11 for presentation and Present a Frame for Every VSync.

With these settings I get the stats posted.

If I change in LAV VIDEO Filter Hardware to use to D3D11 and Automatic the playback is not good. Dropped frames and bad results. I don't understand this behavior because as far as I know D3D11 works better in Windows 10, Directx 11 and madVR.

Thank you again.