AndreaMG
SVP (via avisynth) works before the renderer resizing
Yes. You are right.
if you get SVP to work after resizing is done (maybe via avisynth?) the resources required will be much higher
...if you doing upscale. I think travolter told about reduce CPU Load by downscaling video before smoothing.
It would be great to be able to have SVP working after MadVR resizing
This approach have several principle limits:
- MadVR works with RGB (16 bit per channel internally), SVP - with YV12 only (12 bit on all channels);
- MadVR store frames in video memory, SVP needed for frames in operation memory;
- SVP has tighter integration with avisynth and ffdShow, we need to make independent standalone DS filter first.
But who knows? See it in the future 
travolter
Potplayer and FFdshow provide very similar performance
Thank you for test.