I'm trying to support all GPUs the best I can, so madVR is not favoring any GPU manufacturer. However, there are a couple of things I have no power over which favor either NVidia or AMD:
(1) NVidia can do hardware overlay, AMD can not. Overlay is a pretty good solution for windowed playback (-> pro NVidia).
(2) NVidia has custom resolutions, AMD has not (-> pro NVidia).
(3) I can do lossless native DXVA2 decoding/deinterlacing with AMD, without needing "copy-back", while with NVidia "copy-back" is needed to achieve lossless-ness (-> pro AMD).
(4) There's a chance I could use CUDA to speed up processing in future madVR versions, or to use it for new algorithms. Currently AMD's OpenCL support is to slow to be useful. I might be wrong about CUDA being useful, though, I haven't really tested its speed for madVR use yet (-> pro NVidia).
(5) There's a chance frame packed 3D output could work better on AMD, but I'm mostly guessing here. I could be wrong about this (-> pro AMD).
(6) I've been told that sometimes when using AMD, audio playback could suffer in reliability/latency (-> pro NVidia).
(7) AMD offers a "PC/TV levels" switch for HDMI RGB output, NVidia does not, but there's a hack available for NVidia (-> pro AMD).
I think overall NVidia might currently be the better solution. Mostly because of hardware overlay and CUDA, from my point of view.