Saw this in archived mailing list, but I never saw the actual email from Rob Prowel:
Does the AppImage image contain active and working CUDA/nvenc or is it just a fallback to CPU encoding? My testing seems to indicate the latter, as I'm seeing zero difference in encoding times between libx264 and nvenc.
My ffmpeg transcoding scripts always show about a 1/6th encoding time when using nvenc.
Rob, the AppImages are created on computers with different graphics cards/capabilities then what a user may have and therefore you will not get the same nvenc/CUDA speedups unless you built the binary yourself.  There seems to be no way to create an AppImage that accomodates all of the variations.