After a lot of experiments and vmaf/eye measurments we concluded that SD video we do on cpu decoding/encoding (with libx264 library) with 2 passes (cause quality is more better then any gpu SD encoding)
HD we do with cpu decoding and h264_nvenc encoding (we got 3x speeder than on cpu encoding) and we have good quality
I got some idea, when 2 pass gpu encoding is used, then passlog file creates, but it is zero inside, all data is in memory as i get it.
When we use cpu 2 pass encoding, first pass without audio, which gives ~10x speed, if we do first pass on gpu then it gets 20x, but it does not make sense cause passlogfile is zero (so we cannot use it), and second pass on gpu encoding is bad on SD.
Is there any way to make a first pass on gpu and create passlogfile for use it on second pass with cpu encoding?