0
1
I use FFMPEG with NVENC HVEC and tried several options the last days but finally i was not able to reach the same quality as CPU Encoding. The main issue here is that fast scenes are getting unsharp. I read in some article that Twitch and YouTube are using GPU Encoder to compress videos, so the first question is, is it possible to reach the same quality with GPU Encoding with lower filesize or how are they doing this?
The second questions is if i'm doing something wrong here with my ffmpeg command or some idea to finetune it, as example i found some article that is telling "CRF" does not have an effect with x265 nvenc, other meanwhile told that this still works and this is confusing me. Here is my command
-y -preset slow -c:v hevc_nvenc -vf scale=1920x1080:flags=lanczos -rc cbr_ld_hq -rc-lookahead 32 -q 20 -r 60 -crf 19 -b:v 6.4M -movflags +faststart -c:a ac3 -b:a 320k -pass 1 -2pass -1 "D:\-=streaming=-\concats\x265-website_high.mp4"
Is there something i can do to compress it even better without using the CPU Encoder?
May you explain me a bit more what "YouTube doesn't use real CBR means", I read some article before some time that they might use a kind of neural network to improve the videoquality, but overall it is hard to find information for this. – Deex – 2018-10-08T16:09:18.597
If you play a YouTube video, the bit rate is by far not constant: Fast or detail-rich scenes will use appreciably more bandwidth. While I don't know, I consider it a quite educated guess, that this is YouTube's way to keep up a good subjective visual quality while not wasting bandwidth or storage. – Eugen Rieck – 2018-10-08T16:14:29.857