I'm trying to simultaneously transcode multiple videos using ffmpeg. I do see substantial speed gain when using GPU on a single video file. When transcoding a single file the GPU load is quite low (about 15%), so I thought that I could run several processes in parallel. But running even two transcoding tasks in parallel takes almost 2x time. Why parallelization does not work?
I'm using AWS g4dn.2xlarge instance with Tesla T4 GPU and running a simple command like
ffmpeg -hwaccel cuvid -c:v h264_cuvid -y -i src.mkv -c:v h264_nvenc -b:v 300k out.mp4
Any help/insights on how to accelerate massive video transcoding is greatly appreciated.