I am attempting to test GPU on a p3.2xlarge AWS server (1 Tesla V100 card, 61GB ram, 8 core). and seem to be hard limited by CPU with little to no usage on the GPU.
The setup is a Centos 7 machine running wowza/nginx to stream and transcode a rtmp 1080p 60fps buckbunny video.
I added 5 stream files from the same source URL and am attempting to use the transcoder on the 5 streams (again same source url from nginx).
The preset is as follows:
192000 bits per second (bps)
Use first available GPU
Key Frame Interval
Same as source (required for transrating)
The problem I am having is near 100% CPU usage and only 1-2% GPU usage. Is there something I am doing incorrectly or is there a way I can shift more of the load to GPU?