Long latecny of when using HLS input

Do you mean both CPU loading and GPU loading are 40% for 10 streams? What is the loading of one stream?

How did you monitor the GPU loading? Can you post the log of “nvidia-smi dmon” command?