Advice to speedup model on CPU

Hello,

sorry for slow response. I’m not using AMI, instead I’m building my image in a Singularity container, if it is any important.

I do not use GPU, I use CPU, as it says in the topic. I was monitoring CPU load in htop. I did not use any options to limit system utilization. I just saw that system utilization is lower than it could be and deduced, that I can get faster, if I increase system utilization.

I than that system util should be more than 5 or 6, because I have 24 cores, hence the maximum utilization is 24.

Regarding change in waves, I mean change in utilization which I observe in htop. It is not uniform 5-6, it is rather periodically changing from 18 to 1. The period is about epoch time, but I’m not sure if it is aligned exactly, because htop update granularity was too coarse for me.

Could you please rephrase your last question?