How to use all CPUs when training [Help Needed]

I have 8 CPUs, each has 4 cores. However, when I ran my model, it was always around 800% CPU utils, which is ~25% when I did a top. How could I use more CPUs? I have checked nvidia-smi and it is indeed working. My nproc is 8. I am using data loader with 20 workers.

Also, for a RNN model being trained on GPU, does it sound problematic if my CPU util is such high?

1 Like

Same issue here. Look forward to hearing back.