Poor processing resource usage in large datasets (ADNI)

I have been running fMRIPrep in a large (~300 rs sessions) dataset that is a subset from the ADNI database. I am using a 96 core Xeon server for the task and I noticed most of the time the script makes poor use of processing power, with few cores active. Of course, during computing intensive tasks such as ants registration and freesurfer recon-all the cores are all active, but during most of the less intensive tasks it seems the script doesn’t use parallel processing in an efficient way.

Am I doing something wrong? Is there any way to make better use of processing power when running fmriprep?

Do you have a job scheduler like slurm on your server?

This section of the documentation can provide some guidance: