Run subjects in parallel

I’m trying to understand how I can run standard fmriprep on several subjects in parallel and efficiently. I read this and this, but I’m still confused. If I have 128Gb RAM how would you run fmriprep if I vary the number of cores to be:

4 cores
8 cores
12 cores
16 cores

I think the critical parameters are nthreads and omp-nthreads, but I don’t know how I should specify these per subject given the number of cores above. How many subjects would you run in parallel (if at all) for each number of cores above?