I’m confused with the parameters -nthreads
and -omp-nthreads
. I have 8 physical CPUs with hyperthreading and so 16 logical cores. With these settings I’m attempting at running two subjects in parallel.
What parameters would you choose in my case when running fmriprep-docker
with default options?
Is it more efficient to go with -nthreads 4
and -omp-nthreads 8
? -nthreads 8
and -omp-nthreads 7
as @effigies suggests in this thread? Or something else?
I would suggest using the defaults (for a 16-core machine, it will be --nthreads 16 --omp-nthreads 8
) unless you find there are issues with it.
Thank you @effigies. Is this the recommendation even if I’m attempting to run fmriprep-docker on two subjects in parallel (I run one instance per subject)? fmriprep is getting increasingly slow as more subjects are processed (it is currently processing the same two subjects for the last 2 days). I have sufficient memory and use the flag --mem 20
in each fmriprep-docker call.