Fmriprep memory issue

Hi!

I’m running fmriprep on a single subject with 2 tasks. Each task has around 900 volumes.
I specify an external disk with 2TB as a working directory. I also use --the low-mem flag. I also use 18Gb of RAM for the docker.
I also tried to run fmriprep for each task separately.
But no matter what I do, I run into memory issues. I get an error message about the broken process pool.

Do you have any idea what else can be done for this dataset?

1 Like

Hello,

You can try uploading the data to Brainlife.io and running fmriprep on their servers.

Best,
Steven

It now always stops at
could not run node: fmriprep_wf.single_subject_pilot4_wf.anat_preproc_wf.surface_recon_wf.autorecon_resume_wf.autorecon_surfs

It also shows this error:
Standard error:
ln: failed to create symbolic link ‘rh.white.H’: Operation not supported
Return code: 1

I’m not sure anymore if it’s actually a memory issue

Is this on Brainlife or on your machine?

If on your machine, do you have read/write permissions on all relevant drives?

Why are you providing the low memory flag if you allow 18 gb of memory via docker access? We ran similar images (i.e., # of volumes) using the following----

   --n_cpus '12' \
    --omp-nthreads 8 \
    --mem 20GB \

I would give that a go and see what happens.

Their docs are also helpful—see here:

https://fmriprep.org/en/stable/faq.html#how-much-cpu-time-and-ram-should-i-allocate-for-a-typical-fmriprep-run