Hi FSL Experts,
I am currently processing a large dataset and would like to seek some advice regarding parallel processing.
These are the specifications of my iMac: 3.2GHz , Intel Core i5, No. of processor: 1, Total number of cores: 4, L2 Cache per core: 256kb, L3 Cache: 6MB, Memory: 8GB.
Would it be useful to open 4 terminals and run RECON-ALL on 4 subjects simultaneously? Lastly, is there some sort of -flag I can use to maximize parallel RECON-ALLs ?
Currently, I am running 2 RECON-ALLs on 2 terminals and each subject take on average 7-9 hours for completion, which is rather fast I think.
Many Thanks.
Best Wishes, Elijah
Dear Elijah,
If I'm right, there are two options to accelerate FS execution:
1) The first option is using the FS openmp implementation http://surfer.nmr.mgh.harvard.edu/fswiki/OpenMP. I haven't tested this way but openmp allows to open multiple processing threads, and it is supported for the most time consuming steps of FS.
2) The second is using the GPU capabilities. This needs to have access to a GPU card, and add -use-gpu flag in your recon-all call. This activates the execution of the gpus implementations for CA Reg, SkullLTA, Inflation1 and so on, reducing the execution time (in my expertise, to 4-5 hours per subject).
I recently developed a scheduler to optimize the FS workflow execution using the GPUs implementation and a left and right hemisphere steps in parallel. This will be published soon on the Springer NeuroInformatics journal.
I hope that this will help a bit...
Cheers,
/Jordi
freesurfer@nmr.mgh.harvard.edu
-
Elijah Mak -
Jordi Delgado