Hi Douglas,
You asked a good question which prompted me to do an experiment with just "-threads 8 -itkthreads 8".
The run time has been quite slower (~8 hours) compared to ~5 hours it took with "-parallel -openmp 8".
I agree with you about the underestimation of run time with the latter flags in that page. Though the page stipulates
3 hours, I have never been able to get a freesurfer done in less than 5 hours. Maybe it is related to the resolution of my
data?
-Tashrif
.
-parallel -openmp 4
should run things with 4 processors and
parallelize across hemi where it can The run
time on that page is probably not accurate. But is it faster when you use those options?
On 7/16/2020 11:47 PM, Billah, Tashrif wrote:
This <https://surfer.nmr.mgh.harvard.edu/fswiki/ReleaseNotes>says "-openmp <num>
after -parallel", this <https://surfer.nmr.mgh.harvard.edu/fswiki/BrainstemSubstructures> says "export
ITK_GLOBAL_DEFAULT_NUMBER_OF_THREADS=2", while this <https://surfer.nmr.mgh.harvard.edu/fswiki/mri_coreg>
says "--threads nthreads" for parallel processing. I want to
run recon-all and want all processes under recon-all to
be parallelized. Which one(s) of the above flags should I
provide with recon-all? Please note that the values are not my
interest, the flags are. I know the proper values I should choose for the
flags on my machines.
So far I have been
trying "-parallel -openmp 4" option only but my run time stays over 5
hours contrary to less than 3 hours stipulated in the documentation.
-Tashrif
.
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer