That's true, you need a lot of RAM for this to work. There isn't a way to
multithread probtrackx with only a single copy of the info in the RAM. The
diffusion samples are what take up the most memory in most probtrackx
instances. You can reduce memory usage by cropping the diffusion volume to
not include any empty space around the nodif_brain_mask.
Peace,
Matt.
-----Original Message-----
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf
Of Kenji
Sent: Wednesday, June 23, 2010 6:05 AM
To: [log in to unmask]
Subject: Re: [FSL] Probtrackx on SGE
Hi,
Thanks Matt.
>2) If you have individual pathways that take longer to track, launch
several probtrackx processes with fractions
> of the total number of samples you want to track and different --rseed
values.
In your method described, probtrackx needs many RAM because of reading
merged_th, ph, f .
Do you have plan that probtrackx allows single mask to parallelize tracking
in order to speed up processing ?
i.e. probtrackx splits number of seed voxels into number of threads and
tracks from divided seeds independently.
Thanks
Kenji
----- Original Message -----
From: "Matt Glasser" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Tuesday, June 22, 2010 12:00 AM
Subject: Re: [FSL] Probtrackx on SGE
> You can already paralellize probtrackx in several ways. 1) if you have
many
> subjects or pathways to track, submit each to SGE using fsl_sub. 2) If
you
> have individual pathways that take longer to track, launch several
> probtrackx processes with fractions of the total number of samples you
want
> to track and different --rseed values.
>
> Peace,
>
> Matt.
>
> -----Original Message-----
> From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf
> Of Kenji
> Sent: Monday, June 21, 2010 1:36 AM
> To: [log in to unmask]
> Subject: [FSL] Probtrackx on SGE
>
> Hi,
>
> When is ProbtrackX set up to be parallelized ? 4.1.6 ?
>
> Thanks
> Kenji
>
|