FAST is a single threaded process, so you will have to write a script to
parallelize the processing of separate subjects. For some things, e.g.
BEDPOSTx, FSL works with the Sun Grid Engine (SGE) for parallelization
automatically. You could still install SGE and write a shell script to
submit the jobs to SGE, which would then take care of load balancing and
start the processes as resources were available. Thus, provided enough RAM,
you can double your processing time with 8 cores over 4. I would run FAST
on one of your datasets and find what its highest memory usage is. I would
multiply that by the number of cores you intend to run concurrently and then
get that much RAM.
Peace,
Matt.
-----Original Message-----
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf
Of Murat Bagci
Sent: Wednesday, June 03, 2009 4:41 PM
To: [log in to unmask]
Subject: [FSL] benefits of dual processor / quadcore for FAST
Dear all,
We have large number of MRI volumes to be segmented using FAST
(256x256 slices). We are deciding on a new computer for this purpose.
We are considering the following options:
1) Linux with single Xeon QuadCore processor (4 cores)
2) Linux with double Xeon QuadCore processors (8 cores)
3) Mac Pro
We are leaning towards a Linux based system for cost efficiency. I
have a few questions:
- What is the benefit (processing time?) of running FAST on 8 cores
rather than 4 cores?
- Do we need to install other software to run FAST on multiple processors?
- How much RAM is recommended?
- Is there an advantage to Mac Pro?
Thank you in advance.
--
Murat Bagci
|