the memory requirements of mri_glmfit are dependent on the number of subjects given as input, the greater the subjects, the greater the ram. but generally it is not a ram hog.

compiling with system specific flags might improve performance, but generally not enough to be worth the trouble. freesurfer already is compiled with maximal optimization (-O3). the GPU work is what we are targeting for substantially improved performance.

n.

On Thu, 2010-09-23 at 17:49 +0200, daniel geisler wrote:

Hi Nick!

Thanks a lot for your answer. I guess that I could only benefit from processing several subjects in parallel. Do you also have some estimations of the resources needed by mri-glmfit?

I notice that the source-code of freesurfer is also available. One could suppose that compiling with system-specific optimization flags improves performance in some way. Is it worth the trouble?

Thank you & best wishes from Dresden, Daniel

On Wed, Sep 22, 2010 at 10:50 PM, Nick Schmansky nicks@nmr.mgh.harvard.edu wrote: the second option is best because it uses the x86 processor (amd opteron), which is our common distribution. the first option uses the Itanium processor which we dont actively support because we no longer have access to a build machine.

    the recon-all stream is not parallelized so you wouldnt
    benefit from
    multi-cores or any special parallelization hardware.  runtime
    on the amd
    cluster would be about 25 hours per subject, and consume a
    peak of about
    2.7GB of ram.  each subject consumes around 400MB of hard disk
    space
    (budget 500MB per subject).  negligible temporary hard disk
    space is
    used.

    n.

    On Wed, 2010-09-22 at 19:17 +0200, daniel geisler wrote:
    > Dear Freesurfers!
    >
    > I am a new user of freesurfer and I intend to install it on
    a HPC
    > system. There are two different systems available (see end
    of
    > message). Could you please give me a hint which one is best
    suited wrt
    > hardware and software.
    >
    > In order to use these systems one also has to state some
    estimations
    > of the needed resources. In detail, I have to estimate:
    >
    >  * CPU time
    >  * maximal number of parallel used CPUs
    >  * size of used RAM
    >  * size of permanent HDD usage
    >  * size of temporary HDD usage
    >
    > Could you share some experience how many resources are
    usually used
    > for the segmentation (recon-all) of a single subject?
    >
    > Thank you & best wishes from Dresden,
    > Daniel
    >
    > ---
    > 1.SGI Altix 4700 (Mars)
    >
    >  * For highly parallel and memory-intensive applications
    >  * 2048 cores Intel Itanium II Montecito 1.6 GHz
    >  * 6,5 TB memory
    >  * One partition for interactive operation with 128 cores
    >  * 68 TB SAN disk storage
    >  * 1 PB SAN tape archive
    >  * SuSE Linux Enterprise Server 10, SGI ProPack 5, batch
    system LSF
    >  * 13,1 TFlop/s peak performance
    >
    > 2. Linux Networx Evolocity II PC-Farm  (Deimos)
    >
    >  * For capacity computing
    >  * 2576 cores AMD Opteron 2,6 GHz
    >  * 5,4 TB memory
    >  * 724 nodes with 1, 2, or 4 dual-core CPUs each
    >  * 68 TB SAN disk storage
    >  * SuSE Linux Enterprise Server 10, batch system LSF
    >  * 13,9 TFlop/s peak performance
    >
    > For software specification please refer to:
    >
    http://tu-dresden.de/die_tu_dresden/zentrale_einrichtungen/zih/hpc/hochleistungsrechner/hpc_software?set_language=en&cl=en
    > _______________________________________________
    > Freesurfer mailing list
    > Freesurfer@nmr.mgh.harvard.edu
    > https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer



    The information in this e-mail is intended only for the person
    to whom it is
    addressed. If you believe this e-mail was sent to you in error
    and the e-mail
    contains patient information, please contact the Partners
    Compliance HelpLine at
    http://www.partners.org/complianceline . If the e-mail was
    sent to you in error
    but does not contain patient information, please contact the
    sender and properly
    dispose of the e-mail.

--

........................................................ Drosselstraße 8 02827 Görlitz, Germany mobile: +49 (0) 174 7932227 e-mail: daniel.geisler@gmail.com ........................................................