note that MATLAB does have a new distributed computing toolbox for use
on clusters (not sure if it works on a single multiprocessor machine),
but I assume this would require recoding of the SPM matlab code and/or
mex files to take advantage of.
http://www.mathworks.com/products/distribtb/
cheers
russ
On Jan 4, 2005, at 2:38 PM, Neggers, S.F.W. (Bas) wrote:
> Dear colleage,
>
> for one analysis at a time you do not gain much from dual procs when
> using SPM and the conventional binary libraries, since they cannot
> split themselves into tiny "calculation packages" for parallel
> processing. I just looked up a post on the list that offered some
> tweaked binaries that apparently can split up spm processing, to take
> advantage of 2 procs:
>
> http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0312&L=spm&P=R6876&I=-1
>
> I did not find the time yet to try it out, I might try it later this
> year. When you have experience with it I would appreciate it to hear
> what you find.
>
> With conventional SPM, you do get some advantage because 1 proc is
> completely free for image processing, and the other can do OS related
> jobs for your windowing system, but this gain is modest. How much you
> save here heavliy depends on the linux distro and windows manager you
> use (assuming you do use linux). Windows is not good at either multi
> proc or 64 bit (for the latter no ready MS OS even exists as of yet).
>
> The real advantage (up to 100% faster) occurs when 2 users process
> data simultanously (each in a separate matlab instance) on the same
> machine as compared to a single user, since each one can use their own
> proc.
>
> For exact numbers you should do some benchmarking, or have a look at:
> http://www.psychology.nottingham.ac.uk/staff/cr1/spm99.html
>
> Good luck,
>
> Bas
>
> -------------------------------------------
> Dr. S.F.W. Neggers
> dept. of Psychonomics,Helmholtz Institute
> Utrecht University
> Heidelberglaan 2
> 3584 CS, Utrecht, room 17.09
> the Netherlands
> Tel: (+31) 30 253 4582 Fax: (+31) 30 534511
> E-mail: [log in to unmask]
> Web: http://www.fss.uu.nl/psn/pionier
> --------------------------------------------
> -----Original Message-----
> From: khead [mailto:[log in to unmask]]
> Sent: dinsdag 4 januari 2005 23:21
> To: Neggers, S.F.W. (Bas)
> Subject: Re: [SPM] single vs dual processor
>
>
> At 02:02 PM 1/4/2005, you wrote:
> Hi: I saw the below post on the SPM list and am curious as to exactly
> *how*
> much of an advantage you actually get from a dual operton system. We
> have a
> huge amount of work coming up and I was going to ask my boss if we
> could
> get a faster machine. Is the dual processor a good cost benefit for SPM
> (which is virtually all we do with our machine)? Currently to perform
> optimized VBM it takes about 20 minutes per file and I do upwards of 40
> files at a time. It sounds as if I could even break the job up it
> might be
> worth it in time savings. Thanks for any advice you can give me.
>
>> Dear Carolyn, list,
>>
>> I can second Satra's post. We run analyses of multiple users
>> simultanuously on a dual proc 64 bit opteron system. Note that matlab
>> jobs
>> itself cannot (yet) be divided over 2 procs. Still, a major advantage
>> is
>> that when one user runs an analyses, the next user has almost as good
>> a
>> performance, provided your license allows you to run multiple matlab
>> instances simultanuously. A second advantage I really appreciate is
>> that
>> when you run an analysis, it consumes all the CPU power of 1 proc,
>> and the
>> other does operating system related things, and you work as smoothly
>> on
>> your system as when no job is currently running. With one proc
>> working on
>> the system when an analyses is runnen is a nuisance.
>>
>> I also read on the list some time ago that there apparently are
>> efforts
>> out there to make the compiled C routines (that do all the hard work
>> for
>> SPM and matlab) multithreadable so that even with 1 job running you
>> might
>> be able to use both processors. You would have to search the list for
>> that.
>>
>> Good luck,
>>
>> Bas
>>
>> -------------------------------------------
>> Dr. S.F.W. Neggers
>> dept. of Psychonomics,Helmholtz Institute
>> Utrecht University
>> Heidelberglaan 2
>> 3584 CS, Utrecht, room 17.09
>> the Netherlands
>> Tel: (+31) 30 253 4582 Fax: (+31) 30 534511
>> E-mail: [log in to unmask]
>> Web: http://www.fss.uu.nl/psn/pionier
>> --------------------------------------------
>>
>> -----Original Message-----
>> From: SPM (Statistical Parametric Mapping)
>> [mailto:[log in to unmask]]On
>> Behalf Of Satrajit Ghosh
>> Sent: dinsdag 4 januari 2005 18:28
>> To: [log in to unmask]
>> Subject: Re: [SPM] single vs dual processor
>>
>>
>> We have been very happy with our dual processors, particularly because
>> we get to do several things simultaneously. However, given the
>> increasing size of data sets and memory bottlenecks that come with a
>> 32-bit machine, I would recommend purchasing a 64 bit processor with
>> lots of memory. Matlab 7.1 has support for 64 bit machines.
>>
>> Satra
>>
>> --
>> Satrajit Ghosh
>> Postdoctoral Associate
>> Speech Communications Group
>> Research Lab of Electronics, MIT
>>
>> On Tue, 4 Jan 2005 11:42:42 +0000, Cinly Ooi <[log in to unmask]> wrote:
>>> Volkmar Glauche wrote:
>>>
>>>> On Mon, 3 Jan 2005, Carolyn L. Fort wrote:
>>>>
>>>>> I'm purchasing a Linux workstation that will be devoted to fMRI
>>>>> processing/analysis and we'll primarily be running SPM (though
>>>>> AFNI,
>>>>> FSL and other packages will be loaded as well). I'm wondering
>>>>> whether
>>>>> SPM performs better on a dual processor or if sufficient speed
>>>>> (between
>>>>> 2.8 - 3.2 GB) on a single processor will suffice.
>>>>
>>>>
>>>> Matlab (and therefore SPM) will not make explicit use of a 2nd
>>>> processor,
>>>> but it is recommended if you want to use the workstation
>>>> interactively
>>>> while an analysis is running.
>>>
>>> Having a High speed processor waiting for interactive command is a
>>> bit
>>> of a waste. I would prefer to take the penalty of sluggish
>>> workstation
>>> response by running two analysis process on dual processor.
>>>
>>> Alternative is to get a single processor and hyperthread it. But
>>> analysis speed can be slower, assuming that Data access
>>> (Ethernet/Hard
>>> Disk/Cache) is not a limitting factor. May be dual processors,
>>> hyperthread to 4 processors is a better proposition (3 analysis + 1
>>> workstation activity). Unfortunately you cannot hyperthread only one
>>> processor in a dual processors config.
>>>
>>> Cinly
>>>
>
>
---
Russell A. Poldrack, Ph.d.
Assistant Professor
UCLA Department of Psychology
Franz Hall, Box 951563
Los Angeles, CA 90095-1563
phone: 310-794-1224
fax: 310-206-5895
email: [log in to unmask]
web: www.poldracklab.org
|