 |
 |
Or a version that's multithreaded - matlab is an ideal candidate for the multi CPU system. Jb. Jonathan Brooks (Ph.D.) - Research Fellow MARIARC, University of Liverpool, Liverpool, UK Tel: +44 151 794 5629 Fax: +44 151 794 5635 Web: http://www.mariarc.liv.ac.uk -----Original Message----- From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of Bernd Markgraf Sent: 11 December 2001 18:53 To: [log in to unmask] Subject: Re: spm performance under solaris 8 dear kent, > As a fellow Sun fan, I would suggest upgrading to SPM99, and changing > the maxMem variable in spm_spm.m from the default value to something like > 2^31 (given the amount of ram you have) and that should help. i'm using the latest release and updates from the spm website. i already set maxMem to different values and 2^31 gives roughly 3-4% performance increase. still speed isn't at all comparable to spm running under windows. so something's going wrong here. i tried different benchmarks for matlab (especially mbench) and comparing 5.3 and 6.1 some test showed a significant slowdown (nested for loops took 280sec under 5.3 and 993sec under 6.1). i haven't had any time to take a look at the spm files but if there are lots of for loops it will definitely slow things down at least for the solaris version of matlab. also file io is taking far too much time, there is no swapping and i'm the only user, no other big processes running, matlab is taking ~100MB of ram and uses about 80% of one cpu and i just don't know why it would take that long to write 128kB. > Nice thing about your sun, is it will likely be up and running 99.99% of > the time, whereas, you will be restarting daily with your pc. I am a fan > of stability over raw speed. We normally start scripts for realignment, > normalization, smoothing, and statistics (single subject) and go home. The > suns reliably finish over night, our pc's are not so fortunate. i know what you're talking about ;-) but it's a shame that mathworks isn't able to produce a 64bit version of matlab for solaris. i guess that would speed up things quite a bit... bernd