[gmx-users] Re: scaling issues in GROMACS

[David]

On Wed, 2003-10-15 at 19:19, Dean Johnson wrote:
> Hello,
> I've been trying to get GROMACS running on a 16 dual xeon 1.7/1.8
> cluster, with gigE, and I'm running into some problems. It isn't seeming
> to load-balance very well and cpu utilization for 1 process per cpu is
> on the order of 10-20%. It seems to stay at about that level from 8 cpus
> on. Is this consistent with expectations? If not, any idea of what the
> problem may be? I'm using latest of everything, including mpich.
> 
Our experience (of some time ago, admittedly) is tha LAM is faster. It
also depends a lot on the system that you are simulating. Start out by
trying to reproduce the DPPC benchmark (use grompp -shuffle)

> Also, do you know if the OSX semaphore issue has been resolved on
> Apple's side? We are currently benchmarking G5s vs. Opterons vs.
> Itanium2's, for both GROMACS and Amber, and we want everyone to get a
> fair shake.
The OS-X bit I don't know, performancewise for gromacs they will be
roughly equally fast as a Xeon of the same clockspeed.
They are a bit cheaper though.

According to Erik Lindahl a 900 MHz Itanium is slightly faster than a 3
GHz Xeon for gromacs (next version), and new Itanium chips with higher
clock and lower amount of cache are coming out soon, so that might be an
option. If have recently received a couple of Opterons but haven't had
time to play with them.
> 
Please put further questions to the gmx-users list where more experts
are listening.

>         -Dean
-- 
David.
________________________________________________________________________
David van der Spoel, PhD, Assist. Prof., Molecular Biophysics group,
Dept. of Cell and Molecular Biology, Uppsala University.
Husargatan 3, Box 596,  	75124 Uppsala, Sweden
phone:	46 18 471 4205		fax: 46 18 511 755
spoel at xray.bmc.uu.se	spoel at gromacs.org   http://xray.bmc.uu.se/~spoel
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++