[gmx-users] Horrendous scaling
David van der Spoel
spoel at xray.bmc.uu.se
Wed Jun 25 17:18:00 CEST 2003
On Tue, 2003-06-24 at 16:33, Senthil Kandasamy wrote:
> I guess this topic has been beaten to death, but here it is again.
> My machines are dual processor Athlon 2600+ using ethernet.
> The system is a reasonably large lipid bilayer with proteins(~40000
> atoms.) with PME electrostatics. I use fft_optimize, fft_order=4,
> shuufle and sort options
> 1 processor : in 1 hour walltime, 13.3 ps of simulation
> 2 processors(1 node) : in 1 hour walltime, 17.7 ps of simulation
> 4 processors (2 nodes) : in 1-hour walltime, 8.7 ps of simulation!!
> I can live with running on two processors if I can improve scaling by a
> little bit more.
> I will try to increase the fft_order to 6 and see what happens.
> Also, these machines are part of a super cluster using mpich-1.2.5
> Are there any compilation options for mpich and fftw that would improve
> performance? I do see a lot of suggestions for LAM., but none for mpich.
it looks like mpich isn't using the shared memory on a dual processor
node. Check whether you can find out some details about that. We have
actually seen considerable superscaling on two processors (but not with
PME yet). I would estimate that with well tuned LAM you would get up to
a factor of 1.6 or more than 21 ps/hour for such a system.
> gmx-users mailing list
> gmx-users at gromacs.org
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.
Dr. David van der Spoel, Dept. of Cell & Mol. Biology
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
More information about the gromacs.org_gmx-users