[gmx-developers] max cpu scale-up

hessb at mpip-mainz.mpg.de hessb at mpip-mainz.mpg.de
Sat Sep 20 09:30:41 CEST 2008

> On Fri, Sep 19, 2008 at 1:31 PM, <hessb at mpip-mainz.mpg.de> wrote:
>> > What -ddorder cartesian does?
>> This determines the order of the PP and PME nodes.
>> If the XT4 would really support cartesian communication this would
>> be a lot faster, but unfortunately it does not.
>> Still it will probably be faster when all the PME nodes are physically
>> close toghether. -ddorder cartesian will do this.
>> -ddorder pp_pme will also work well, but I expect cartesian
>> to be slightly faster, you can try both.
> For one of my test systems interleave is fastest on XT4. But the
> difference
> is very small (5%). My system with 120k atoms scales to 160 cores (with 48
> cores for PME) giving 21.1ns/day. XT4 latency is really high. So that in
> some cases it is 50% slower than on low latency clusters (Myri 10G).
> Roland

You might also want to try -nosum option.
mdrun will tell you this at the end of the run.
At the end of the log file there is lots of information
on performance, load imbalance etc.


More information about the gromacs.org_gmx-developers mailing list