It would not surprise me if sander.MPI's efficiency degraded after 16
processors or so (I've never tried to measure efficiency using top's CPU
utilization percentage -- I'm more interested in ns/day and other measures
of time-until-simulation-completion). You should try using pmemd instead
(pmemd.MPI for Amber11), since it does a better job of load balancing, and
minimizing communications and cache misses (which means fewer wasted CPU
cycles).
HTH,
Jason
On Thu, Dec 15, 2011 at 7:57 AM, <abinayar.imsc.res.in> wrote:
>
>
> Dear Jason,
>
> Thank you very much for your immediate response.I was using
> the SANDER.MPI module. Using the command "top", I was able to
> see the %cpu efficiency of each processor, that indeed kept
> changing every second.
>
> I am planning to run a 50 ns simulation in 10 stages of 5 ns each.
>
> Regards,
> Abinaya Raju,
> Project Student,
> Institute of Mathematical sciences
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Dec 15 2011 - 07:00:02 PST
