Re: [AMBER] Amber md in HPC

From: David Case <david.case.rutgers.edu>
Date: Tue, 4 Jul 2017 09:57:28 -0400

On Tue, Jul 04, 2017, Garima Singh wrote:
>
> I am using fewer nodes this time using
> below pbs script. Job run and and still continue but it only give 7
> ns /day which is very slow.How to get fast result from HPC. Is it slow
> because of Sander module.Can i use PMEMD to get fast calculation using
> HPC.

> mpirun -np 160 sander.MPI ....

I thought I had replied to this earlier, but maybe the answer got lost...

1. 160 is *way* to many threads for sander.MPI. Try something like 8 or 16.
Experiment with your system to find the best number.

2. By all means, run pmemd.MPI: as long as you don't require features unique
to sander, pmemd will almost always be faster, sometimes by a lot (especially
with larger numbers of threads.) Again: run experiments to find the best
number of threads.

....dac


_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Jul 04 2017 - 07:00:03 PDT
Custom Search