Re: [AMBER] 256 CPUs restriction in AmberMD?

From: case <case.biomaps.rutgers.edu>
Date: Sun, 6 Nov 2011 21:10:28 -0500

On Sun, Nov 06, 2011, Garo Garabedyan wrote:

> Are you aware of restrictions in the AmberMD code (like hardcoded max
> number of) about parallel CPUs to 256?

Sander indeed has a MAX_MPI_PROCESSORS set to 256 in parallel.h. Before
you think about changing this, see if you get acceptable scaling up to 256
threads. BlueGene/P may differ from other platforms, but for most situations,
sander's scaling would not go so high, and might even be slower at 512 than
256.

If you do want to change the code, see the comments in parallel.h. I don't
think it would be very hard to do.

For large numbers of threads, we strongly recommend the use of pmemd, not
sander, which had a lot of effort put into to it to get better parallel
scaling. I don't believe that there is any such limit on the number of
threads in pmemd.

...good luck....dac


_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sun Nov 06 2011 - 18:30:02 PST
Custom Search