AMBER Archive (2002)

Subject: Re: parallel sander on T3E

From: Natasja Brooijmans (nbrooij_at_itsa.ucsf.edu)
Date: Sat May 25 2002 - 10:44:34 CDT


Look at the amber benchmarks page
(http://www.amber.ucsf.edu/amber/amber7.bench4.html) for speed ups with
different number of nodes. Generally nice speed ups are seen up to 8
nodes, beyond that you don't gain a lot of speed up.

Best,

Natasja

Natasja Brooijmans
Graduate Program in Chemistry & Chemical Biology
Department of Pharmaceutical Chemistry
University of California, San Francisco
San Francisco, CA 94143-0446
phone: 415-476 8291
fax: 415-502 1411
e-mail: nbrooij_at_itsa.ucsf.edu

On Sat, 25 May 2002, Peter Gannett wrote:

> Amber users:
>
> We ran two MD simulations on two solvated DNAs, a duplex and a triplex. Both runs were 1 ns and 16 nodes (64 CPU) were used. Both runs took about twice as long as we had expected based on a single cpu run. I am guessing that there must be some point at which additional CPUs don't get you any speed up. If true, than how can you make a rough estimate of the optimal number of CPUs to use for a given job. If I'm wrong in my guess, then what in our setup might be the problem. Thanks.
>
> Pete Gannett
>
>