On Thu, Apr 29, 2010, David Watson wrote:
> In my experience, mpich2 performs well for Amber on Mac.
> I am not sure about Amber 11 performance, but for my needs (REMD) mpich2
> allowed me to start a job which needed "4" CPUs on a Core Duo or Core 2
> Duo without complaining that there weren't enough CPUs.
Any version of MPI *should* allow this (more threads than physical cores).
Certainly openmpi also does this on Macs.
Just to be clear: the "configure_openmpi" script is just there to help those
who don't have some other version of MPI they really like. This script should
be portable and give reasonable performance on many platforms, but certainly
won't be optimal on all machines.
If you are having troubles with other MPI's, one debugging strategy is to
use configure_openmpi. If that works, (a) you at least have a working code;
(b) you can try to figure out what the difference is between openmpi and
whatever you have that is not working.
....dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Apr 29 2010 - 08:30:03 PDT
