Just provide more information. This strange thing does not happen if
all processors are in one node.
dawei
On Wed, Nov 25, 2009 at 12:05 PM, Da-Wei Li <lidawei.gmail.com> wrote:
> Dear All:
>
> I recently try explicit water replica exchange MD using Amber9. The md
> run is very slow, compared with normal constant T runs. A close look
> at the profile_mpi reveals that "Verlet update time" take about 3/4 of
> the CPU times at processor 0. This is not normal. This does not happen
> with constant T MD. I use CentOS and mvapich2.
>
> Anyone has some idea?
>
> best,
>
> dawei
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Nov 25 2009 - 09:30:05 PST
