Hi,
Amber doesn't scale well in CPUs, as the time required to communicate the
nodes and processes starts to bottleneck the calculation. You can find this
in the mailing list being discussed multiple times (e.g.
http://archive.ambermd.org/201808/0160.html). Do test runs with increasing
number of CPUS/nodes to see when the increase in performance levels off, or
if you need more performance evaluate migrating to a GPU system with pmemd.
Hope it helps,
El vie., 14 sept. 2018 a las 10:37, Leena Aggarwal (<leena.hrc.gmail.com>)
escribió:
> Dear Amber Users
> We are running MD simulation using sander.mpi in High Performance Computer
> Cluster with 56 cores. As we decrease the number of cores, the speed of
> simulation should decrease but according to "mdinfo" file the speed of
> simulation has increased as given below:
>
> 56 cores 13.05 ns/day
> 54 cores 14.33 ns/day
> 52 cores 14.85 ns/day
> 50 cores 14.78 ns/day
> 48 cores 14.53 ns/day
> 38 cores 13.62 ns/day
> 36 cores 13.52 ns/day
>
> It would be really beneficial, if anybody will suggest why is this
> happening?
>
> From
> Leena Aggarwal
> Department of Chemistry
> University of Delhi
> Delhi-110007
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
--
Stephan Schott Verdugo
Biochemist
Heinrich-Heine-Universitaet Duesseldorf
Institut fuer Pharm. und Med. Chemie
Universitaetsstr. 1
40225 Duesseldorf
Germany
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Sep 14 2018 - 02:00:02 PDT
