I think you'd be extremely lucky to scale significantly past 96 CPU cores
(8 nodes)...
And even then you might do better running 10 processes per node instead of
12 (80 cores)...
Finally, disable hyperthreading.
OTOH a single M2090 GPU ought to be able to be able to run at 24 ns/day or
so depending on the mdin file parameters.
All IMO of course...
On Sat, Apr 14, 2012 at 4:45 AM, Albert <mailmd2011.gmail.com> wrote:
> hello:
>
> I am going to use Amber for hundreds of ns of a 40,000 atoms system.
> Does any body has any idea how many core should I use the most so that
> the efficiency can be the best? suppose each nodes has two sockets, each
> socket with 6 cors and 12 cores per modes.... Will 504 cores be a good
> choice?
>
> thank you very much
> BEST
> Albert
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sat Apr 14 2012 - 09:30:02 PDT
