Hi Adam,
Just to add to Jason's comments in case your email didn't specifically
mean running a single calculation in parallel across the two cards. There
should be no problem having a GTX680 and a 580 in the same physical node.
Running two calculations, one on each GPU will work very well and this is
probably your best option. Running a single calculation across the two
cards will work but I suspect the performance will be less than just
running on either card individually. There are no considerations in the
pmemd.cuda.MPI code for running across heterogeneous GPUs sine this would
just massively overcomplicate things for little or no payoff.
All the best
Ross
On 9/21/12 10:00 PM, "Jason Swails" <jason.swails.gmail.com> wrote:
>
>
>On Sep 21, 2012, at 8:02 PM, Adam Jion <adamjion.yahoo.com> wrote:
>
>> Hi,
>>
>> Is it possible to use Amber12 to run a Fermi card (i.e GTX 580) with a
>>Kepler card (i.e. GTX 680) in parallel?
>> Will performance be affected?
>
>I would suggest against it. The 680 will be limited by the 580. I expect
>that the 680 can run faster by itself than the 2 of them together.
>
>HTH,
>Jason
>
>--
>Jason M. Swails
>Quantum Theory Project,
>University of Florida
>Ph.D. Candidate
>352-392-4032
>_______________________________________________
>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 Sep 22 2012 - 11:00:03 PDT
