On Thu, Mar 14, 2013 at 11:20 PM, Mary Varughese <maryvj1985.gmail.com>wrote:
> Sir,
>
> Infact this is a single GPU with 24 cores as i understand.
>
No. AMBER does not split the workload up between CPUs and GPUs.
Therefore, for every thread you request, it will try to use a GPU. If you
have 8 GPUs on a single node and you launch 24 threads to use every CPU,
then pmemd will try using each GPU to run 3 threads!
bugixes have been done.
>
Then they have not been done correctly. Check the output of
./patch_amber.py --patch-level
from AMBERHOME. For me it looks like this:
swails.Batman /usr/local/amber12 $ ./patch_amber.py --patch-level
Latest patch applied to AmberTools12: 35
Latest patch applied to Amber12: 15
Don't forget to recompile every executable you want to use -after- applying
the updates. The updates are only visible after rebuilding the programs.
However, trying to launch 24 GPU threads will certainly cause the problems
you are seeing.
Good luck,
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
Received on Fri Mar 15 2013 - 07:00:03 PDT
