On Mon, Sep 14, 2015 at 3:46 AM, Bala subramanian <bala.biophysics.gmail.com
> wrote:
> Friends,
> I am submitting a job, using the following control list, i am using
> pmemd.cuda.MPI in amber12 version.
>
A couple things to try:
- Is your Amber 12 installation completely up-to-date? You can run
"./update_amber --check" inside $AMBERHOME to check for updates. Make sure
all updates have been applied.
- Does the simulation work in serial?
If the simulation works in serial, but not parallel, then I suggest just
running separate simulations on each GPU instead of trying to use all of
them for the same simulation. GPU parallel scaling suffers fro the fact
that pmemd.cuda is highly optimized and the GPU-GPU communication is so
slow compared to the throughput of the GPU itself (and since Amber 12 does
not have P2P parallelism, to my knowledge). So even if it *did* work to
run in parallel, you would likely see very little speed improvement upon
doubling the number of GPUs you're using.
HTH,
Jason
--
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Sep 14 2015 - 06:30:04 PDT
