Hello,
Currently, multi-CPU/multi-GPU builds are not supported. i.e., if you had 8
CPUs with 4 Fermi cards, you could run 4 separate pmemd.cuda simulations
that will take up 4 CPU cores and all available GPUs. MPI and CUDA are not
currently compatible in the amber build. There's not much need for multiple
CPUs, though, since the intensive labor is done on the GPU and the CPU is
just there to "supervise" as I see it, though others can give a better
explanation. I think the bottleneck currently is communication off-board
for the GPU (i.e. each time the GPU has to send information back to the CPU
to be printed), discounting the actual calculation for a decently sized
system (i.e. parallelizing CPUs doesn't seem worthwhile). Others may chime
in and correct minor misconceptions I have about this, but the amber11
release does currently only support 1 CPU and 1 GPU, and I think
parallelization will go in the direction of making it 1 CPU for multiple
GPUs, not the other way around.
All the best,
Jason
On Wed, May 26, 2010 at 5:09 PM, Sasha Buzko <obuzko.ucla.edu> wrote:
> Hi all,
> has anyone done any comparison testing to determine the optimal number of
> CPU cores per executing GPU?
> I'm thinking of putting together a GPU system to run Amber. In this case,
> would a single quad core Xeon be enough for four (4) Fermi cards (GTX480,
> for example)? One core per card, in this case. Or should I really go for a
> 2:1 ratio and a dual quad-core cpu with fewer GPUs?
> Are there any other considerations for a multi-GPU hardware configuration?
>
> Thanks for any advice
>
> Sasha
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Graduate Student
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed May 26 2010 - 15:30:30 PDT
