Hello!
On 10/18/2013 01:24 PM, Muthukumaran R wrote:
> Dear Amber users,
>
> I plan to run amber 12 jobs simultaneously on GPU and CPU (Pmemd.MPI
> version), will the CPU jobs affect the efficiency of GPU JOB?
As long as you leave one CPU core per GPU simulation, the performance of
the GPU simulation will not significantly be affected by other CPU
intense tasks running on the same machine. Each pmemd.cuda process wants
to consume one CPU core on its own. If it gets one for itself, it's happy.
>
> configuration of our Work Station
>
> Processor(s)
>
> 1 x Intel® Xeon® E5-2630V2 processors (6-Core, 2.6GHz, 15M Cache)
A valid configuration for concurrently running simulations on that
machine would be:
- 4 or 5 CPUs for the CPU simulation (4 or 5 pmemd.MPI processes)
- 1 GPU simulation (1 pmemd.cuda process)
- nothing else running on that machine that consumes a lot of CPU.
In that configuration, top should tell you that your machines consumes
about 500 or 600 % of CPU, the load average should be 5 or 6. If the
load average goes beyond 6, you must be worried that your simulations
are not running as fast as they could.
I suppose that as long as you don't use 6 CPUs for the CPU simulation
and leave the machine alone otherwise, you should be fine.
If you are not sure, you can always perform some benchmarks on your own.
Cheers,
Jan-Philip
> GPU
>
>
>
> 1 x Nvidia Tesla K20C GPU Compute board with 5GB
>
>
>
> Chipset
>
>
>
> Intel® C602
>
>
>
>
>
> RAM
>
>
>
> 64GB DDR3-1333 ECC RDIMM
>
> HDDs
>
>
>
> 6 x 1000GB, 7200 RPM, Enterprise SATA, hot-plug HDDs
>
>
>
>
>
> Graphics
>
>
>
> Nvidia Quadro K600 with 1GB
>
>
>
>
> Thanks in advance.
> _______________________________________________
> 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 Fri Oct 18 2013 - 07:00:03 PDT
