Thanks Scott. That makes a lot of sense.
On Fri, Sep 14, 2012 at 2:46 PM, Scott Le Grand <varelse2005.gmail.com>wrote:
> MPI performance of GTX 690 is abysmal because the two GPUs share the same
> PCIEX adaptor.
>
> That will improve down the road somewhat.
>
> In the meantime, I think you'll be happy at the performance of two
> independent runs (one on each GPU): 98+% efficiency when I last checked...
>
>
> On Fri, Sep 14, 2012 at 11:40 AM, Jonathan Gough <
> jonathan.d.gough.gmail.com
> > wrote:
>
> > Dear All,
> >
> > I'm curious if anyone is using a nvidia GTX 690 card. It has 2 cuda
> cores
> > on it, and when using pmed.cuda.MPI I am not getting much of a speed bump
> > when using both cores.
> >
> > Curious if others have experienced this or if there is something that I
> am
> > missing, or maybe this is what to expect. Just asking...
> >
> > Jonathan
> >
> > for example - on the benchmarks - using the standard inputs and
> > pmed.cuda.MPI:
> >
> > GB-myoglobin
> > 1 core - ns/day = 144.09
> > 2 core - ns/day = 156.14
> >
> > TRPCage
> > 1 core - ns/day = 698.11
> > 2 core - ns/day = 527.23
> >
> > Cellulose_prod_NPT
> > 1 core - ns/day = 3.60
> > 2 core - ns/day = 4.39
> >
> > JAC_production_NPT
> > 1 core - ns/day = 57.33
> > 2 core - ns/day = 67.92
> >
> > JAC_production_NVE
> > 1 core - ns/day = 72.42
> > 2 core - ns/day = 80.76
> >
> > FactorIX_production_NPT
> > 1 core - ns/day = 15.57
> > 2 core - ns/day = 18.50
> > _______________________________________________
> > 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
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Sep 14 2012 - 12:00:04 PDT
