The code works at my end as well (there is a bug in the energy fas I
expected from that suspect piece of code in gpu.cpp I found this past
weekend which you can see by running with 4 processes) but I am not seeing
the crash that Scott is seeing here. It almost looks like DO_PARALLEL is
somehow not defined.
On Mon, Nov 5, 2012 at 11:12 PM, Ross Walker <ross.rosswalker.co.uk> wrote:
> Hey Scott,
>
> What compiler and MPI combination are you using here? I am trying to
> reproduce this using the current git tree with mvapich2-1.4.1p1 and gnu
> 4.4.5-6 with NVCC 4.2 and every works great in DPDP mode on either 1 or 2
> GPUs. Do you have a simple repo that you see problems with?
>
> See: http://bugzilla.ambermd.org/show_bug.cgi?id=205
>
> All the best
> Ross
>
>
> On 11/5/12 11:36 PM, "Scott Brozell" <sbrozell.rci.rutgers.edu> wrote:
>
> >Hi Ross,
> >
> >Was a notice posted somewhere ?
> >
> >If not then please put one on the amber gpu page.
> >It's a fair amount of work to investigate failing tests;
> >let's spare our users the several hours of work i already spent.
> >
> >thanks,
> >scott
> >
> >On Mon, Nov 05, 2012 at 10:29:08PM -0700, Ross Walker wrote:
> >> Hi Scott,
> >>
> >> We know. Scott Le Grand and I are looking into it although other pesky
> >> things keep getting in the way. Hope to have a fix soon along with CUDA
> >> 5.0 support.
> >>
> >> For now just run single GPU if using NMROPT.
> >>
> >> All the best
> >> Ross
> >
> >_______________________________________________
> >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 Tue Nov 06 2012 - 06:00:03 PST
