Hi Cong,
The Etot difference is because of the EGB difference.
What kind of GPUs are these? Are the GPUs the volta architecture e.g V100?
I know there were some issues with racemic conditions on the volta
architecture and register spillage from one of my GPU kernels (though this
kernel is in amber20 and not in amber18) for the GB calculations on the
GPU.
Dave Cerutti knows more about this since he was been working on it/worked
on it. I know this was happening in newer version of Amber but I think it
is maybe the same issue in Amber 18 if it is one of the newer GPUs.
Regards,
Kellon
On Tue, Apr 14, 2020 at 11:25 AM Cong Liu <cong.liu.1.stonybrook.edu> wrote:
> Does anyone else have this issue when compiling Amber18 with CUDA10.1 and
> GCC/8.3.0? Is there any clue how to fix it?
>
> Thanks,
> Cong
>
>
>
> On Sat, Apr 11, 2020 at 10:25 AM Cong Liu <cong.liu.1.stonybrook.edu>
> wrote:
>
> > Hi David,
> >
> > Thanks for your response. Here is the screenshot of one of the failure:
> > [image: Screen Shot 2020-04-11 at 10.22.57.png]
> >
> > I guess the problem is more serious, since the EPtot and EGB are quite
> > different.
> >
> > Best,
> > Cong
> >
> >
> >
> > On Sat, Apr 11, 2020 at 8:03 AM David A Case <david.case.rutgers.edu>
> > wrote:
> >
> >> On Fri, Apr 10, 2020, Cong Liu wrote:
> >> >
> >> >I recently compiled Amebr18 with GCC/8.3.0 and CUDA/10.1. The
> compilation
> >> >went through well. However, *amber_cuda_serial *testing indicates 5
> >> "*file
> >> >comparison failure*": DPFP myoglobin_igb7, DPFP myoglobin_igb8, DPFP
> >> >myoglobin_igb8_gbsa, DPFP irest1_ntt0_igb7_ntc2, DPFP
> >> irest1_ntt0_igb8_ntc2
> >> >with "IEEE_UNDERFLOW_FLAG IEEE_DENORMAL" floating-point exceptions.
> >>
> >> The underflow flag is not worrisome, but you should look at the .dif
> >> files (collected in the logs folder) and see if the discrepancies look
> >> like roundoff errors or something more serious.
> >>
> >> ....dac
> >>
> >>
> >> _______________________________________________
> >> 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
>
--
Kellon A. A. Belfon, Graduate Student
Carlos Simmerling Laboratory
The Laufer Center for Physical and Quantitative Biology
The Department of Chemistry, Stony Brook University
Stony Brook, New York 11794
Phone: (347) 546-4237 <(347)+546+4237> Email: kellon.belfon.stonybrook.
<kellon.belfon.stonybrook.edu>edu
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Apr 14 2020 - 11:00:02 PDT