On Wed, 2014-03-05 at 21:09 +0100, pavel.banas.upol.cz wrote:
>
> Dear all,
>
> we tried to compile AMBER with gnu compilers and it does not help a lot. The
> segfaults are less frequent but they persists and giving the same error
> message. The overheating is also not the case as in most cases (but not all)
> the segfault occurs at the very beginning of the simulation, when the
> temperature on the gpu is even significantly below the operating
> temperature. In the mean time, we updated drivers on different cluster, so
> we will be able to put our titans to that cluster and test their hardware.
> Anyway, still it seems to me that all 64 card being faulty is not very
> likely.
>
> Please, do you have any other idea, what might be wrong...bios, motherboard
> drivers, etc?? Please, does anybody have the same architecture (GPU
> SuperWorkstations 7047GR-TPRF with Super X9DRG-QF motherboards)?
Did you run the validation suite that Ross gave you a link to?
>
> thanks a lot, Pavel
>
> --
> Pavel Banáš
> pavel.banas.upol.cz
> Department of Physical Chemistry,
> Palacky University Olomouc
> Czech Republic
>
>
>
> ---------- Původní zpráva ----------
> Od: pavel.banas.upol.cz
> Komu: AMBER Mailing List <amber.ambermd.org>
> Datum: 4. 3. 2014 9:13:03
> Předmět: Re: [AMBER] pmemd.cuda segfaults
>
> "Hi Ross,
> thank you very much. Actually I have not seen nvidia-smi or amber output for
> titan-black card until now. So now it is clear that we have titan cards, not
> titan-black. Tomorrow we will be able to compile amber with gcc (as our
> admin has vacacy today), run all tests and I will let you know.
> thanks a lot,
>
> Pavel
>
>
--
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Mar 05 2014 - 12:30:07 PST
