[gmx-users] MD workstation for Gromacs

[Szilárd Páll]

On Mon, Oct 27, 2014 at 2:35 PM, Adelman, Joshua Lev <jla65 at pitt.edu> wrote:
>
> On Oct 27, 2014, at 3:14 AM, Carsten Kutzner wrote:
>
> Hi,
>
> On 27 Oct 2014, at 07:11, Mohammad Hossein Borghei <mh.borghei at gmail.com<mailto:mh.borghei at gmail.com>> wrote:
>
> Thank you Szilárd,
>
> So I would be really thankful if you could tell me which configuration is
> the best:
>
> 2x GTX 980
> 2x GTX 780 Ti
> 2x GTX Titan Black
> I would choose between the 980 and the 780Ti, which will give
> you about the same performance. Buy whatever card you can get
> for a cheaper price. The Titan Black will be too expensive in
> relation to performance.
>
> Carsten
>
>
>
> Just a note about the GTX 780 Ti. While I don't know if people have had problems running with Gromacs, the Amber developers are recommending against this card due high failure rates:
>
> See the "Supported GPUs" section of:
> http://ambermd.org/gpus/
>
> and the following mailing list thread:
> http://archive.ambermd.org/201406/0289.html
>
> Josh

I have never seen or heard of such issues, but, admittedly, I have
personally not used the 780 Ti a lot except for testing and
benchmarking. While I do not doubt that Scott and Ross from AMBER have
done extensive testing and have a good reason to flag the 780 Ti, the
issues they may simply not affect GROMACS. That's of course just my
educated guess based on the fact that they mention:

"Caution: The GTX-780Ti cards are NOT recommended at this time due to
the high failure rate, believed to be due to insufficient CAS refresh
voltage for the GPU DIMMS, of these models."

and that GROMACS kernels are compute-bound and utilizes the GPU global
memory quite little (<20% of peak bandwidth). It could also be that
factory-overclocked cards produce the high failure rate.

Hence, I still suggest proceeding with caution rather than excluding
the 780 Ti when considering accelerators to use/buy.

Cheers,
--
Szilárd

>
>
>
>
> On Mon, Oct 20, 2014 at 12:24 AM, Szilárd Páll <pall.szilard at gmail.com<mailto:pall.szilard at gmail.com>>
> wrote:
>
> Please send such questions/requests to the GROMACS users' list, I'm
> replying there.
>
> - For GROMACS 2x GTX 980 will be faster than one TITAN Z.
> - Consider getting plain DIMMs instead of ECC registered;
> - If you care about memory bandwidth, AFAIK you need 8 memory modules;
> this will not matter for GROMACS simulations, but it could matter for
> analysis or other memory-intensive operations;
>
> --
> Szilárd
>
>
> ---------- Forwarded message ----------
> From: Mohammad Hossein Borghei <mh.borghei at gmail.com<mailto:mh.borghei at gmail.com>>
> Date: Sun, Oct 19, 2014 at 12:34 PM
> Subject: MD workstation for Gromacs
> To: pall.szilard at gmail.com<mailto:pall.szilard at gmail.com>
>
>
> Dear Mr. Szilárd
>
> I saw your comments in Gromacs mailing list and I thought you can answer
> my question. I would be really thankful if you could tell me whether the
> attached configurations are appropriate for GPU computing in Gromacs. Which
> one is better? Can they be improved by not any increase in price?
>
> Kind Regards,
>
> --
> Mohammad Hossein Borghei
>
>
>
>
> Sent with MailTrack
> <https://mailtrack.io/install?source=signature&lang=en&referral=mh.borghei@gmail.com&idSignature=22>
>
>
>
>
> --
> Mohammad Hossein Borghei
> --
>
>
> --
> Dr. Carsten Kutzner
> Max Planck Institute for Biophysical Chemistry
> Theoretical and Computational Biophysics
> Am Fassberg 11, 37077 Goettingen, Germany
> Tel. +49-551-2012313, Fax: +49-551-2012302
> http://www.mpibpc.mpg.de/grubmueller/kutzner
> http://www.mpibpc.mpg.de/grubmueller/sppexa
>
>
>
>
> --
> Gromacs Users mailing list
>
> * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-request at gromacs.org.