[gmx-users] info about gpus

[Szilárd Páll]

On Fri, Feb 1, 2019 at 4:58 PM paul buscemi <pbuscemi at q.com> wrote:
>
> I’ve run these conditions recently. The scale from as single 1080ti to 2080ti is proportional to the core number. the increase I’ve seen with 1080+1080 is 1.3-1.5 times as fast  and perhaps 1.3 -1.6 times as fast depending on the model.

Careful with extrapolating across generations of GPUs especially
without considering simulation size effects!

Some GPU architectural improvements lead to performance improvements
much larger than what in "core" count (or more appropriately
core-count x frequency) would suggest. For instance, pair-interaction
computation shows a markedly larger (per core and per GHz) improvement
from e.g. Pascal to Turing than from Maxwell to Pascal. Other
computation does not improve nearly as much.

Additionally, performance and scaling will be a lot very different
depending on the size-regime; e.g. the difference between a 1080 and
1080 Ti may be marginal for smaller inputs while it can be 20-30% for
larger ones.


> In a million atom membrane model the two 1080ti’s ran at 12 ns/d in an equilibrated npt and the 1080ti +2080ti at ~15 ns day.  In either case more than enough time to go get some coffee.
>
> Personally  i feel that unless you  are building really large systems ( 10+ GPU’s ) and are working with models under about 200k atoms, then the two 1080ti are just fine.  Don’t expect a doubling on any case. Unless you are really strapped for time using two 1080’s may work.  going from 150 ns/day to 200 is a good time saver for smaller systems.
>
> If you Google  gromac benchmarks you will find several articles on the issue.
>
> PB
> > On Feb 1, 2019, at 5:58 AM, Szilárd Páll <pall.szilard at gmail.com> wrote:
> >
> > Hi,
> >
> > It greatly depends what your use-case is, i.e. simulation system and
> > type of study (but if you want to scale I assume you want few longer
> > trajectories).
> >
> > Cheers,
> > --
> > Szilárd
> >
> > On Thu, Jan 31, 2019 at 6:17 PM Stefano Guglielmo
> > <stefano.guglielmo at unito.it> wrote:
> >>
> >> Dear all,
> >> I am tryin to set a new workstation and I would like to know if there is a
> >> significant improvement in performance with two gpus (gtx 1080 ti or rtx
> >> 2080) rather than just one, and eventually with which cpu/ram requisite.
> >>
> >> Thanks in advance for any advice and suggestions
> >> Stefano
> >>
> >> --
> >> Stefano GUGLIELMO PhD
> >> Assistant Professor of Medicinal Chemistry
> >> Department of Drug Science and Technology
> >> Via P. Giuria 9
> >> 10125 Turin, ITALY
> >> ph. +39 (0)11 6707178
> >>
> >>
> >> <https://www.avast.com/sig-email?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=webmail>
> >> Mail
> >> priva di virus. www.avast.com
> >> <https://www.avast.com/sig-email?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=webmail>
> >> <#DAB4FAD8-2DD7-40BB-A1B8-4E2AA1F9FDF2>
> >> --
> >> 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.
> > --
> > 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.
>
> --
> 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.