[gmx-users] Number of Xeon cores per GTX 1080Ti

Szilárd Páll pall.szilard at gmail.com
Wed Apr 4 17:02:55 CEST 2018


Please don't shift the topic of an existing thread, rather create a
new one. Will reply to topics relevant to this discussion here, the
rest separately.


On Wed, Apr 4, 2018 at 4:01 PM, Raman Preet Singh
<ramanpreetsingh at hotmail.com> wrote:
> Szilard:
>
> Thanks for sharing this valuable info. Very helpful.
>
> We are looking forward to procure a Xeon Phi processor based system. Specs are yet to be finalized.
>
> I am not sure if you have tested Xeon Phi but can you give me insights on the processor. Can I expect the high number of cores to translate into a proportional increase in performance with Phi processor as compared to Xeon E5 series with lesser cores? Can the general estimation "7-10 "core-GHz" per fast Pascal GPU" be applied here too?
>
> I am also concerned that the Pascal GPUs come with 700+ to 3000+ GPU cores.

As a side-note that those are "marketing cores" -- Pascal GPUs come
with up to 30-60 cores (that NVIDIA calls SM's).

> How will the generalization "7-10 "core-GHz" per fast Pascal GPU" change with the number of GPU cores.

Note that this is a *very* rough estimate of resource need for a
balanced hardware setup; the balance can depend more on the simulation
setup than on whether you have 8 or 12 "core-GHz" CPU compute.

GROMACS does offload-based acceleration with GPUs, so whichever
computation (CPU or GPU) is the limiting factor in a certain run will
benefit from more resources. If you have, e.g. a Xeon Gold CPU per
Tesla P100, it's quite likely that the run will be GPU-bound and a
faster GPU will make a large difference. Conversely, if your runs are
severely CPU-bound, faster CPUs will help more.

It is really hard to make an accurate general assessment on how will a
certain CPU-GPU combination perform compared to another. What's
certain is that low-end (e.g. "700 core") GPUs with fast CPUs won't be
very worth it.

--
Szilárd

> Thanks!
>
> Raman
> --
> 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.


More information about the gromacs.org_gmx-users mailing list