[gmx-users] new i9 processor

[Szilárd Páll]

On Mon, Oct 16, 2017 at 2:56 PM, Harry Mark Greenblatt
<harry.greenblatt at weizmann.ac.il> wrote:
> BS”D
>
> Dear Bernard,
>
>   Thank you for your reply and info.  Our experience with a second graphics card (when coupled with the appropriate number of cores) is that we garner close to a 100% performance gain.  We have access to nodes on a cluster with 16 cores (2 x E5-2630 v3) + 1 K80 (dual gpu on each card).   If I recall correctly, in our initial tests of those nodes, we compared 8 cores + 1 gpu vs 16 cores + 2gpu’s, and got close to double the performance.  This is also consistent with our performance on our workstations, most of which have a hexacore Xeon coupled to a Geforce GTX 770 (or 970 in more recent versions).  Use of the 16 cores + dual GPU compute nodes gives us close to double the performance relative to the workstations.  I suspect that your dual K40’s are only showing a 5-10% gain over the single Geforce card because the GTX 1080i on your workstations should be significantly faster than the K40 on the nodes.  This is due to the newer architecture (Pascal vs. Kepler), higher core count (3584 vs. 2880), and the higher clock speeds (1480 base vs. 745 base) on the 1080ti.
>
> I recall a statement by Szilárd once, in some email, about getting very good scaling (i.e., close to linear) with multiple GPU’s.

I hope I did say exactly that because in general it is not true: both
because we're accelerating (for now) a single task and because offload
to multiple GPU requires domain-decomposition.
The switch from no DD (only multi-threading) to DD has an initial hit
and going from 1 to 2 GPU you can therefore often see fairly moderate
scaling. From e.g. 2 to 4-way DD you can get near-linear scaling, but
it will depend on the system size.

Scaling to multiple GPUs, especially if by that you mean using the
same number of cores and increasing the GPU count will certainly not
give linear scaling in performance (time-to-solution) as only part of
the computation is reduced -- if that scales at all.

So generally: you'll need a well-balanced CPU cores to GPU ratio and
increasing the GPUs count only will only help if the run in GPU-bound
(and even than, only by a limited amount).

We're hoping to get some new features in the development branch soon
which will change some of these assumptions!

Cheers,
--
Szilárd

> Thanks again,
>
>
> Harry
>
>
> Dear Harry,
>
> we have several nodes with single i9 7900 decacore processors and one gtx 1080ti each. So far they show very good performance compared with dual 16 core xeon nodes with two K20x and even K40x cards. We reach a performance of >40ns per day for >90000 atoms with the i9 system, compared to 50 ns/day with the dual xeon + dual K40x and 30 ns/day with the dual xeon + dual K20x. So your configuration should easily outrun the xeon setups... since you want to use the 16 core i9 the second gtx you propose might even make sense but mb you should reconsider if the 5-10% speedup with the second card is really worth the buck ...
>
> Best,
>
>
>
> --------------------------------------------------------------------
> Harry M. Greenblatt
> Associate Staff Scientist
> Dept of Structural Biology           harry.greenblatt at weizmann.ac.il<../../owa/redir.aspx?C=QQgUExlE8Ueu2zs5OGxuL5gubHf97c8IyXxHOfOIqyzCgIQtXppXx1YBYaN5yrHbaDn2xAb8moU.&URL=mailto%3aharry.greenblatt%40weizmann.ac.il>
> Weizmann Institute of Science        Phone:  972-8-934-6340
> 234 Herzl St.                        Facsimile:   972-8-934-3361
> Rehovot, 7610001
> Israel
>
> --
> 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.