[gmx-users] Workstation choice

[Szilárd Páll]

On Tue, Sep 11, 2018 at 5:16 PM Wahab Mirco <
Mirco.Wahab at chemie.tu-freiberg.de> wrote:

> On 07.09.2018 20:41, Olga Selyutina wrote:
> > ...
> > These sets of CPU and GPU are suitable for price (in our region):
> > *GPU*
> > GTX 1070 ~1700MHz, cuda 1920 - $514
> > GTX 1080 ~1700MHz, cuda 2560 - $615
> > GTX 1070Ti ~1700MHz, cuda 2432 - $615
> > GTX 1080Ti ~1600MHz, cuda 3584 - $930
> >
> > *CPU*
> > Ryzen 7 2700X - $357
> > 4200MHz, 8/16 cores/threads, cache L1/L2/L3 768KB/4MB/16MB, 105W, max.T
> 85C
> >
> > Threadripper 1950X - $930
> > 4000MHz, 16/32 cores/threads, cache  L1/L2/L3 1.5/8/32MB, 180W, max.T 68C
> >
> > i7 8086K - $515
> > 4800MHz, 6/12 cores/threads, cache L2/L3 1.5/12MB, 95W, max.T 100C
> >
> > i7 8700K - $442
> > 4600MHz, 6/12 cores/threads, cache L2/L3 1.5/12MB, 95W, max.T 100C
> >
> > The most suitable combinations CPU+GPU are as follows:
> > 1) Ryzen 7 2700X + two GTX 1080 - $1587
> > 1.1) Ryzen 7 2700X + one GTX 1080 + one GTX 1080*Ti* - $1900 (maybe?)
> > 2) Threadripper 1950X + one GTX 1080Ti - $1860
> > 3) i7 8700K + two GTX 1080 - $1672
> > 4) Ryzen 7 2700X + three GTX 1070 - $1900
> > My suggestions:
> > Variant 1 seems to be the most suitable.
> > Variant 2 seems to be suitable only if the single simulation is running
> on
> > workstation
> > It’s a bit confusing that in synthetic tests/games performance of i7 8700
> > is higher than Ryzen 7 2700.
> > ...
>
> Sorry for jumping into the thread at this point, but depending on
> the problem size and type, it might happen that:
>
>    - a single R2-2700X possibly cannot always saturate TWO
>      GTX-1080/1080Ti (maybe Szilárd Páll can add some rule of
>      thumb whether and when it can),
>

Depends on the definition of "saturate"; for a typical ~100k membrane
protein simulation with h-bonds constraints, already four low-end Xeon
cores per GTX 1080 are sufficient to get ~90% of the ~ about peak perf (say
what you'd get 2-3x more cores/GPU). With a 1080 Ti and ~4 wimpy Xeon
cores, that'll be more around 75-80% of the peak perf. (Side-note: that's
with a single run/GPU -- for small inputs like Olga's I'd definitely try to
have 2 runs per GPU for efficiency.)

Ref: https://goo.gl/8DVqPV

Now, considering that the 2700X has 8 _fast_ cores, I'd expect it to work
fine with two 1080s (and likely with 1080 Ti's too), though it does depend
on the amount of CPU work.


>    - a larger Threadripper may suffer from memory bandwidth issues on
>      many parallel threads in Gromacs,
>

The number of cores will outweigh most bandwidth issues.


>    - due to AVX2_512, 6C/12T "Coffee Lake" i7 may be somehow faster than
>      even 8C/16T Ryzen2 (at some scenarios significantly?)
>

Without a GPU it may, with a GPU it will not. As I said earlier, the
benefits of AVX512 are mostly gone with the nonbondeds offloaded to a GPU.


> But it most probably depends on the circumstances. If you want to
> spend your money effectively, you have to perform at least some tests.
>
> I can provide some performance numbers I got after updating
> Gromacs to v.2018.3 lately on our aging lab hardware.
>
> Regards,
>
> Mirco
>
> ==> (I can provide the input file if required) ==>
> --
> 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.