[gmx-users] RE: Looking for GPU benchmarks

[Szilárd Páll]

Hi,

On Mon, Aug 27, 2012 at 5:34 PM, Mathieu38 <mathieu.dubois at bull.net> wrote:
> Hi,
>
>
>
> Thanks for your answer.
>
>
>
> The thing is that I am a benchmarker and I have no knowledge of the
> physics behind GROMACS. My goal is to demonstrate the interest (or not)
> in using GPUs for a customer test case.

I see.

> So my idea would be to do one run using the CPU only version of GROMACS,
> then modify the input file to add “cutoff-scheme = Verlet” and run the
> GPU version.

Turning on GPU acceleration is pretty much as simple as that, but
optimizing the performance is nowhere near as simple. This is, partly,
because the the problem is inherently complex, e.g. comparing the
performance of the algorithms present in 4.5.x with the new Verlet
scheme/NxN kernels is not really fair.

To give you a rough idea, with a typical bio-molecular input system on
a (balanced) Westmere + Tesla C20xx machine you can expect roughly 3x
speedup (when comparing the CPU-only runs using the Verlet scheme with
GPU accelerated runs).

> I am not sure how to modify any other parameters without changing the
> physics of the problem.

That depends very much on the simulated system and the kind of
properties one wants to look at. However, only one of my tips refers
to changing simulation parameters: picking the shortest cut-off
possible. I guess you could ask from the customer about this.

If you could share the log files of your runs it would be much easier
to spot potential bottlenecks and make concrete suggestions.

Cheers,
--
Szilárd

>
>
> Regards,
>
>
>
>
>
> Mathieu Dubois
>
>
>
> Hardware Accelerators expert
>
> Applications & Performances Team
>
> tel : +33 (0)4.76.29.70.56
>
> BULL, Architect of An Open World
>
> http://www.bull.com
>
>
>
> P Pensez à l’environnement avant d’imprimer / Before printing, think
> about the environment.
>
>
>
>
>
>
>
> De : Szilárd Páll [via GROMACS] <ml-node+s5086n5000577h4 at n6.nabble.com>
> [mailto:Szilárd Páll [via GROMACS]
> <ml-node+s5086n5000577h4 at n6.nabble.com>]
> Envoyé : lundi 27 août 2012 17:26
> À : Mathieu38 <mathieu.dubois at bull.net>
> Objet : Re: Looking for GPU benchmarks
>
>
>
>
> Which system did you run? What settings?
>
> A few tips:
> - Use CUDA 4.2 (5.0 on Kepler);
> - Have at least 10-20k atoms/GPU (and more to get peak GPU performance);
> - Use the shortest cut-off possible to allow CPU-GPU load balancing;
> - Due to initial domain-decomposition/parallelization overhead,
> scaling from one to two GPUs is affected by this overhead.
> (- If load balancing is limited, try using multiple MPI ranks per GPU.)
>
> --
> Szilárd
>
>
> On Mon, Aug 27, 2012 at 1:31 PM, Mathieu38 <[hidden email]> wrote:
>
>
>> I have tried the basic approcah of taking some of the input files that
> are
>> provided with the sources of Gromacs or on the website, and adding the
> line
>>
>> cutoff-scheme = Verlet
>>
>> in the grompp.mdp file
>>
>> However, I have not find a case where use of GPUs (2 MPI Tasks, 4
> OpenMP
>> Threads per MPI tasks, 2 GPUs) lead to a significant speed up,
> comparing to
>> a CPU only version using 8 cores on a single node.
>>
>> I don't know if this is because GPU implementation is still not
> performant
>> or if this is because my test cases are unappropriate.
>>
>> Any help here would be very much appreciated.
>>
>> Thx
>>
>>
>>
>>
>> --
>> View this message in context:
> http://gromacs.5086.n6.nabble.com/Looking-for-GPU-benchmarks-tp5000377p5
> 000567.html
>> Sent from the GROMACS Users Forum mailing list archive at Nabble.com.
>> --
>> gmx-users mailing list    [hidden email]
>> http://lists.gromacs.org/mailman/listinfo/gmx-users
>> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
>> * Please don't post (un)subscribe requests to the list. Use the
>> www interface or send it to [hidden email].
>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> --
> gmx-users mailing list    [hidden email]
> http://lists.gromacs.org/mailman/listinfo/gmx-users
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
> * Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to [hidden email].
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
>
>
> ________________________________
>
> If you reply to this email, your message will be added to the discussion
> below:
>
> http://gromacs.5086.n6.nabble.com/Looking-for-GPU-benchmarks-tp5000377p5
> 000577.html
>
> To unsubscribe from Looking for GPU benchmarks, click here
> <http://gromacs.5086.n6.nabble.com/template/NamlServlet.jtp?macro=unsubs
> cribe_by_code&node=5000377&code=bWF0aGlldS5kdWJvaXNAYnVsbC5uZXR8NTAwMDM3
> N3wtNTc4MTY2NzIx> .
> NAML
> <http://gromacs.5086.n6.nabble.com/template/NamlServlet.jtp?macro=macro_
> viewer&id=instant_html%21nabble%3Aemail.naml&base=nabble.naml.namespaces
> .BasicNamespace-nabble.view.web.template.NabbleNamespace-nabble.view.web
> .template.NodeNamespace&breadcrumbs=notify_subscribers%21nabble%3Aemail.
> naml-instant_emails%21nabble%3Aemail.naml-send_instant_email%21nabble%3A
> email.naml>
>
>
>
>
> --
> View this message in context: http://gromacs.5086.n6.nabble.com/Looking-for-GPU-benchmarks-tp5000377p5000578.html
> Sent from the GROMACS Users Forum mailing list archive at Nabble.com.
> --
> gmx-users mailing list    gmx-users at gromacs.org
> http://lists.gromacs.org/mailman/listinfo/gmx-users
> * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
> * Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists