[gmx-users] Optimal GPU setup for workstation with Gromacs 5

Kutzner, Carsten ckutzne at gwdg.de
Sat Apr 25 11:02:42 CEST 2015


> On 25 Apr 2015, at 03:39, Jingjie Yeo (IHPC) <yeojj at ihpc.a-star.edu.sg> wrote:
> 
> Hi Carsten,
> 
> Thank you so much for the information! Seems like the GTX is the way to go. One last question, regarding the Maxwell architecture, would there be any kind of backward compatibility issue, say if I need to use both Gromacs 4.6 and 5.0, since it is the newest architecture?
No problem there. Both 4.6 and 5.0 work fine with GTX 980.

Carsten

> 
> JJ
> 
> -----Original Message-----
> From: gromacs.org_gmx-users-bounces at maillist.sys.kth.se [mailto:gromacs.org_gmx-users-bounces at maillist.sys.kth.se] On Behalf Of Kutzner, Carsten
> Sent: Friday, April 24, 2015 5:51 PM
> To: gmx-users at gromacs.org
> Subject: Re: [gmx-users] Optimal GPU setup for workstation with Gromacs 5
> 
> Hi JJ,
> 
>> On 24 Apr 2015, at 10:53, Jingjie Yeo (IHPC) <yeojj at ihpc.a-star.edu.sg> wrote:
>> 
>> Hi Carsten,
>> 
>> Thank you so much for the information! I checked with my vendor on his opinions on the GTX and his reply was that GTX cards will not minimize calculation errors the way Tesla and Quadro cards will. Do you think this is an issue?
> The GeForce cards do not offer ECC memory, so they cannot pick up memory errors occurring on the GPU, which are however very unlikely. Run an extensive memory check (e.g. memtestCL) before first usage of the cards. We have tested nearly
> 300 GeForce cards, and only 7 of them had problems with memory, so we replaced them.
> 
> Carsten
> 
> 
>> 
>> JJ
>> 
>> -----Original Message-----
>> From: gromacs.org_gmx-users-bounces at maillist.sys.kth.se 
>> [mailto:gromacs.org_gmx-users-bounces at maillist.sys.kth.se] On Behalf 
>> Of Kutzner, Carsten
>> Sent: Friday, 24 April, 2015 16:40
>> To: gmx-users at gromacs.org
>> Subject: Re: [gmx-users] Optimal GPU setup for workstation with 
>> Gromacs 5
>> 
>> Hi JJ,
>> 
>>> On 24 Apr 2015, at 03:02, Jingjie Yeo (IHPC) <yeojj at ihpc.a-star.edu.sg> wrote:
>>> 
>>> Hi Carsten,
>>> 
>>> In this case of 2 x GTX980, as far as I can tell, the clock speed and 
>>> GPU ram is significantly lower. For simulations of more
>> The GTX 980s have a clock rate of about 1200 MHz, whereas the clock rate of the Tesla K40 is 732 MHz. The latter has more CUDA cores, however.
>> A rough estimate on how well the GROMACS short-ranged kernels perform on a card is the product of clock rate and CUDA cores. In this metric, the GTX 980 is slightly better than the K40. In addition, the GTX 980 also has the newer Maxwell generation chip, yielding somewhat higher performance due to better instruction scheduling.
>> 
>> The amount of GPU memory is almost never an issue with GROMACS unless you want to run enormously large MD systems. The largest system that we benchmarked had 12 million atoms and this was using 1.2 GB of GPU memory, so you could even run a couple of these on a 980.
>> 
>> In our tests with a 2 million atom system on a node with 2x E5-2680v2 processors, using two 980s resulted in a 14% increased GROMACS performance when compared to using two K40s.
>> 
>> Note that from the money you save by buying GTX instead of Tesla cards 
>> you can get another node to run another simulation on :)
>> 
>> Carsten
>> 
>> 
>>> than a million atoms, would it be advisable to go for more cores or more clock speed, or having both is the best case scenario?
>>> 
>>> JJ
>>> 
>>> -----Original Message-----
>>> Date: Thu, 23 Apr 2015 08:03:46 +0000
>>> From: "Kutzner, Carsten" <ckutzne at gwdg.de>
>>> To: "gmx-users at gromacs.org" <gmx-users at gromacs.org>
>>> Subject: Re: [gmx-users] Optimal GPU setup for workstation with
>>> 	Gromacs 5
>>> Message-ID: <FE42569F-159B-49B1-A644-86A866AAACF4 at mpibpc.mpg.de>
>>> Content-Type: text/plain; charset="us-ascii"
>>> 
>>> Hi,
>>> 
>>>> On 23 Apr 2015, at 08:03, Jingjie Yeo (IHPC) <yeojj at ihpc.a-star.edu.sg> wrote:
>>>> 
>>>> Dear all,
>>>> 
>>>> My workstation specs are 2 x Intel Xeon E5-2695v2 2.40 GHz, 12 Cores. I would like to combine this with an optimal GPU setup for Gromacs 5 running simulations with millions of atoms. May I know what are the recommended setups? My vendor proposed doing a dual K40 Tesla GPU setup, would that be optimal?
>>> I would propose to put two GTX 980 into that machine instead.
>>> This will give you the same GROMACS performance at a fraction of the price.
>>> 
>>> Carsten
>>> 
>>>> 
>>>> Best Regards,
>>>> JJ
>>>> --
>>>> 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.
>> 
>> --
>> Dr. Carsten Kutzner
>> Max Planck Institute for Biophysical Chemistry Theoretical and 
>> Computational Biophysics Am Fassberg 11, 37077 Goettingen, Germany 
>> Tel. +49-551-2012313, Fax: +49-551-2012302 
>> http://www.mpibpc.mpg.de/grubmueller/kutzner
>> http://www.mpibpc.mpg.de/grubmueller/sppexa
>> 
>> --
>> 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.
> 
> --
> Dr. Carsten Kutzner
> Max Planck Institute for Biophysical Chemistry Theoretical and Computational Biophysics Am Fassberg 11, 37077 Goettingen, Germany Tel. +49-551-2012313, Fax: +49-551-2012302 http://www.mpibpc.mpg.de/grubmueller/kutzner
> http://www.mpibpc.mpg.de/grubmueller/sppexa
> 
> --
> 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.

--
Dr. Carsten Kutzner
Max Planck Institute for Biophysical Chemistry
Theoretical and Computational Biophysics
Am Fassberg 11, 37077 Goettingen, Germany
Tel. +49-551-2012313, Fax: +49-551-2012302
http://www.mpibpc.mpg.de/grubmueller/kutzner
http://www.mpibpc.mpg.de/grubmueller/sppexa



More information about the gromacs.org_gmx-users mailing list