[gmx-users] Re: On the choosing of GPU for MD runs

[James Starlight]

Szilárd,


thanks for suggestions.

IT's only not quite understand for me in what GPU ( gtx or tesla) I'll
get best performance with the typical ssystem consisted of 50-80k
atoms with explicit solvent and PME ( assuming that I'm using small
cut-offs as you told). For example on the 400$  gtx 670 I have
performance similar to your at the GTX 680. I't intresting to me is
there any reason to buy modern tesla card ( which prise is > 2000$ )


James

2013/1/25 Szilárd Páll <szilard.pall at cbr.su.se>:
> On Mon, Jan 14, 2013 at 7:20 PM, James Starlight <jmsstarlight at gmail.com>wrote:
>
>> 1)As I understood from nvidia site at present time GPU's from the
>> tesla serie are the best for the calculations like md. For instasnce
>> NVIDIA Tesla C2075 have performance ~ 1 TFLOPS in the
>> single-precission calculations. I wounder to know how many 8-core CPU
>> must have typical cluster to obtain such performance ? Have someone
>> tried to use tesla GPU with gromacs ? What real performance in ns\day
>> have been obtained with the explicit solvent systems?
>>
>>
>> 2) Today I performed test gpu-based calculation with 2 different gpu's
>> ( GTX 670 vs GT 640). Both of that gpu's have the same GPU frequency
>> but GTX 670 has 256 bit memory interface width with ddr5 ram worked on
>> higher frequency. As the result I've obtained 50% more perfomance with
>> the GTX 670. Does the memory interface width as well as ram frequency
>> affect on total GPU performance in addition to GPU frequency?
>>
>>
> As mentioned before, the memory does not matter as much as the number of
> processing cores on the GPUs and their frequency.
> - GTX 670: 1344 "CUDA cores" (7 multiprocessors * 192 cores) running at 915
> Mhz.
> - GTX 640: 384 "CUDA cores" (2 multiprocessors * 192 cores) running at 900
> Mhz.
>
> The raw GPU performance with our current algorithms has a roughly linear
> relationship with the number of cores and frequency, so the raw GPU kernel
> performance difference between 670 and 640 should be around 3.5x. It is a
> bit surprising that you've only got 50% higher total performance with the
> 670.
>
> --
> Szilárd
>
>
>
>> James
>>
>>
>>
>> 2013/1/14 James Starlight <jmsstarlight at gmail.com>:
>> > Dear Gromacs Users!
>> >
>> >
>> > I wounder to know some detailes about choosing of the gpu for md with
>> > gromacs. In particular on what properties of the videoadapter should I
>> > pay most attention ? What modern gpu nvidia-series might give best
>> > performance (gtx 6xx, tesla or quadro series) ? Could you provide me
>> > with some bechmarks besides the information present on the Gromacs web
>> > ?
>> >
>> > For instance with the gpu Geforce GTX 670 + core i5 (4 cores) I have
>> > performance 10ns\day for explicit system with 67000 atoms ( protein in
>> > tip3p water). Have someone better results with common home-like
>> > desktop? :)
>> >
>> >
>> > James
>> --
>> 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
>>
> --
> 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