Thanks Ross for your prompt answer !
Regarding to Amber GPU calculations on AMD systems I have good cca year
experience
with Amber GPU calculations on:
Asus M4N98TD EVO - nForce 980a SLI
AMD Phenom II X4 955 Black Edition
2 x Tesla C2050
We have also several AMD computers (I don't have specs) in lab equipped
with GTX 470s
and Amber GPU calc. are also just fine here if one did not mention some
rare
"unstability" issues ...
Just to be complete I have to say that I never tried parallel AMBER/GPUs
jobs, just
satisfied with single GPU jobs.
Best wishes,
Marek
Dne Thu, 15 Sep 2011 19:37:01 +0200 Ross Walker <ross.rosswalker.co.uk>
napsal/-a:
> Hi Marek,
>
>> I have just two quick technical questions
>>
>> Am I right that contrary to gaiming/graphical applications, SLI is not
>> actually so important for
>> Amber parallel GPU calculations ( just deducing from this contribution
>> http://archive.ambermd.org/201106/0005.html )
>>
>> Or Am I wrong ?
>
> Correct. There is no way to use SLI for CUDA. Effectively SLI just allows
> two GPUs to rending in the same frame buffer - e.g. one GPU renders the
> top
> half of the screen and one renders the bottom. There is no way to send
> actual useful computing data over this link so it makes no difference
> whatsoever for general purpose computation.
>
>> Another thing.
>>
>> If one thinks also about AMD alternative to build multi-GPU system (
>> for
>> example "MSI 990FXA-GD80 - AMD 990FX" + 3 x GPU)
>> is it worth to wait a little for new AMD FX "Bulldozers" ? Based on
>> several discussions here I think, that speaking just
>> about Amber GPU calculations, difference in using "AMD Phenom II X6
>> 1100T
>> Black Edition" or new 6 or 8 cores Bulldozer
>> will be negligible. So only in case, that the machine will be used also
>> for non-GPU calculations ( for example MM-PBSA
>> analyses ) one can obtain some interesting gain using new AMD FXs so in
>> such a case it is worth to wait some last weeks ( hopefully :))
>> ) for these nice 32nm guys.
>>
>> Am I right ?
>
> Yes. Very well put. You are pretty much spot on. There may be some small
> improvement in parallel GPU runs if the CPU memory bandwidth is higher
> with
> the new Bulldozers but I have not looked at the actual specs for AMD
> chips
> in a while. Plus I have not actually tried any AMBER GPU calculations on
> machines powered with AMD chips - maybe others here have tried.
>
> So yes the real help is if you want this to be a total general purpose
> machine and want to run CPU calcs as well as GPU ones.
>
> All the best
> Ross
>
>
> /\
> \/
> |\oss Walker
>
> ---------------------------------------------------------
> | Assistant Research Professor |
> | San Diego Supercomputer Center |
> | Adjunct Assistant Professor |
> | Dept. of Chemistry and Biochemistry |
> | University of California San Diego |
> | NVIDIA Fellow |
> | http://www.rosswalker.co.uk | http://www.wmd-lab.org/ |
> | Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
> ---------------------------------------------------------
>
> Note: Electronic Mail is not secure, has no guarantee of delivery, may
> not
> be read every day, and should not be used for urgent or sensitive issues.
>
>
>
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
> __________ Informace od ESET NOD32 Antivirus, verze databaze 6465
> (20110915) __________
>
> Tuto zpravu proveril ESET NOD32 Antivirus.
>
> http://www.eset.cz
>
>
>
--
Tato zpráva byla vytvořena převratným poštovním klientem Opery:
http://www.opera.com/mail/
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Sep 15 2011 - 12:00:02 PDT
