While AMBER will run on any Compute 1.2+ GPU, there is *no* official support for Ge Force cards. None whatsoever. AMBER is officially a Tesla-only app. And while they are very similar cards on the surface except for the significant price difference, the Tesla cards have undergone far more rigorous QA w/r to running CUDA apps for extended periods of time, a qualitatively different use of them as opposed to rendering stuff at 30+ fps.
GTX590 MPI scaling should flat out suck. The end. Just don't go there. If you must, buy two GTX 580s.
Finally, if you're going to do development, do yourself a favor and buy a Tesla card for it. You'll *never* debug the code without one because you'll never know whether the bug is you or some effect of the lesser QA for Ge Force. Of course, if you are a slave master and like to see your postdocs and grad students suffer because it builds character and all that I can think of no better way to achieve this. Might I suggest a 400W or less power supply along with a 14" or smaller VGA CRT to complete such a debugging h311 ensemble? Just kidding...
-----Original Message-----
From: filip fratev [mailto:filipfratev.yahoo.com]
Sent: Wednesday, March 30, 2011 15:17
To: AMBER Mailing List
Subject: Re: [AMBER] Amber CUDA calcualtion on GeForce GTX 590 ?
Hi Marek and Ross,
I am a bit confused. GTX590 is roughly 1.5 times faster than GTX580. Moreover the individual GPUs scale very well (what I saw till now) between each other. I think that in the definition 1.5 was included the internal SLI factor too. On the other hand, from the explanations till now here seems that actually we speak about two individual not one cards i.e. what’s the point - scaling. If we spoke about some different applications I would definitely answer you Marek that there is no GTX580 model that will be faster than GTX590. Secondly if you run one job, ok in parallel, we will have again totally 3GB and you can run XXX thousands atoms simulation.
Thus my question here is probably for Ross:
1) How you expect the individual GPUs of GTX590 to scale in an Amber job?
I can not believe that if now two individual C2050 scales 1.4 it will be the same in the case of GTX590 cores. I don’t want to believe too that buying GTX590 we only safe one more slot when using Amber.
2) My second question is: how actually to run GTX590, using cuda.MPI or? Here I am a bit confused too because if I am not wrong I saw Amber tests on GTX295 before parallel Cuda version of Amber to be released.
Regards,
Filip
P.S. Marek, did you receive any answers from Nvidia about the Linux driver? Thanks in advance!
--- On Wed, 3/30/11, Marek Maly <marek.maly.ujep.cz> wrote:
> From: Marek Maly <marek.maly.ujep.cz>
> Subject: Re: [AMBER] Amber CUDA calcualtion on GeForce GTX 590 ?
> To: "AMBER Mailing List" <amber.ambermd.org>
> Date: Wednesday, March 30, 2011, 10:26 PM
> OK,
>
> so if one takes care more about the memory (for simulation
> of big systems),
> the "Gainward GeForce GTX 580 Phantom 3GB" is probably
> definitely the best
> choice regarding GTX cards available on the market because
>
> #1
> It has 3GB memory
>
> #2
> It is moreover a little bit faster than regular GTX 580 and
> so even more
> faster than GTX 590.
>
> #3
> Probably no problems with linux drivers since it should
> work with the
> latest driver (260.19.44)
> which supports regular GTX 580.
>
> Am I right ?
>
> Marek
>
> PS: Which is actual status regarding citation of your
> paramfit routine ?
>
>
>
>
>
>
>
>
> Dne Wed, 30 Mar 2011 21:21:54 +0200 Ross Walker <ross.rosswalker.co.uk>
>
> napsal/-a:
>
> > Hi Marek,
> >
> >> Just the additional question. How it is with
> memory of that
> >> dual card. I mean if I use the GTX 590 just for
> one single job
> >> (so if I understood well just one half of this
> card will be used = 512
> >> cores)
> >> will be for this job available also just half of
> the total memory (0.5
> >> x
> >> 3GB = 1.5GB)
> >> or the memory is shared with both separate parts
> and that job will have
> >> in
> >> disposal full 3GB ?
> >
> > No unfortunately. My understanding, although I have
> not actually tried
> > one
> > of these cards is that they really are just two
> completely self-contained
> > GPUs mounted side by side. There is no sharing of
> memory other than
> > through
> > the attached PCI-E connection which they share. As
> such if the card's
> > total
> > memory is 3GB then each GPU gets 1.5GB of memory
> available to it. There
> > is
> > no possibility for the other GPU to use the other GPUs
> memory. Hence
> > really
> > consider this to be two 1.5GB underclocked GTX580
> cards sitting in PCI-E
> > x8
> > slots.
> >
> > 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 6000
> > (20110330) __________
> >
> > 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
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
-----------------------------------------------------------------------------------
This email message is for the sole use of the intended recipient(s) and may contain
confidential information. Any unauthorized review, use, disclosure or distribution
is prohibited. If you are not the intended recipient, please contact the sender by
reply email and destroy all copies of the original message.
-----------------------------------------------------------------------------------
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Mar 31 2011 - 13:00:03 PDT
