Dear jason and filip,
Thanks for your review.
regards
setyanto
On Mon, May 16, 2011 at 8:47 PM, filip fratev <filipfratev.yahoo.com> wrote:
> Hi Setyanto,
>
> We have already discussed about GTX590 and I will use this message to give
> some additional information and details. I have two of these cards.
> My initial benchmarks were not so correct because it turned out that
> (during my tests) when I use more than one GTX590 the cards work with
> limited speed of 553Mhz, i.e. they reach only to the performance level 2.
> With simple words my tests were performed with downclocked GTX590's. At
> full speed of 612Mhz I obtained:
>
> DHFR NPT benchmark
> GTX590 1 core 22.05 ns/day
> GTX590 2 cores 31.89 ns/day
> GTX470 18.82 ns/day
> GTX580 about 28 ns/day
>
> As you can see from these values indeed the GTX590 is faster than GTX580
> but if you use two GTX580's you will probably reach 40 ns/day against 36
> with two GTX590.
> For the more realistic case:
>
> FactorIX benchmark
> GTX590 1 core 4.55 ns/day
> GTX590 2 cores 7.26 ns/day
> GTX470 3.65 ns/day
> GTX580 about 5.5 ns/day
>
> Here I was a bit surprised and enjoy because as you can see the difference
> between 2 GTX580 and 1 GTX590 is not so big (about 8 ns/day against 7.3) and
> GTX590 is really about 100% faster than GTX470. However again, if you use 4
> GTX580 they will be faster than two GTX590's.
>
> I was not able to solve the downclock problem with my GTX590's till now. To
> be honest the Nvidia support was not great too. They supposed that it is
> just a PSU issue but after many tests both my and their results showed that
> indeed the PSU's are not the problem and it is "something" in my systems,
> but never received any suggestions what could be and how to solve the
> problem.. I think it is just a driver problem, because the cards work great
> with the older drivers but ..not for CUDA applications. Anyway...
> I also have to mention that GTX590 works very good for QM and some docking
> applications. For instance for QM calculations two GTX590's are really great
> performer and even clocked at 553Mhz they are much faster then pair of
> GTX580's.
>
> So if you plan to use many cards and only for MD calculations then GTX580
> is definitely the better choice, but if you plan to use only a single card
> then GTX590 will perform better. Also if you plan to run multiple jobs at
> reasonable speed, with GTX590's you can safe PCI slots. Personally I use
> GTX590's and GTX470's and will upgrade the latter's to GTX580's.
>
> Regards,
> Filip
>
> P.S. According to the rumors/news a new revision of GTX590 is expected in
> June, which hopefully will solve some of the power issues.
>
>
>
> --- On Mon, 5/16/11, setyanto md <stwahyudi.md.gmail.com> wrote:
>
> > From: setyanto md <stwahyudi.md.gmail.com>
> > Subject: [AMBER] need suggestion about NVIDIA GTX 590
> > To: "AMBER Mailing List" <amber.ambermd.org>
> > Date: Monday, May 16, 2011, 5:34 AM
> > Dear Amber user and developer,
> >
> > Our laboratory will prepare to build GPU based Computer for
> > computation. We
> > are consider to use GTX 590 since according to nvidia its
> > has 1024 core.
> >
> > But, when I see in this review
> > http://forums.nvidia.com/index.php?showtopic=196244
> > NVIDIA technical support
> > give warning about overclocking. It can damage the card.
> >
> > My question:
> > 1. If we are using PMEMD simulation for long time runs, is
> > it overclocking
> > or not ? so can damage the card according NIVDIA technical
> > support.
> > 2. Can we use GTX 590 for PMEMD simulation using Amber 11
> > ?
> > 3. Do we need extra cooling for this card ?
> > 4. Are there anybody here have experience with GTX 590 ?
> > and what is the
> > recommendation ? good or not ?
> >
> > Thank you very much for your attention
> >
> > regards
> > Setyanto Tri Wahyudi
> > Institut Teknologi Bandung
> > _______________________________________________
> > 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
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed May 18 2011 - 04:00:02 PDT
