Hi Raik,
i7 will be much
better than some middle class Xenon's. The speed of the CPU, and of course many
other factors, are important for the GPU scale. For instance quad core Xenon
2.66Ghz and GTX580 3GB perform 48ns/day on JAC against more than 57ns/day with i72600K.4.6Ghz.
That's the reality..
If I am not
wrong NAMD uses only double precision, thus Tesla will be better but why to
spend money when you can use Amber. Moreover the NAMD GPU version is somewhat
basic and Amber GPU is much much better!
My advice at the
moment is i7+few GTX580 3GB. In fact a good configuration is listed on the
Amber website:
http://ambermd.org/gpus/
Paragraph: Building
your own System
Keep in mined
that Kepler will be realized after few months.
All the best
________________________________
From: Aron Broom <broomsday.gmail.com>
To: AMBER Mailing List <amber.ambermd.org>
Sent: Wednesday, January 11, 2012 12:39 AM
Subject: Re: [AMBER] i7 extreme for Amber GPU computing?
I've used the M2070 cards with the Xeon 6-core processors, and also the
GTX570/580 with 2-core intel or 6-core AMD. If you are wanting to
potentially run NAMD also, then I would highly suggest going with the
6-core CPU (I didn't see an advantage for the intel over the AMD in my
benchmarks), but if you are just doing AMBER then keep in mind that AMBER
under GPU acceleration only uses 1 cpu per gpu.
Also, you can see from the AMBER benchmarks page that additional GPUs don't
scale all that well past 2 GPUs. In particular you need to consider your
motherboard, if you have a single 16x and putting two cards in reduces each
to 8x, then it really isn't worth your money to put two cards in. I'd
suggest either making sure you get a motherboard with 2 separate 16x slots,
or just build multiple cheaper systems (unless there is some massive system
you need to simulate).
In terms of the expensive M2070 cards versus the GTX570 or 580 (~1/5 the
price), the GTX cards are faster because the memory clocks are actually
higher. The main draw-back is memory. For NAMD this isn't much of an
issue as its memory usage is rather small on the GPU, but for AMBER you
want to make sure you have enough memory. You can buy 3GB GTX580 cards
though. There are some guidelines on the amber page concerning how much
memory will allow you to simulate a certain system size, so unless you need
more than 3GB (the M2070s have 6GB for instance) then I would save the cash
and go for the cheaper and faster GTX580. Of course there is also the ECC
on the M2070, but (and if others have a different opinion I'd love to hear
it) if you are talking about molecular dynamics only, then the occasional
extremely rare error isn't going to be an issue at all.
That's my two cents,
~Aron
On Tue, Jan 10, 2012 at 3:34 PM, Raik Grünberg <raik.gruenberg.gmail.com>wrote:
> Hi there,
>
> I am trying to configure a new desktop / workstation for GPU
> accelerated Amber simulations. The top-notch high-end configuration
> seems to be based on dual Xeon with two or four NVidia Tesla. Does
> anyone have experience with a system built around the (cheaper but
> supposedly faster) Intel i7 extreme processor?
>
> Thanks in advance!
> Raik
>
> --
> ___________________________________
> Raik Grünberg
> http://www.raiks.de/contact.html
> ___________________________________
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
--
Aron Broom M.Sc
PhD Student
Department of Chemistry
University of Waterloo
_______________________________________________
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 Tue Jan 10 2012 - 15:30:02 PST
