Re: Recent Geforce Cuda GPU cards(GT 440, GTX 560ti)

From: Axel Kohlmeyer (akohlmey_at_gmail.com)
Date: Wed Apr 13 2011 - 09:25:58 CDT

On Wed, Apr 13, 2011 at 9:21 AM, Burgess, Don E <deburgess_at_uky.edu> wrote:
> I am interested in using cuda-enabled VMD to analyze trajectories from MD simulations.
>
> Has any one used these Geforce cards (GT 440, GTX-560ti) with either VMD or NAMD?

not yet. i have a GTX 560Ti sitting on my desk and waiting to be tested.
last time i checked, there was only an obscure beta driver that supported the
card, so i am waiting for a proper driver, which will probably coincide with the
full release of cuda 4.0, that would require a driver update in any case.

> How does the performance of a GTX-560ti compare to a GTX-580 (or a tesla c2050)?

you have to distinguish between compute bound tasks and memory
bandwidth bound tasks. classical (bio) MD with neighbor lists and
simple potentials is typically more bound by memory bandwidth than
speed and number of GPU cores. so i would expect the GTX 560Ti
to do about as well as a C2050 on MD like tasks, especially when
you operate the C2050 with ECC enabled (which is an important
reason to get one). things will get more complicated if you have
applications that can take advantage of large GPU memory and
are optimized tuned specifically for fermi type cards.

cheers,
    axel.

> Thank you for your comments.
>
>

-- 
Dr. Axel Kohlmeyer
akohlmey_at_gmail.com  http://goo.gl/1wk0
Institute for Computational Molecular Science
Temple University, Philadelphia PA, USA.

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:56:58 CST