Re: [AMBER] is there a difference between the GTX 680 GPU's

From: Jason Swails <jason.swails.gmail.com>
Date: Thu, 29 Nov 2012 18:19:58 -0500

On Thu, Nov 29, 2012 at 5:03 PM, Jonathan Gough
<jonathan.d.gough.gmail.com>wrote:

> Hi All,
>
> I was wondering in terms of performance/accuracy/precision is there any
> difference between the different EVGA Nvidia GTX 680 GPU's?
>
> I thought I remember Jason saying you didn't want the overclocked version,
> but I couldn't figure out what the differences were between certain models
>
> specifically I am referring to:
>
> EVGA 04G-P4-2686-KR GeForce GTX 680 w/ Backplate 4GB 256-bit GDDR5 PCI
> Express 3.0 x16 HDCP Ready SLI Support Video Card
>

According to Newegg.com, this card has (at $519):

1536 CUDA cores
4 GB 256-bit GDDR5
Core clock at 1019 MHz
Boost Clock at 1084 MHz
Effective Memory clock at 6008 MHz


>
> EVGA 04G-P4-3687-KR GeForce GTX 680 FTW+ w/Backplate 4GB 256-bit GDDR5 PCI
> Express 3.0 x16 HDCP Ready SLI Support Video Card
>

Also according to Newegg, this card has (at $549):
1536 CUDA cores
4 GB 256-bit GDDR5
Core clock at 1084 MHz
Boost Clock at 1150 MHz
Effective Memory clock at 6008 MHz

I think all of our applications (i.e., computational science programs) are
memory-bound. That is, the main determinant of speed these days is how
quickly the processor can be fed data to crunch. (This is mostly why the
GPU can run programs so much faster*, as I've been led to understand -- it
has much more memory bandwidth than CPUs). Therefore, I expect the 6% jump
in core clock speeds is all but meaningless to pmemd.cuda. You would
probably not be able to measure a statistically significant difference in
performance between them.

The far more significant difference is that the second card comes with
Assassin's Creed 3 and Borderlands...

HTH,
Jason

-- 
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Nov 29 2012 - 15:30:02 PST
Custom Search