[Beowulf] crunch per kilowatt: GPU vs. CPU

Gus Correa gus at ldeo.columbia.edu
Tue May 19 12:36:18 PDT 2009

Joe Landman wrote:
> Nifty Tom Mitchell wrote:
>> On Mon, May 18, 2009 at 01:35:43PM -0400, Joe Landman wrote:
>>> David Mathog wrote:
>>>> Although the folks now using CUDA are likely most interested in crunch
>>>> per unit time (time efficiency), perhaps some of you have measurements
>>>> and can comment on the energy efficiency of GPU vs. CPU computing?  
>>>> That
>>>> is, which uses the fewest kilowatts per unit of computation.  My guess
>>> Using theoretical rather than "actual" performance, unless you get 
>>> the  same code doing the same computation on both units:
>> Another issue is the upgrade cost and upgrade potential.  Looking across
>> my room I see an older AMD-64bit single core Athlon box.   It would be
>> less expensive and much less of a task to swap out and upgrade the GFX
>> card than it would be to update the processor.   In some cases multiple
>> GFX cards are also possible.
>> Also it makes little sense for me to upgrade the GFX card based on
>> graphics display needs only.   If I toss CUDA in the mix a high end
>> GFX card upgrade does get interesting.
> What makes this very interesting is that you can get reasonable CUDA 
> performance for a number of apps for ~$500.  Add in the powersupply you 
> will likely need ($300) to support more than one card ...
> ... the cost benefit analysis just keeps getting better ... *if* your 
> code can use it.  There is no guarantee on this, but it is worth 
> considering.
Hi Joe, Tom, list

Assuming the code is CUDA-friendly,
is it worth the investment ( $GPU_card + $power_supply)
to GPU-revamp a 2-3 year old system,
keeping a motherboard with PCIe 1.0 slots and bus?

Gus Correa
Gustavo Correa
Lamont-Doherty Earth Observatory - Columbia University
Palisades, NY, 10964-8000 - USA

More information about the Beowulf mailing list