I am working on a gpgpu project (Mostly mathematical computations).
I have a Geforce 7300GS and 7600GT on the lab.
can I use CUDA to perform computational operations on the GPU on this cards.
Nvidia say only from 8 series but I heard there is an option for some operations to use CUDA even without GeForce series 8.
is this true?
or should i try a different language
CUDA only works on GeForce 8. You may have heard about “emulation mode”, but emulation mode is much, much slower than the real thing because it runs your CUDA program on the CPU for testing purposes. (No GPU acceleration at all)
If you want to do GPU computing with a GeForce 7, you should take a look at BrookGPU:
In addition to what Seibert said, I’d recommend you at least give it a shot to try to get a small budget for an 8xxx card in the lab.
$200 8xxx cards are pretty powerful, and CUDA is v. nice IMO.
The academic justification for an 8xxx card may be that you have more tools to approach the problem, and that you could do some measurement and evaluation as to what’s the best hardware to use.
Thank you so much guys
your advice was really helpful.
since I can’t get the budget I’ll try either Sh or Brook.
can you tell me which one is better for my purposes?
and can I get a real acceleration with these?
I only briefly worked with Brook before switching to CUDA. It was pretty straightforward for simple problems, like array operations and reductions. I have no idea how it would work for more complicated operations.
This really depends on what you’re going to do on GPU.
Unfortunately I do not have pre-CUDA GPU background, but some forum visitors do. If you describe what kind of problem you’re trying to solve on GPUs I’m sure they will comment on this.
I’d go with Brook. Brook is a standard in GPU computing, while Sh seems to be nearly dead.