Performance differences for parallel computing with GTX 275, 280 and 285

Hi - I’m going to buy an NVIDIA GPU card so I can use CUDA. I am looking at the GTX 275, 280 and 285.

Will the extra money for the 285 be worth it in terms of performance difference? They all have the same # of processor cores, just different hertz for things. Will I see a large leap going from the 280 to 285, or from the 275 to the 280? Which card is the best price/performance pair?

Thanks for answering - this has probably already been asked. I just don’t want to pay $100+ extra for the 285 if the performance difference isn’t huge.

Even GTX295 will be outperformed shortly due to Moore’s law. I think you may need to consider your possible upgrades in the future and spend enough money for the backbone now, i.e. mainboard/rack. Buying a multi-PCIe-slot mainboard will let the GPU staying longer with you.

The cheapest card with 1.3 hardware (GLX260) may be a sufficient initial investment. From my testing result, GLX280 (30 MPs) is only 1.18x faster than GLX260 (the latest revision with 27 MPs), but their prices differ quite a lot.

I am happy with GLX260. Speed was not as important for me as the 1.3 Compute Copatibility as the software I am developping is ultimately not for me but for someone else.
Even the best hardware cannot outperform good algorithm…

Hi: on New egg the GTX 260 is about $175, the GTX 280 is about $265, the GTX 285 is about $350. What is the “GLX260”? I can’t find it. Do you mean GTX260?

The GTX260 has 192 processor cores, while the GTX280 has 240 processor cores. But, as you measured cvnguyen, this translates to only about 1.18 times as fast? In your opinion is this worth the extra $90 or so?

Is there general agreement that the GTX260 is probably the best bang for the buck, and that spending $90+ more on the 280 or $175+ more on the 285 really isn’t worth it?

There are in fact two variants of the GTX 260. The original GTX 260 has 192 cores, but shortly after release NVIDIA also came out with a 216 core version, usually called the “GTX 260 Core 216”. So you have an entire spectrum of options. The wikipedia page has a nice summary:

http://en.wikipedia.org/wiki/Nvidia_gpus#GeForce_200_Series

and now it looks like someone has even done the multiplication to figure out the peak GFLOPS for each card. That column along with the memory bandwidth column should be the two things you look at.

Sorry just a typo, I meant the GTX260. My GTX260 card, according to the retailer, is the latest edition with 27 multiprocessors (3 MPs more than the original edition, seibert is right). The higher cards (except GTX295) contain only 30 multiprocessors, so speed-up won’t be so amazing.

One more thing to consider is memory bandwidth and memory speed; GTX260 seems to have the worst memory. But this thing only affects global memory access, and I am not using global memory intensively.

The price tag is also dependent on the vendors. I prefer ASUS although there are cheaper brands such as Gigabyte or EVGA.