Recommendations of GPU for cuda programming

[phantom7026]

Hello~

I want to build a machine for cuda programming.
Can you recommend gpu card?
Since the budget is not enough, I cannot afford to buy tesla .

Originally, I considered GTX 590 just because it is the best model.
Now, I’m not sure it is the best, I mean a SLI combination of the lower model may be the better or not. I don’t know.

I can pay around $1500 for gpu.

Which is the best option with the budget ?

[melonakos]

A couple of comments:

• SLI is not applicable to GPU computing efforts, so forget about that - it doesn’t matter.
• GeForce is the cheapest way to go
• At $1500, you might be able to get a Tesla C2050. There are used ones on Amazon for $1359: http://www.amazon.com/s/ref=nb_sb_noss?url=search-alias%3Daps&field-keywords=tesla+c2050
• Memory size is the main reason to go to Tesla. But there are other reasons listed here: The Official NVIDIA Forums | NVIDIA

Shameless plug: To avoid software costs and headache, you can use the free version of ArrayFire.

Good luck!

[pasoleatis]

If you go with a GTX you need to keep in mind something regarding the compute capability. If you plan to program on your card and later use them on Tesla cards you should buy just a card with compute capability 2.0. The cards with capability 2.1 have 48 core per one SMP while the 2.0 has 32. Since a warp has 32 threads this means that a program optimize for 2.1 will not run optimal in a 2.0 card.

[seibert]

The GTX 570 or 580 are excellent choices, and cost substantially less than the Tesla. If you need 3 GB of memory, the GTX 580 has a 3 GB version.