Choose a graphic card for GPU CUDA?

What exactly are you trying to achieve???
Are you just trying to learn to program using CUDA, or do you actually want to run large tasks???
big difference between Hello World vs Folding @ Home etc...

I have to develop some algorithms for computer vision. I have to learn, but it's for industrial applications so it's not just an "Hello World" program. I really have to implement complicate algorithms. These algorithms are slow in Matlab and C++ programming. So, we gonna try to speed up them.

Latest card with fastest GPU clock would be my guess, does CUDA work with sli? because that will also affect the card or cards chosen.

It should be way faster if you look at nvidia site instead of asking for advices here http://www.nvidia.com/object/tesla_computing_solutions....
Also look carefully at the additional info column, there're some important links there.

So, I've chosen a PNY 8800 GTX. I just want to say that there are 30 graphic card supporting Cuda. The choice is not that easy and this is why I was asking for an advice. Instead of saying to me "search on nvidia", try me be smarter : answer the question or do not reply!

I've never programmed using a GPU (I just do regular CPU programming   ) I did however do a parellel processing unit back at uni - one of the things is that if you do your program properly, it will scale.
Hence I would ASSUME (someone correct me if this is a wrong assumption) that a GT or a GTX would work the same, but the GTX would be quicker

Thanks Hassa, thanks for your answer  

I have read the CUDA specs and installed the package with the necessary drivers in the past. The ideal CUDA card is the GTX or the Ultra according to all I've read and an SLI setup of either is ultimate.

Find your CUDA literature and read it.