My understanding was that the GPU use for computation was only possible with Nvidia GPU, as Nvidia developed special software for that.
Does this still hold or the fact of having Nvidia or ATI cards makes no difference which is the ideal solution.
I have been since the beginning a fan of first 3DFX graphic cards and later Nvidia all the way, so my devices are all Nvidia equipped as a standard. I tried the GPU (GTX 285) on Seti@Home shortly to see if it worked and the crunching power was impressive. With Fermi the performance should nearly double. I noted also that the power consumption, noise and heat also shot up in a sensible way.
However it is very important for crunchers to be clear on this issue when upgrading a Video card, as they may go to other projects which use the GPU.

WCG is planning on a CUDA project, but in general investigating possibilities for the more broader applying Open-CL, which my recent card acquisition does in addition to CUDA.

There are a number of different GPU projects that run outside of the WCG.
Present projects include, GPUGrid, Collatz, Milkyway, Folding@home, SETI and Einstein.
You can run these as well as WCG tasks, though Einstein will eat up a lot of CPU time!

Given the current situation with NVidia manufacturing and CUDA development, it is unlikely that there will be any CUDA projects here until April at the earliest.
I am not sure what the point would be in releasing an Open-CL project at this stage either. There would be no guarantee of good performance on NVidia’s next generation of top cards, due out around March.
The latest ATI (5000 series) cards do have current potential and ATI cards are now also recognised by Boinc.
So the question is two segregated card systems (ATI, NVidia) or one (Open-CL)?
If the WCG wants two systems, development for the newer ATI cards could be underway as we speak. If not (more likely, saying how busy they are with Betas and how reluctant the WCG have been to go down this road, and rightly so) then it cannot happen for at least 3 months.

By the way, that is the tip of the iceberg; the above situation is far more complicated, and who knows what impact the Tegra 2 (ARM9/ARM11/NVidia computer on a board) devices will eventually have? They will be in mobiles, TVs consoles and computers – done deals, and the next 18moths will probably see 1 Billon sold!

The most power efficient and system acceptable (no extra power cables) card I have looked at is the GT 240. It is also reasonably priced, runs Very Cool and Quiet!
Although it wont work on Milkyway (requires CC1.3 and the GT 240 is CC 1.2), it works perfectly on GPUGrid (7200 Boinc Points per day) and I don’t know of any reason why it would not work on any other project.
So, if you want a reasonably fast card that wont set you back too much and does not cost much to run (£72 per year crunching 24/7) it is a good choice. If you want an ATI card, get a 5000 series. If you want a top of the range NVidia card, save up until March/April.

Seems we have already passed another launch date. I refer to the arrival on March 1st of IE8, the anti-trust version, which allows you to have any popular other browser without any clash.
This will be followed, on 12th March, by Intel's launch of the new multicore CPU's.

If I could find a smiley with crossed fingers I would tell you what this has done for my ageing pewter