Latest drivers aren't always the best or most stable, so you may have to do a lot of research to find what does for your system. I have found that even Microsoft's drivers are now capable of crunching most projects or at least on the older cards, so the above statement on stock drivers isn't always true. Considering I just picked up a first gen nVidia GPU card for $4.99 after MIR (US), even people without the up front money can get in the game. However, we haven't seen what WCG has in store for us, so no idea how low you can truly go at this point.

On a side note, for those interested in protein sciences, POEM@Home announced on their front page that they are finally releasing a Linux version of their GPU wu's. Evidently Windows compatibility will come after they get Linux ironed out.

@RT, it is possible to get some experience with GPU crunching (and to make a scientific contribution) on very limited graphics cards.

As mentioned above, POEM at home recently made GPU applications available for protein folding. I am running this under Linux on a netbook with an AMD E-350 fusion chip. I believe that right now the applications for ATI cards are working under both Linux and Windows (can't vouch for the Windows version myself though), but that the CUDA version for Nvidia cards is still having issues.

Albert at home (testing project for Einstein at home) has applications for ATI and Nvidia cards under both Windows and Linux. I am successfully running these under Linux on an Nvidia ION as well as on the AMD netbook.

I run GPUgrid on my Nvidia GT 430 card (low end of the GT400 series).

Right now I have only cheap, slow GPUs that aren't recommended for crunching, but I have them for other purposes and don't think they should be idle. I've learned a lot about graphics drivers, BOINC, and GPUs by using them, so I'll be all ready to go when WCG's beta test starts. And I'll have a better idea of what I'm doing if/when I update my hardware for more powerful crunching.

Thank You kateiacy! I need to follow in your footsteps on this. I really don't care enough about any other distributed computing projects (other than WCG) but it does seem that it might be a good learning tool and I do have some graphics card capability. In another week+ I should complete 100 years of HCC and will then take the opportunity to mess about a bit with this unless something else in this circus-side-show of a life of mine happens.

I just put two X6 boxes on line, one a 1090T and one an FX-6100 and BOTH motherboards support MULTIPLE gpu's!!! That means it may be better to leave the good one elsewhere and put TWO gpu's, of the exact same model, in the newer machine. Just a thought. In my case I took 4 dual core machines off line and brought on line two 6 core machines, for a net gain of 4 cores! But all the gpu's out of those 4 machines went into the two X6 machines, so no loss of gpu power, each dual core had one AMD 5770 gpu.

I agree