if there are no legal reasons preventing me from doing so, I am going to modifiy the Mandelbrot SDK sample to run on 2-4 GPUs in parallel - source code will be posted here (given that I am successful, of course). A simple tiling of the coordinate work space should be possible, followed by copying all results into a texture on the primary graphics card for display. Maybe the copy operation can be overlapped with computation as shown in some other SDK sample.
The reason for this is that I want to get some hands on experience with CUDA and the best way of doing so is probably to modify the SDK samples.
The hardware (MSI K9A2 Platinum mainboard + 2 GPUs) is already ordered and if feasible, I’ll add 2 more GPUs. These are inexpensive Geforce 9600GSO boards with 96 shader units each. Pretty much bang for the buck! I am only worried about heat development in my PC casing - maybe I’ll have to leave it opened. I might also use the PC for folding@home. Overall the costs will stay below 700 Euros.