I have two almost identical GPU's on my system and BOINC will only select one of them. I understand you use certain criteria to establish what GPU to use, but would like for you to re-consider the logic.

Here is the analysis done by BOINC on my two cards (not very powerful by today's standards)

11/14/2009 7:17:12 PM NVIDIA GPU 0: GeForce 8400 GS (driver version 18635, CUDA version 2020, compute capability 1.1, 512MB, 29 GFLOPS peak)
11/14/2009 7:17:12 PM NVIDIA GPU 1 (not used): GeForce 8400 GS (driver version 18635, CUDA version 2020, compute capability 1.1, 256MB, 29 GFLOPS peak)

As you can see, the only difference between them is the amount of memory available and not their capability. Can I suggest a logic change to find the minimum capable card on the system, and then enable use of all similar or better GPU's. This would still keep from using less capable GPU's causing problems but allow call apable cards to be used, to the projects' advantage.

That only the best card is picked is a shortcoming of the Nvidia API, it's beyond BOINC's control at this point. You can use different methods to assure that cards are used, see the CUDA FAQ on that, section "OK, so now I disabled SLI and BOINC only recognizes one video card. What is up with that then?"

---