@tacojohn: Known issue, talked about a few pages back. Blame CUDA 5.5 and just run 1 instance per card.


If you don't mind being half as fast as a 660  


Edit: I'm up there in the VIP section with cbuchner1 now: