I read that there was a bug at a point in the nvidia drivers, where the system did not recognize a nvidia card as CUDA device unless a screen was connected (be it only a dummy DVI) but that it should work, only PhysiX isn´t working (thanks for that nvidia -.- )
As the whole GF100+ GPU GeForce generation absolutely blows in terms of OpenGL performance I was thinking on getting a HD6970 for display and use GF100 GPUs in the same system for Octane.
And yeh, I know of the alternatives.
(o) I could use a quadro for display - I refuse to buy one as Nvidia obviously thinks they need to cripple OpenGL in GeForce cards deliberately, denying any problems and ignoring support tickets to force people into buying quadros. Thanks for nothing nvidia.
(o) I could simply put the GF100 for CUDA in one of my other systems - well, to me its inconvinient, I rather have it all in one machine, but I guess it´s the only way I currently see.
(o) I could use a G92 series GPU for display along with GF100+ GPUs - I am also a passionate player and the G92 is simply too weak, again a matter of convinience having all in one machine.
So yes, it´s basically I creating my own problem, and I know I need to point it out, else one tends to get good advice rather than an answer to the initial question

One other thing, anyone here with a GF110 (GTX570/580) can report if the OpenGL issues also persist with those GPUs?
I´d be glad if buying a few of those would solve my problem as the GF110 is quite superior to the AMD Cayman besides certain AA modi and adaptive power management.