That doesn't seem like much of an "issue" as it were. I mean, having two cards totally dedicated to the render plus significant boost from the third, except that Windows experiences the "jitters". When I'm working on a project and using all of my resources, I'm so excited about how the render is progressing, I say, "To hell with what Windows needs!"My three cards are identical (560 ti) so I either leave one of them free to drive the OS or if I want the extra speed boost then I use all three. Obviously if I were to use all three to render then I would once again get the problem with some of the memory being eaten by windows.

Which brings up a question...Is it possible to de-allocate a CUDA device in the middle of a render? That would be nice, if you want all of the resources spent on the render, with the option of taking a card out of the loop to let it concentrate on Windows.