Well, maybe. I'm getting the standard "you need to buy a Quadro card" from NVidia so far. This does indeed look like a problem you might run into without Unified Back Buffer support, but I'm trying to convince them there's more to it than that. Without UBB, it's possible you can get your card into a super memory thrashing state if you open up a lot of double buffered applications, but it looks like NVidia's driver doesn't allow this state... if you requirest buffer space and you are out of VRAM it just denies acceleration alltogether, and when buffer space becomes available again it doesn't reallocate it to apps that need it so they keep running unaccelerated. The fact that this little virtual console switch can allow you to run a few more apps accelerated than were running before the switch shows that you really can squeeze more apps into VRAM if you "repack" the allocations, and so their normal allocation method is not being as efficient as it should.
Anybody have any tips or thoughts for or against this theory?