On the discrete note, Fermi isn't that bad of an architecture, it just took too long to get here. I have a 470 which can chew up anything I throw at it, and contrary to what most people tell me (including the Sales Associate who sold me the card - he tried to talk me out of it) it doesn't run supernova on air cooling.
Its just all too little too late. I was already well sick of waiting for the cards to debut months before they hit shelves, and while I was enthusiastic about getting some new hardware it kind of fell off the radar until the day I walked by it on the shelf. So while it outbenches the 5000 series in tessellation, it was just another upgrade. "w00t, I can plug two more monitors in!", not "w00t new card!".
I think NVIDIA needs to break back into the chipset business. Regardless of the flaming, ION is a pretty successful platform and integrated NVIDIA chipsets were not (always) total sh!t. I think it is going to be hard for the company to break in to AMD and Intel chipsets - Intel + NVIDIA is a bittersweet marriage between two companies which really don't like eachother but need to compete with the company that now does both (AMD) rather than real drive to get the competitive edge.
NVIDIA needs one of the following:
-Come out with an AmazingCard™ that blows everyone away before any competitor can react, let up on the rebranding
-Somehow, get back onto AMD and Intel boards. Buy Lucid or something, find a way in. I guarantee AMD + Intel are holding that door shut.
-Get their own x86 action going on, and graduate from a graphics chipset manufacturer to a chipset manufacturer.
I think #3 is their long term ticket to success, if they can release something well designed and competitive. Ultimately it is contributing the the stagnation of chip progress (x86 blows chunks) but it would probably save the company from the Voodoo fate. #2 would also work well if they can get AMD+Intel to open the floodgates, but #1 isn't happening with their current business model. Their life raft right now is the 460.
Maybe something else will happen, I don't know. Maybe OpenGL and DirectX will die in favor of raytracing, and then it wont matter how much anyone rebrands.