Quote Originally Posted by Chumbucket843 View Post
fermi is fine. almost every chip in the last 10 years has been designed for process scaling. i wouldnt be surprised if they could get a 1024sp fermi on 28nm. they may do an Si spin and get a considerable improvement on 40nm too. i think GTC will reveal what they are planning.

AMD's losses were not only from inferior products but buying ATi and the TLB bug fiasco. nvidia doesnt have those issues.
Whats to stop AMD from doubling up with SI or NI. If they can get more performance for less space, they will keep on getting better and better each generation.

Nvidia needs to get more performance out of the transistors they have and keep the size the same or lower. The simplest way to do this would be to up that shader clock if they can. It was originally rumors thought that this card was going to have a shader clock between 1600-2000mhz. If they can keep the core clock down while increasing the shader clock, this architecture will start to have legs.

The problem with this generation compared to the prior is that they removed the MUL operation which supposedly would not have a drop in performance(it was found in the gtx 280), however it actually did(or drivers still havent reached maturity). If the gtx 295 was clock like a fermi card, it would almost certainly be faster. Also the gtx 480 loses pretty soundly to SLI gtx 285. Per transister, fermi is worse than the gtx 280, which is pretty bad considering it is a new architecture.

The only thing I can think of to turn fermi around at this point is get the power down, up the shader clock and get those original TMU reenabled. Fermi needs 25% more performance at least to be considered a success and to justify its power consumption.