According to TSMC:
"The advantage of low-k dielectrics is significant for certain designs. Agere Systems, which recently announced that it is using TSMC?s low-k based 0.13-um process for its DSP16411 digital signal processor (DSP) chips, now in volume production, has reported a performance improvement of 20 percent and a power savings of another 20 percent compared to competitor chips without the low-K technology."
Linkage
20% increase in performance and 20% power savings.
ATI utilized low-k first and I think it was with the 9600's who's clocks speeds ramped up quite high as a result. Good call by ATI.
If you noticed, the X800XL was produced on the 110nm process BUT without low-K (as opposed to the rest of the x800series cards) hence its lower clocks and less than stellar o/c ability with some exceptions here and there as per our own forum members here.
Nvidia AFAIK, has never utilized low-k in their GPU's, so the G71 "may" be the first. 750MHz should be a cinch. Nv is getting 550+ at 110nm and 24 pipes on a normal process (non-low-k). Now shrink that to 90nm PLUS implementing low-k dielectric process, and we should have a Winnah!.
But, we shall see. Anything can happen.
Just for kicks: and not in any way scientifically proven
6800U - 130nm - non low-k - 400MHz (reference)
7800GTX - 110nm - non low-k 430MHz (reference)
7800512 - 110nm - non low-k 550MHz (reference and after refinement) very impressed.
So Nvidia/TSMC went from 400MHz 130nm 16 pipes to 550@110nm 24pipes. 150MHz increase.
G71 a.k.a 7900Ultra - 24 pipes - low-k and matured TSMC 90nm process. I don't think 150 to 200MHz increase is such a far cry.
Considering that 7800GTX512 owners seem to be getting around 580MHz with the stock coolers. I think its a shoe in. But, that's just me.
Take it as you will.
I'm more interested in what the architecture can do rather than how fast the core runs.
I speaketh of G80 of course.