I think we all want this shrink to trend like all recent TSMC node process changes, where chip reduction improved price per chip and that got passed along to use consumers'. This time TSMC said, "Sorry this time no real price adjustment". That basically wiped the normal pricing advantage of moving to a new shrink. I think Nvidia might have a slight advantage in GK104 die-size performance, but wonder as to the clocks they can garner from cutting out entire graphics processing cluster that provides the advantages from the more challenging HG HkMG process. AMD stuck with a straightforward 28Nm LP production, and prices seem high at first, but as we now see with the 7950/7970 there’s room to wiggle, once Nvidia has cards to lay out.
We might see a GTX660(Ti) like a 7850 (who knows) and asking $270 upon release, I don’t see many of that variant encroaching into the 7870 it will bee a volume leader and pricing will be agressive.
The real question is where, what, and how many GK104 can Nvidia mix in the $300-450 opening? Sound like they have a GTX670 that’s pretty much a full chip disabling only one SMX units, but I’m doubt with supplying it with “boost clock”. The question I keep trying to get a consensus on is can Nvidia provide the components and PCB to offering dynamic clock support at say $400? I see just them going with traditional approach; holding to 950 MHz top FTW limit, because I think they’d likely want to hold to two 6-pins. While another scenario is they skip the "Ti" on the GTX 660 (discussed above) and disable some other sections (more SMX units?) and insert another GTX660 that has the Ti? I almost think they need 4 variants to cover the spread and get full use of GK104 yields.