I think 7xxx series will buy them some time but they are still stuck with GCN. I have a feeling that AMD will needs to exceed the 400nm2 and jump into or near the 500nm range to compete with gk110, which might be against their design philosophies. If Gk104 performance indicates anything, we might get a gk110, that is 40-60% faster than a 7970; that is if gk104 is 10% faster in real world testing than the 7970 and the die size of gk110 is 60-70% bigger. GCN is AMD's new architecture and it is here to stay at least one more generation. It is called GraphicsCoreNext for a reason.

Increasing the frequency will help of course, but 1305mhz 7970 consume in excess of 300 watts and 62 more watts than a power hungry gtx 580. The biggest deterrent for high clocks is chip reliability over time and binning of course. Also there have been rumors of Keplar clocking quite, similar to what 7xxx has been doing so this won't help that much either. Considering the high clocks at stock voltage, i think this rumor is actually likely.

http://www.hardocp.com/article/2012/..._card_review/9

A 360mm2 high clocked a card would probably still lose to gk110. Losing the performance per watt/die size at the high end has alot of long term consequences. Unlike the lower end, they can't rip out unnecessary components for GPU compute because they still want to tackle this market. A big performance based pure gaming part(uncompromised gaming performance with no GPU compute features) doesn't make sense for any company. When your competitor has the more efficient architecture and is willing to make chips much bigger than you, it will be crazy hard to catch up. AMD has to either have another architecture altogether or be willing to make big ass chips like Nvidia. I would happy with scenario 2 but I don't think AMD will do this.

Instead AMD may be more content with second place again and I think is a more likely scenario, try to compete for the bang for your buck area and avoid the monolith. AMD's market share is much smaller than Nvidia's in the professional market and without the assurance of revenue coming from the professional market(as Nvidia is the industry standard here) there is much less incentive to make the monolith monster dies which funds these projects and cards.

Unless they pull a shark out of their ass or GK110 simply is a terrible products which loses everything that makes gk104 run well, 8970 could be very much slower than the GK110. If Fermi wasn't rushed and gtx 580 was the chip released initially, the 6970 would obviously still be slower even though it was second generation. From what it sounds like, with a product to sell, Nvidia is in no rush like Fermi to get gk110 on the market. Partners have something to sell in the form of gk104, gk107 and I am not sure about gk106. I can imagine Nv are likely trying to get it right the first time as their competitors don't even have their professional market products out and they would rather avoid another gtx 480 debacle(henced why gk100 was scrapped). I don't think anyone is sure what AMD will do at this point, but there is no simple solution and I have a feeling there is going to be a lot of compromises made by the company and hopefully price drops this generation.

All this speculation is based on a lot of rumors but I am also following the history of what both companies have done in the past and of course chip planning is done years in advance.