RV770-based products can be sold for as low as $150, and I'm talking about finished ones. ATI has delivered wtf-levels of performance at a very low price point. G92/G94 is history. Only UNICEF rescue-aid kind of pricing can sell it in today's market. RV770 wasn't designed to take on G200 either. It's become a policy for ATI not to develop large monolithic GPUs ever, instead develop powerful, cheap GPUs that can be used in dual-GPU boards to compete against monolithic high-end from NVIDIA, or single ones in high/mainstream boards. That scheme didn't quite take-off with RV670, apparently with RV770 it just worked.
(Not really a reply to your post, just my analisis of the situation)
I have to disagree with Ati's strategy. If we talk about actual products, I agree that Ati has delivered the better mainstream ones this round. But to answer your last sentence, RV770 just worked, because Nvidia failed in the first place. Had Nvidia used 55nm/GDDR5/256bit (lower prices) or simply had better yields (better prices again), or would have been able to clock the cards where they wanted (higher performance that would have made RV770 slow in comparison), RV770 wouldn't have been so appealing. Also IMO NONE of them has delivered a good high-end card. Yes ATi's X2 is FASTER, but I would never say BETTER for many reasons. There's a lot more to a card than pure performance, like power consumption, heat, etc.
That leads me to the point. We have talked about the chips, let's talk about architectures. Ati has talked so much about the benefits of multi-small-GPUs versus monolithic GPUs, but in the end what the final products show, for those of us that are not blind, is that the monolithic architecture has won this battle hands down. Ati chips had many advantages against Nvidia's ones in this particular battle:
1 - 55nm.
2 - GDDR5.
2b - 256 bit memory interface.
3- No CUDA support. Today that's a huge advantage. Nvidia chips have actual hardware which is specific to CUDA. Basically that's adding many transistors (a good chunk BTW -> full 64 bit computing...) that don't help increasing framerates. Because CUDA is not established yet, that's a dissadvantage, but we'll see in the future.
Even with those advantages and Nvidia's 40% yields fiasco, Ati has not really won this round, because Nvidia has been able to lower the prices in order to regain the performance/price lead and still make money. Well that's a bit of speculation on my part, but one of the main problems was those low yields and if 216 SP GTX260 is feasible now, yields have improved a lot for sure. Contrary to what many people think, PCB complexity is not such an issue anymore. First of all, things are expensive when they are new and scarce (and they are not now, that is) and secondly, dual 256bit designs, although not as complex as 512bit ones, it is also complex enough so that the difference gets smaller.
Nvidia already had the performance/watt one, something that doesn't help protraying Ati's multi-GPU strategy very well to say the least. In the next round, both Ati and Nvidia will play with the same weapons (40nm, GDDR5, low memory buses) and common sense says that Nvidia has many chances to win every front by a huge margin with the simple evolution of their architecture. Unless Ati does something a lot more revolutionary than evolve/refine the strategy they have today, they won't have anything to do against Nvidia's next gen IMHO. Ati is been lagging behind despite their fab process advantage and that's a big one of advantage. RV770 was the secret weapon and has won, but not to the point it should have in order to demostrate their strategy is the best one long term.