I think a lot of people are looking at this card all wrong. When you actually think about things, this quite is much better than a lot of people are thinking, or atleast, AMD's strategy with this card is much more commendable. Few points to consider:
1. When the 5870 first came out, it wasn't an amazing overclocker. Reference card got about 8%, and as it matured and non reference cards came out, average was about 15-18%. The
reference 7970 cards seem to be getting about 25% overclocks and the overclock scaling as far as performance goes is quite a bit better than the 5870. So once more mature cards and non reference cooling comes out, I'd expect to see that go up to 30, even 35%. The people overclocking the card said,
"Without any voltage mods AMD says that enthusiasts and overclockers should be able to hit 1.2GHz and if someone wanted to do some voltage mods that they might be able to able to hit 1.3 GHz!"
1.3 GHz would be a massive OC of 40%, and I'd say with voltage mods and a decent MSI or Asus cooler, 1.25 GHz is really gonna be the norm, which, as I said, would be an OC of 35%. The highest I ever really see GTX 580's go is about 15-20% OC's, so assuming both are at maximum OC, the 7970 is in actuality about 30% faster than the GTX 580, assuming similar OC scaling, which it seems to be that way.
2. The 7970 has lower idle, and load power consumption than the GTX 580, and while that doesn't matter to most enthusiasts, it is worth mentioning. Idle noise is basically quiet, and load noise could be better, I'd definitely agree to that. Fortunately, Asus and MSI non-reference models are usually quite a bit quieter, but the same could be said for the GTX 580 and so on, so we'll call that even.
3. Almost no new games that are even remotely demanding are being released until
atleast 2013 when the new consoles come out. BF3, Skyrim, MW3, many of these games really aren't all that demanding. If AMD really wanted to go after 50% faster than the GTX 580, I'm sure they could have by souping up a VLIW4 architecture. But when a GTX 580 is pulling 80 FPS in Skyrim, does it really matter all that much if your 7970 is doing 120 FPS? No, of course not, and I like the fact AMD recognized this fact. So instead of maxing out pure pixel-pushing performance, AMD took a play out of Nvidia's book and added a lot of new features onto this card to make it more universally useful. Specifically dealing with compute performance, this card actually has a decent bit of die space dedicated to that, where as in the VLIW4 architecture, compute performance was an afterthought.
Specifically here, the GTX 580 is over 69% faster than the 6970 in this compute performance benchmark. Generally speaking, the GTX 580 is only about 15% faster than the 6970, so having such a big performance gap is much more damaging to the card than an extra FPS here and there. And so with the 7970:
Same benchmark, wildly different result. The 7970 is 58% faster than the 6970, and 12% faster than the GTX 580. The 7970 is usually about 40% faster than the 6970, so thats a pretty tangible additional performance increase in compute, compared to what would have been an irrelevant few extra FPS in Skyrim.
4. Similar to the point above, AMD diversified this cards performance by making tessellation so much better as well.
The 7970 gets a gain of 56% better tessellation performance over the overall 40% better performance it has over the 6970. Tessellation is becoming a more and more used feature in games, and so AMD caught on to this and made this card much more tessellation capable compared to the 6970.
5. As we continue to go down the nodes, performance returns will continue to shrink a little. The 5870 was about 50-60% faster than the 4870 on release and that was considered a massive gain and a great new architecture. The 7970 is about 40% faster than the 6970, so obviously a little less. Some of that can be contributed to decreasing returns on node shrinks, but mostly I'd say this architecture just isnt as raw performance driven, because it doesn't need to be given the rather undemanding games of today. If you really want to look at details, technically the 5870 was the last "new" architecture and die shrink, so you could compared the 7970 to that (where it is 50-60% faster) and the refresh of the 7970, say the 8970 to the 6970. But lets be honest, everybody is comparing the 7970 to the GTX 580 so it really doesn't matter.
6. Yes, the price is relatively high. But what do you expect? AMD needs money, they have no real competition right now, and so they really
need to charge all the market can bare for this series. Intel on the other hand, who has even less competition in the CPU department, has plenty of money and so can afford to not maximize profits and sell their CPU's relatively cheap. This is also a enthusiast level GPU, its supposed to not have great price/performance. If you want cheap GPU that can max most games, just buy a GTX 560Ti, but if that's you, this card isn't meant for you, so don't judge it.
7. New architectures always gain the most from new and improved drivers. We are comparing the 6970 on mature drivers, to the GTX 580 on mature drivers, to the 7970 on new drivers. You can always factor in about 5-10% better performance across the board when a few driver revisions come out.