Max OC to MAX OC the 7970 is faster than 680 but reference for reference the 680 is faster than 7970 because of the higher reference clock speed of the 680. Clock for Clock the 7970 is faster than the 680. 670 is slower than both 680 and 7970 end of story LOL. PS it's not correct to say a 670 is faster than 680 when the 670 you are referring to is OCed and the 680 is @ reference clock speeds LOL. OC to OC and stock to stock is fair and correct.
I think I see where you are coming from but trying to compare performances based on clock speeds is only indirectly relevant between cards. It's the same as comparing an FX4100@4GHz to a 2500k@4GHz and expecting the same performance. Even comparing the same card made by different 3rd party manufacturers doesn't often produce the results you expect.
Take a look again at that chart:
The 680 is the reference design which sports a 1006-1058 core clock including turbo.
The 7970 is the gigabyte OC design which comes with a 1000MHz clock.
So based on those figures you would expect the 680 to be 0-6% faster than the 7970 yet the results show the opposite trend with the 7970 performing noticeably faster in 4/7 games and the 680 is better in only 1.
Secondly the 670 that TTL tested was the reference design which had a core clock of 915-980MHz so you would expect the 680 to be 9-10% faster which it really isn't.
Thirdly the 670 is far from being categorically slower than both cards...
It maybe that technically the 680 is a faster card. After all it has a greater clock speed (at stock) more cores etc but these are just numbers and words.
For whatever reason, whether games are unable to take advantage of the extra power, whether the cards are bottlenecked by something or even that those advantages actually do nothing at all - regardless we see that in real world games (where it actually matters) we just aren't seeing what we expected.
We certainly aren't seeing anything where the performance even remotely matches the extra cost over the other cards.
