i wouldn't be surprised if its a case of taking the end of each scale, as in in the best situations these numbers look about right but in the worst of cases Charlie's numbers are more correct.
Once you factor in newer drivers than when those tests were done, and some pretty nice performance improvements since and apparently some great ones coming in 10.3, I'd think the 470gtx isn't on average 5% ahead right now. Charlies "source" said that was the average performance, not worst case, and the best case was only in one benchmark. Its very likely we'll come across a game the 5970 sucks badly in crossfire with, with crossfire basically not working or even worsening scores and seeing the 480gtx compared in that game. In reality, I wouldn't have expected it to be more than 30% faster in real world use in Nvidias best games, much like the 285gtx was now and then. Considering its missed its clock targets by 20-25%, I'd be unsurprised to see it below in some games, ahead in others and average of 5-10% ahead.
Rroff, not sure what you mean, or not sure you know what I mean.
To date we've rarely if ever seen a 100% performance jump from one generation to another, new architecture or not, 80% is about what you'd aim for. SO I'm saying the 480gtx would be aiming for about that, if its aiming to be 80% faster, but missing gpu clocks by 25%, you're down to being 55% faster, minus 10% because a lot of transistors went on things that won't help gaming in the slightest(infact a heck of a lot more than that most likely), and you're down at around 45% for the 480gtx, and with 12% or so less shaders i'd say the 470gtx will only end up about 35% faster than the card its trying to beat.
Considering if they'd hit their clock targets, that would be about 55-60%, obviously thats clearly way out of the leagues of an overclocked 200b chip. Problem is, they have missed clock targets considerably, you can't just ignore the fact it was never in a million years going to double performance as designed, this isn't the 2 pipelines up to 4 pipelines graphics days with that being the only limit involved. When you factor in a 25% drop in performance, because that is what you get from 25% lower clocks, then you're clearly going to be 25% short of where you were aiming.
Not only would an aim of 60%(for the lower end of the new gen, 80% for the high end) be WAY out of the realms of an overclocked 200b, they of course did try to shrink a 200b, and they couldn't do it AT ALL, let alone with massive overclocks involved.