TPU tested a load of games, and got the 580 as being an average of 25% faster in all games they tested. So cant we just use 25% as how much faster it is?
I linked TPUs review summing all resolutions. GTX580 = 100. HD5870 = 77. Mathematically this means GTX580 is 30% faster or HD5870 is 23% slower.
I do not think they are biased. By the way, I think it is a bit misleading to average out all resolutions, with AA on and then off. As others noted, I would object to this because 1) The extra video memory would have given it a distinct advantaged when running the insane settings (2560x1600 w/AA).
Averaging all resolutions doesn't help anyone determine anything, especially if you are trying to inform people. I don't care what the difference between A and B is at 1280x1024, 1680x1050, 1920x1200 and 2560x1600 all averaged together if I'm playing at 1920x1200.
In my #1 perspective point, you can view
any resolution by itself with and without AA. I was not trying to provide misleading results. An average, being a typical amount, rate, quantity, rating, or the like that represents or approximates an arithmetic mean. This is how I arrived at 36% "all resolutions average".
However, here is the complete breakdown (GTX580 vs. HD5870):
1680x1050 0AA = 32%
1920x1200 0AA = 30%
2560x1600 0AA = 31%
1680x1050 4AA/16AF = 40%
1680x1050 8AA/16AF = 33% (shows that AMD is more efficient at 8AA)
1920x1200 4AA/16AF = 38%
1920x1200 8AA/16AF = 31% (shows that AMD is more efficient at 8AA)
2560x1600 4AA/16AF = 53%
If you consider any of the above resolutions, GTX580 is
at least 30% faster. Even if we discount 2560x1600 due to VRAM limitations, you are still looking at 34% difference across what remains (that's why I said GTX580 ~ approximately 35% faster).
The initial review had the GTX 480 barely faster than the 5870 (less than 10% overall) and now it appears to be a solid 20% faster. This, when combined with the faster GTX 580, makes the 5870 trail by at least 35-40% in my opinion.
First thing I would like to point out is that the 480 speeds have seen quite a few increases since launch. If any of the reviews were with launch drivers, then I believe at this point they should be ignored.I think that a good effort was made to show evidence of this 35% claim, using multiply sources.
At the end of the day, the differences will be less or more depending on the games, settings and resolutions. AMD cards tend to perform well in Crysis, BF:BC2, F1 2010. So I definitely expect HD6970 to win in those 3 games. If the current PowerColor rumours of HD6970 being 30-50% faster than HD6870 are true, then HD6970 may even come out with a 10%+ performance lead overall.