^In some tests it equals or tops the 5970. So while technically true that on average it 'loses' to the 5970, it's hardly by much. If the ATI fanboys were dismissing 16% increases as 'tiny' or 'negligible' when the 480 first came out compared to the 5870, I think the Nvidia fanboys can do the same for the 580/5970 considering the performance gap between them is much less than 16%
The issue is not purely performance.
It's a combination of performance, price, power consumption and availability (with the 5000 series being available very early).
The 480 had better performance than a 5870 at the cost of ~100W higher power consumption and more money, as well as not being available for a while.
In the case of the 5970 and GTX 580, the 5970 has better performance, lower cost and equal power consumption.
Yea not to mention the fact that its 2 GPUs vs 1
Usually a card with two GPU's costs more and has higher power consumption, which are the reasons for a comparision being unfair or bad - but that's not the case with the 5970 vs a GTX 580.
The GTX 460 I think is the best card Nvidia has ATM when considering power consumption and price.
It's competitive against the 6870 and 6850 - the 6870 is slightly faster and costs slightly more. So price:
performance is quite close.
I'm not really sure why ATI decided to use the 6870 name for their mid-range card though.
Nvidia are struggling right now in terms of die area efficiency. The GF104 is a step forward, but since ATI moved from a 5D to 4D shader design for Cayman and Barts they've moved the goalposts further. Cayman XT appears to be faster than a GTX 580 while at the same time being smaller.
Moving from 5D to 4D shaders now means that each shader cluster can be saturated easier given the same workloads - i.e fewer shaders in each cluster are being left unused (though I'm not sure why they did 5D in the first place).
Hence how the 6870 has close to 5870 performance with 1120 shaders vs 1600