I think the argument that R580, with the same memory and core clocks as R520, wouldn't be dramatically faster than R520, is the crux.
First you have to find games with the heavy duty shaders that are going to use that extra ALU power.
All the evidence we have so far is that R520's texturing, rendering, render-target blending and AA are bringing performance improvements of 10-50% more than an equally clocked X800XT. The performance of all of these isn't going to change in R580, I expect, because they're based upon improvements in texture caching and the memory interface/ring bus.
Texturing may get more efficient, I suppose, because with 3 shader quads sharing a texturing quad the texturing quad's idle time may fall to nearly zero. Depends, really, on the scheduler. Maybe the texturing quads in R520 already have near-zero idle time for the same shader code. I wish we knew more about this...
The lack of the new memory controller in RV515 makes comparisons between RV515 and RV530 pretty awkward - and therefore hard to use as a basis for guesstimating R580 on the basis of R520. And the double-Z ROP capability in RV530 further blurs things. But, I hope Dave's going to do an analysis of RV515 v RV530 at some point, anyway...
Jawed
