yes I'm also a tad confused about that considering pci-e works natively at 100MHz and anything over that would previously fry whatever you tried running in that interface...what they seem to be implying is that there is some sort of tweak with r5** cores that might allow them to get more out of the unified pipelines, specifically the r580 in this case, from running at a higher pci-e bandwidth
This concept just doesn't seem to make sense considering even two x1900xt cards in crossfire I don't think would use anywhere near the bandwidth of pci-e 8x...although MAYBE this has something to do with increasing the execution time and trying to decrease the latency of core to core communication because they are still using the off-die memory controller and FSB
I'm just talking crap though, I have no idea...would be interested to see who decides to manufacture boards with this chipset and how it compares with nf4 or an nf5 possibly?!