leaked things about tessellation (off old amd slides):
Now scalable (so it ll work probably down to small pixels like nvidias does)
Off chip buffering (apperntly the 5xxx 68xx cards where task-swapping, slowing down tessellation and ROPs effecientcy)
As I already commented, 'scalable tessellation' doesn't mean anything, it's hollow marketing talk.
So we pretty much know they have fixed the tessellation with the new 69xx cards.
No, we know that AMD is apparently worried enough about tessellation to give it media attention.
We have no clue what they improved exactly (they didn't specify, which is usually not a good sign), and how much this is going to help them.
Theres also benchmarks from china, where the 6970 is beating the 480 in extreme tessellaion setup with Unigine by about 25%.
We'll have to wait and see if those are real, then.
Again from china (chiphell) rumors saying the 6970 and 580 perform about equall to one another in Vandtage mark.
Well, if the 580 and 6970 are matched for performance, I'd say nVidia did well with the 580.
On paper, AMD had the most room for improvement, starting from a smaller die size with lower power consumption. If AMD didn't manage to turn that advantage into the fastest DX11 GPU, then nVidia has made a nice comeback. nVidia's GF100 was plagued with manufacturing issues and people considered it way too large and powerhungry. People didn't give nVidia any chance.
All nVidia would need now is a decent dual GPU card, to compete for the performance crown.