Its a respectable jump with the combination of changes to GCN2.0 (which is mostly addressing HSA enablement and some other changes of which I can't give hints) and most importantly process changes.
GCN2.0 is not going to be the radical change we saw from VLIW4 to GCN1.0. That should be a good indicator what to expect.
I know what you mean, there are wild rumors out there about just how much of a performance jump these are supposedly going to be, those were mildly exaggerated. Energy efficiency is a different story. Those numbers I don't know but process improvements will definitely give more room to save power or use the saved power to increase clocks. Even after knowing the final clocks I wont be able to determine what the final power draw will be because a sizable board/component change can lead to different power draw. Reference board and then AIB dependant.
On the mobile front: Turns out GCN2.0 is indeed in the line up. My friend who suggested its a tweak of GCN1.0 was not wrong technically. He said what he noticed on the test program level. He doesn't have to know the design/changes, he wont be able to see it unless he asks his higher ups. Everything is compartmentalized on a need-to-know basis. So the story on silicon being different is checking out. But the family of Shader processors were not really revealed to the press. Press have interpreted what they felt. Even its not confirmed that the 8800 series is a re-brand as originally thought. That was press speculation not AMD's confirmation I went through a couple of the latest press core reports on this piece. Couldn't find a direct quote from AMD rep saying anything of the sort.
There is a good strong reason for this change in assessment on my part because I have now a better understanding of changes between GCN1.0 and GCN2.0 so based on that I am going to go out on a limb and say all 384 shader parts revealed yesterday are indeed complete GCN2.0 parts. The 8800M sereis 640 shader part is also if my understanding is correct GCN2.0 but I could be wrong since there are corresponding SI parts to match, then again re-spinning an older generation is cost and manhours, does not sound very effective financially to me when they can just focus on CI based silicon. But those 384 shader parts are pretty much GCN 2.0.
Spinning new silicon for an older generation when the new generation is already there is not business sense for AMD especially in these trying times. They are also trying to push HSA forward as much as possible so improvements in GCN2.0 cant come any sooner for them. Developers need time if they want to work on something if they want to have it out in 1-1.5 years time as Beta.
Let me put it this way, whatever AMD told the press was correct the only thing is they didnt tell the press the families or the core differences that is saved for a later date. I don't know when. Anandtech says Jan 2013? Maybe, this I don't know.
To me it seems reasonable they don't want to talk about GCN2.0 right now. Everyone likes to have some fanfare when they make big announcements so maybe thats why they didn't disclose anything on the solar system lineup.
From my understanding for the same shader count SI and CI die sizes will be virtually identical not really easy to see the size difference unless we are talking about the larger models even then its going to be very small.
Yeah thats all I have now. If anyone is waiting on these cards its all good. They will be happy with their purchases for sure.