http://www.techpowerup.com/gpudb/1963/radeon-hd-8870.html
And this description of Hainan is clearly on the nose! These fantasy web descriptions have been around for a long time, and they are just that - fantasy.
What was it, as it's now deleted?
Totally guilty as a fantasy-descriptor or more-so spec-assumer, but I truly fail to see what makes sense other than 1792. Be it something like 800/5000 in mobile (super power efficient) to ~1100/7000 (top-end of a realistic reference clock plus fastest available ram) on the desktop, it simply makes the most sense. The real question is when ya'll feel is the right time to set it upon the world given the current niche of Tahiti variants, or perhaps both Tahiti and Pitcairn. The current stack wins on either value and/or performance with Pitcairn having to be the cash cow; the new stack could be more efficient with units/bw and fit certain TDPs great, but perhaps settle in odd niches compared to competing products diminishing their ASP/margin. If it were the case, one could see the hesitancy to replace 7000; essentially spending more money to sit in the same performance hierarchy (although with better relative performance) because of tdp differences.
The stock clocks of 770 (and leaks of a product to settle between 660ti/670) all-but-cemented this as AMD's plan at some point in my mind. In fact, 770 is straight-up obviously built in anticipation of exactly that...would not be surprised if 760 is similar. It's pretty tough to say GK104 at 1085/7000 and a greater than 225w TDP (for overclocking) is not looking for a niche between 7970, which absolute performance it will be difficult to beat given the unit/bus setup, and such a part as described that would be more efficient within 225w because better die size/clock/bandwidth engineering. Same goes for 1536sp and relatively low clocks vs 6SMX and higher clocks. Ya'll would be more efficient within 150w, but nvidia will probably shoot against the TDP of Pitcairn. I could see how with nvidia organizing their stack like that new products may fail to perform better while perhaps being more expensive to produce than the products they are replacing one market up in the current stack. I just hope we end up with efficient products rather than products (perhaps held so they can be) shoe-horned to compete or mangled because of market conditions relative to the older stack.
That means: I want a Bonaire with 7ghz ram that competes with 650tiB. I want a 150w Hainan that kills anything else in that segment. I don't want a 6ghz Bonaire because it is cheaper and/has a niche/because 7850 exists. I don't want a 150-225w 6ghz/1536sp-1792 part because it competes better with >660 or because it is cheaper while still competing with 660ti+...7870XT/7950 can do that.
I'm not saying it's not possible you're planning something else, or that if it was ever the plan it has changed, but in my mind I certainly can see the possible predicament and conceivably the different directions this could be going.
Point being, outside of clocks/memory choice/tdp placement, I find it really difficult for anyone to argue spec. We all know precisely what these products need to be...including nvidia.
