I thought someone had already said it was for bios/firmware flashing of one kind or another. Same way a lot of ssd's have the exact same kind of switch but most production models the switch is hidden behind casing. Pre-production models often have a few switches for either debug modes or to make bios/firmware flashing easier, or other reasons aswell.
information is too silly at the moment, I think AMD made a mistake with not letting someone unofficially leak some reliable info from somewhere we pretty much trust, raised expectations are only bad if your product isn't fast though.
I did say a LONG time ago that 35% would be pretty decent for same process and that AMD has two options really on the same process, bigger, lower yields, uber performance, expensive, similar size and a bit of a performance boost, or same size/marginally smaller, similar performance way cheaper.
Lets be honest, that was the 6870, if they make something 50% bigger than the 6870, with 15-25% more performance I'll be frankly, shocked, and appauled, and may have to sell my AMD stock(literally, it might go down

).
But I can't see it, if the 6970 was too new, and the 6870 brought with it efficiency increase Cayman can't match, why make Cayman at all, just bring out a 6970 that is 2x 6870 and thats it, all done, save money, it would be out already and it would beat anything Nvidia can release. I just can't see them doing that, if the 6870 was THAT close to the 6970 that, at 50% bigger would cost circa 60-65% more to produce(255mm2 WILL have higher yields than a 380+mm2 card, thats just how it works), it would be canceled by now.
Theres a slim chance a lot of the missing performance is in something as simple as driver that doesn't know the 4 basic shaders can get together and do stuff the 5th shader normally does. But come on, seriously, what are the chances that a 50% bigger core is less than 30% faster? I'll give you a hint, none.
Then theres one last thing, maybe AMD screwed up, maybe they went with a new metal layer or transistor type from TSMC and its just not working for them, maybe simplified core logic was supposed to enable a higher clock speed and it didn't work out, but AMD haven't really made a big process mistake in quite some time, even 2900xt is astonishing as its a 65nm product backported to 80nm in 6 months and well, worked, thats impressive, it was also fairly decent clocks and beat the 8800 in things like Bioshock in dx10.