As for me, I'm very curious to see just how big and how fast the 390(x) turns out to be.
Just out of curiosity, a while ago I drew (in mspaint) a mock-up of what Hawaii would look like with HBM on an interposer. All three are to scale to actual chip sizes (20 pixels = 1mm). Perhaps I could have re-arranged them to look more how it probably will (two chips per side...probably a 90 degree turn of chips) but I was too lazy to go back and make the line density thinner, as it would get pretty tight (which would make sense, by design). It should give an idea though of what they have to work with in terms of space.
I know we all hope it is 20nm, but on 28nm you can more-or-less guesstimate how big it will/could be (also note Hawaii is pretty much the perfect size to fit chips on two sides given the size of a regulation interposer...which is what I was curious about in the first place.)
I'm also curious if they stick to 64 ROPs. That would mean it would (generally) need to be clocked higher than GM200. Not by a lot, mind you (couple percent), but with their general performance scaling it could add up (5-6% total). That's not counting the extra performance the mem bandwidth would obviously give (probably enough to make up the slightly lopsided rops/shaders if 512Gbps), nor the fact extra shader/compute/sfu resources can help in certain situations (versus nvidia's '3840'), but when we're talking estimated clockspeeds that are very similar (say ~1ghz) with chips that could be relatively similar size (say 24x24 [but perhaps more rectangular] vs 25x25) every little bit matters.
I think the point is it *should* perform well-enough considering where Titan-X will probably be placed. There's certainly a question of how much power it vs GM200 will be given and/or how well GM200 will clock (especially within 300w), but if nvidia is truly going to launch the first parts at $1350 (or some other obscene figure), AMD doesn't really need to worry until nvidia makes an apples-to-apples consumer part, and even then it could be interesting (as amd will probably be running at power efficient clocks where-as a cut-down GM200 would have to run higher and vicariously less power efficient. That will probably be offset vs Titan-X by having less ram, but I digress).
Should be a relatively interesting Q2 this year.