So if I can decipher DM's post, theoretically a 390X with 16gb of HBM1 is possible right now if they stacked the mem and used the 4 slots?
I would say no, not at this time.
What I've been saying since the start is HBM 2 isn't relevant to any particular specification. HBM 2 say having higher density chips doesn't prevent an HBM 1 chip having higher density... HBM 2 is essentially entirely irrelevant to what can be produced today, everyone should entirely ignore it.
WIth a new memory the situation is that we don't know where they are at, with an established product we know when the first chip with X specs was available and thus have a good idea when the second chip and what it can do will be available. IE with a 2000 shader 40nm gpu you can make a good guess as to what you'd get at 28nm, circa 4000 shaders possible. Without the 40nm chip you'd be making a complete stab in the dark at what the 28nm gpu would have.
This is where we are at with HBM, we know the starting spec, but we don't know where the start of the development time line is. Lets say sticking 1GB stacks together to act like a 2GB stack comes 4 months after the first chips, and 8hi stacks comes another 4 months later, then 2x 8hi stacks together comes another 4 months later... then HBM 2 rated chips 8 months after that. Without knowing where we are on that time line you can't know when one spec will be available. When you know where we are on that time line, you can make good educated guesses at when the next thing will be possible.
An educated guess would be maximum capacity and highest speed come with a process node shrink as the image suggested with HBM 2 with 2Ghz spec and 4Gb chips listed as a result of a node shrink.
Personally I'd see the time line of HBM development as :-
HBM 1
1Gb/4 hi/1Ghz > 2 stacks merged > 8 hi > 2 8hi stacks merged
process node shrink giving HBM 2 starting at
4Gb/4 hi/2Ghz > 2 stacks merged > 8 hi > 2 merged > 4 hi 8Gb chips > 8 hi 8Gb
Due to high end product requirements, I would guess we start off effectively at stage two of that development cycle. Giving the 390x the option of 4 stacks and having 1 or 2GB per stack, giving us 8 and 4GB variants of the 390x.
Per capacity cost is higher for HBM than GDDR5, 16GB is both 100% not required for ANY gaming at all and a complete waste of money. It's likely also not ready yet.