I find it difficult to believe that AMD would release a GPU that's bigger than GP102 yet barely competitive with GP104 - especially if it requires high-cost HBM2. At that point they'd be better off cancelling the consumer version altogether and only releasing the Radeon Pro equivalents (where the higher TFlops will be more important than gaming performance) rather than selling at a loss and getting a slew of bad reviews. This is especially true since if Zen is as good as the leaks indicate, they'll need all the foundry capacity they can get.
There are going to be two Vega GPUs in 2017. This preview could be from Vega 11, which is said to be the smaller of the two. Alternatively, it could be Vega 10 but have badly immature drivers that hurt performance. The third possibility is that AMD could drop the ball again - but this would be a pretty big fumble, almost as bad as Bulldozer, and would basically spell the end of RTG as a viable competitor in gaming.
In 2015, AMD's Fiji release was underwhelming - but they didn't have much warning, since Titan X was only released three months earlier, too soon to make any changes except some minor clock speed bumps and driver tinkering. Even then, Nvidia's victory was mostly based around releasing a cut-down GTX 980 Ti with much better price and performance than most expected. If Nvidia had stuck with the Titan X only, and kept it at $999, then AMD's card would have looked a lot better. AMD was clearly caught off-guard with how much Nvidia managed to improve from Kepler to Maxwell on the same 28nm node. But this situation is different. The GTX 1080 was released last May. We've got a "H1 2017" release timeframe for Vega, which means it will come out probably at least a full year after Pascal hit the streets. If Vega is really as uncompetitive as Termie thinks, I believe that Raja Koduri would have been let go, since that wouid indicate gross incompetence.