The power savings from HBM2 will more then offset the additional power consumption of the larger die IMO.
Assuming that is true, you are still looking at a much larger die in GP100 plus expensive HBM, but not really any more gaming performance, at least from what we know and can infer. We are confident that the two dies have the same number of FP32 ALUs. Now, we do not know (as far as I am aware) what kind of fixed function graphics hardware (TMUs, ROPs) the GP100 has. However, given its primary purpose as an HPC chip, I strongly doubt that the GP100 has any advantage over GP102 in this area.
So, without the promise of more clock speed or ALUs, the only advantage from a gaming perspective that GP100 has over GP102 is HBM2. For that to significantly matter in gaming, you would have to assume that GP102 is horrendously bandwidth starved even with GDDR5X on a 384-bit bus. I really doubt that.
So, again, from strictly a gaming perspective, I fail to see how Nvidia with the GP100 is going to set the bar for performance much higher than a fully functional GP102. Which brings me back to my original perplexity over Nvidia's (lack of) marketing of this chip. I do not believe we'll see a GP100 consumer chip, it just doesn't get Nvidia anything in the gaming sector. So, I think the best we'll see of the Pascal generation is a fully enabled GP102, which is not going to be much better than the cripped GP102 in Titan XP. It seems far-fetched to think Nvidia is going to offer in just a few short months an even more titan Titan (the Titanest of them all) which may offer, what, 15 or 20% more performance tops, at, what, $1500?