As expected, it's a salvage part of GP102. It disables a little bit of a lot of things, from compute units to memory channels. Nvidia decided not to repeat the GTX 970 "4 GB" fiasco and will market it as an 11 GB card. Both the GPU itself and the memory will be clocked higher than the Titan X, making it likely a hair faster in typical situations.
With reviews not yet out because the card isn't launched, we don't officially know exactly how it will perform yet. But really, we pretty much do from the paper specs, as it's a salvage part of a GPU already on the market.
With the GTX 1080 Ti available at $700, a GTX 1080 at $600 looks rather overpriced. Nvidia agreed and slashed prices on the latter to $500. They're also cutting prices on the GTX 1070 founders edition to $400, but you can get other GTX 1070s for cheaper than that, so that's not terribly interesting.
One could ask why it took Nvidia 7 months after the launch of the Titan X for the salvage part to arrive. My guess is that yields weren't very good, and as Nvidia had the top end all to themselves with the GTX 1080 anyway, they didn't feel the need to rush. Do respins, let the process node mature, or whatever it takes for better yields and get good yields before you start really producing GP102 in high volume.
This sets the bar for AMD Vega, due out in the second quarter of this year. Vega should beat a GTX 1080 Ti in most of the paper specs (TFLOPS, memory bandwidth, etc.), but that's been the case for most of the last decade, and Nvidia was able to counter with the ability to more efficiently use the hardware available. For a CPU analogue, think of it as Nvidia had better IPC but AMD had more cores--except that in the GPU world, unlike consumer CPUs, this sometimes meant AMD won.
GCN was a nice architecture in its day, but AMD has been selling derivatives of it for over five years now. If Vega is nothing more than a minorly tweaked GCN, AMD's GPU side could be in trouble, as that's not going to catch Pascal. A scaled up Polaris isn't going to catch a GTX 1080 Ti inside of 300 W and 500 mm^2. That might be why AMD didn't bother to produce such a chip last year. GCN was competitive in 2012 and 2013, but Nvidia made Maxwell a lot more efficient in 2014, and AMD has been generally behind ever since.
If Vega is an all new architecture, or at least a massive overhaul, then it could be just about anything. It could be this generation's equivalent of the Radeon 9700 Pro that lapped the competition, or it could be this generation's equivalent of the Radeon HD 2900 XT that was hot, late, and slow.
The upshot of this is, if you want a high end gaming card, you really should wait until next week. I don't see a dire need to wait for Vega, as I'd be very surprised if Vega blows away Pascal or forces Nvidia to sell the GTX 1080 Ti for $500 and the GTX 1080 for $350. If the top end Vega does match a GTX 1080 Ti in performance, it will probably about match it in price tag, too.