*Saw the graphic and bench result
*Predict the Nvidia pulled their good old approach of chopping the CUDA core counts but increase the stock clock speed to make the card look better like 670->760 and Titan X->980Ti when in fact they are slower clock for clock
*Click on link to see article
*Confirm what I predicated and see core clock speed increase from 1607MHz to 1860MHz (thus reducing the overclocking margin available to user comparing to the 1080) for the sake of showing off better performance in all the benchmark for "stock" performance
You are getting a bit too predictable now Nvidia

Assuming both the 1080 and 1070 can overclock to 2150MHz (I'm not talking about FE, but custom cards that will eventually avoid the temp throttling), 1080 overclocking from 1607MHz to 2150MHz is 33% overclock, where as 1070 overclocking from 1860MHz to 2150MHz is just 15% overclock- which is less that half the overclocking headroom comparing to the 1080.
Comparing an overclocked 980Ti to an overclocked 1070 instead of stock vs stock, or dropping the 1070's clock speed to same as 1080, the 980Ti is probably going to beat it, assuming Nvidia won't have amnesia about the need of optimising for cards that are previous gen for new and upcoming games.