nVidia did always advertise Maxwell as having 3x the double-precision GFLOPS/s per watt compared to Kepler. Maybe they finally decided to trade gaming performance for more mining/compute performance. D:
First of all, that wouldn't make sense from a business perspective, since it would risk cannibalizing sales of their profitable $999 Titans (and even more profitable Tesla pro cards).
Secondly, mining doesn't rely on double-precision, or any kind of floating-point at all. The primary reason AMD cards do better is that they can perform an integer rotate in 1 cycle, compared to 3 for Nvidia cards. Fixing that shouldn't be hard for Nvidia, and it shouldn't require sacrificing performance in gaming applications.