Most of alexis work is based on my opensource optimized maxwell ccminer kernels. I have more than 1000 commits to github. Don't forget this... Anyway, the unreleased nist5 private is doing 55mhash. Ccminer 2.1 45 mhash and alexis 1.0.0 48mhash.
Gtx 1070 same clocks and launch config.
Lie!
Alexis was the first to implement shared mem technique that gives the significant boost
In nist5? nope.
Alexis improved AES by moving some of the precalc tables from shared mem to the level 1 cache. This give a significant boost in most of the X algos. (x11,x13,x15,x17)
The nist 5 isn't using AES.