I think one of the big things that a lot of people overlooked in the previous RISC revolution several decades ago was that it was only partly a hardware phenomenon. It went hand in hand with great advances in compiler technology and it was also inseparable from changes in the typical tasks to which processors were applied. The dragon book was brand new then! IMO, there's no such thing as a good hardware design in isolation from the other two. This revolution sounds somewhat similar in that the typical workload have changed. The 1980's chip revolution happened along with an exponential increase in the use of processors to handle Unix-like mixed workloads characterized by having humans throwing a blizzard of diverse tasks at them. A decade earlier the important workload to support had fewer humans in it--it was classic "data processing." CUNY still had a card punch for the mainframe when I started CS in 1986. I suspect that todays typical workload has changed almost that much again. Hardware design follows algorithmic advances just as much as the other way around.