There is a wonderful talk from 9 years ago for the Stanford EE380 class by Bob Colwell titled "Things CPU Architects Need To Think About" which remains relevant

There is a wonderful talk from 9 years ago for the Stanford EE380 class by Bob Colwell titled "Things CPU Architects Need To Think About" which remains relevant today. Bob was leader of the Pentium Pro through Pentium 4 projects at Intel. There are fascinating tidbits in there including things like at most 4 people actually understood the processor as a whole, how you have to introduce random delays in order to improve performance, complexity, and even how the Itanium was accepted. He also mentions what happened with the Pentium FDIV bug. Go to http://www.stanford.edu/class/ee380/ay0304.html http://www.stanford.edu/class/ee380/ay0304.html find 18 February 2004 and then click on the icon on the right - it goes to http://stanford-online.stanford.edu/courses/ee380/040218-ee380-100.asx http://stanford-online.stanford.edu/courses/ee380/040218-ee3... which can be played with VLC on Linux. What this highlights is how complexity is creeping in everywhere - the CPU that looks like a deterministic black box to software developers is actually extremely complex. Heck even a "network card" these days isn't something that just shovels packets to/from the wire. Many even have an ARM processor on them. The same is true for storage controllers.