at about 10 minutes into this talk i become much more susceptible to @Kartik Agaram’s burn it all down and start again philosophy. memory layout randomness/processor caching etc can introduce performance variability of up to 40% and can lead to things like performance differentials based on which user is logged in (changes environment variables, changes the cache layout)

Heh, Emery Berger used to sit two cubes over from me for a couple of years in grad school. Definitely much smarter than me. I'm way past the time when I can use being younger as a fig leaf. To clarify my own position, though: My philosophy is to start again, but not necessarily burn it all down. That may be excessive until we have a proven alternative. But if we find ourselves in a hole, maybe we should stop digging.

That is a good talk by Berger. It confirms my own tests where i turned off all the fancy intel instructions and didn't even get 1% difference in performance. It proved to me that Intel is just wasting everyone's time adding nutty instructions since the Pentium that basically accomplish a form of copy protection by forcing cloners to waste massive amounts of engineering time implementing worthless instructions merely for compatibility. Thank goodness for ARM which is driving energy efficiency at Intel; at least CPU's take a lot less power to run.