Comments by "Mikko Rantalainen" (@MikkoRantalainen) on "The Universe is Hostile to Computers" video.

  1. What does surprise me is not using ECC memory for any critical process. For avoiding problems in CPU or caches, you can just execute all logic twice in series and report error if second result does not match the first result. So the max to avoid this whole error class is 2x CPU load plus the price of ECC over regular RAM. Modern CPUs have multiple cores so you can (and should, to avoid unnoticed repeatable errors in CPU) run both the "original" and "verification" tasks in parallel so there's no extra latency needed.
    5
  2. Did your computer use ECC memory? Most common problem for random crashes is slightly faulty memory and without ECC RAM you only see the consequences instead of getting a warning or error message about faulty RAM module.
    1