• Markaos
      link
      fedilink
      arrow-up
      10
      ·
      3 days ago

      That’s a reasonable per-core size, and it doesn’t make much sense to add all the cores up if your goal is to fit your data within L2 (like in the article)