• 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)