Comment on lemmy.world down...

• tempest@lemmy.ca ⁨1⁩ ⁨day⁩ ago

  It’s still not that common to have one die. Servers have ECC memory which can get you by longer even as a DIMM starts to die. It’s also rarely run at speeds consumer RAM is run at.

  The thing with server memory is when you get to point where you have hundreds of servers and each one has 12 to 24 DIMMs the chance you have a bad one somewhere increases.

  source

  • dan@upvote.au ⁨1⁩ ⁨day⁩ ago

    Some servers have ECC. If you get a cheap one (like a Hetzner auction server), it’s less likely to have ECC. ECC protects against bitflips, but it won’t help if the RAM is starting to die. ECC isn’t magic - it just has an extra 8 bits of parity data per 64 bits of data. It still uses the same type of RAM chips.

    source

    • tempest@lemmy.ca ⁨16⁩ ⁨hours⁩ ago

      It isn’t magic but RAM doesn’t tend to just stop working. You’ll start getting reported error correction events on a bad dim before it might fail which usually means it’s replacement time.

      source
  • Quill7513@slrpnk.net ⁨1⁩ ⁨day⁩ ago

    nah for sure, but i’d give a bet ruud was running on consumer hardware which has far less fault tolerance built in

    source