Apparently data centers routinely burn through water at a rate of about 1.9 liters per KWh of energy spent computing. Yet I can 🎮 HARDCORE GAME 🎮 on my hundreds-of-watts GPU for several hours, without pouring any of my Mountain Dew into the computer? Even if the PC is water cooled, the water cooling water stays in the computer, except for exceptional circumstances.

Meanwhile, water comes out of my A/C unit and makes the ground around it all muddy.

How am I running circles around the water efficiency of a huge AI data center, with an overall negative water consumption?

  • Melvin_Ferd@lemmy.world
    link
    fedilink
    arrow-up
    3
    arrow-down
    5
    ·
    5 days ago

    I think it’s more likely that yellow journalism is making an issue out of something that isn’t as big a deal as it is.

      • Melvin_Ferd@lemmy.world
        link
        fedilink
        arrow-up
        2
        arrow-down
        10
        ·
        edit-2
        5 days ago

        When haven’t they?

        Fear and anger sell.

        This AI shit is the leftist version of “illegal immigrants are stealing yur jobs”

        • brucethemoose@lemmy.world
          link
          fedilink
          arrow-up
          3
          ·
          5 days ago

          To be fair, the “infinite scaling” vision Altman and such are selling is quite a dystopia. And they are the ones pushing it.

          It’s not reality at all. But it’s kinda reasonable for people to hate that specifically.

            • brucethemoose@lemmy.world
              link
              fedilink
              arrow-up
              4
              ·
              edit-2
              5 days ago

              No.

              The path I see forward for ML is small, task specific models running on your smartphone or PC, with some kind of bitnet architecture so it uses basically no power.

              That’s the hope, anyway, but all the pieces already exist. Bitnet works, extreme task specific training works with a paper that just came out, NPU frameworks are starting to come together.

              If that sounds incompatible with corporate AI, that’s because it is.