• Grimy@lemmy.world
    link
    fedilink
    English
    arrow-up
    1
    ·
    edit-2
    6 days ago

    I did some quick math with metas llama model and the training cost was about a flight to Europe worth of energy, not a lot when you take in the amount of people that use it compared to the flight.

    Whatever you’re imagining as the impact, it’s probably a lot less. AI is much closer to video games then things that are actually a problem for the environment like cars, planes, deep sea fishing, mining, etc. The impact is virtually zero if we had a proper grid based on renewable.

    • boor@lemmy.world
      link
      fedilink
      English
      arrow-up
      2
      ·
      edit-2
      1 day ago

      Please show your math.

      One Nvidia H100 DGX AI server consumes 10.2kW at 100% utilization, meaning that one hour’s 42 day’s use of one server is equivalent to the electricity consumption of the average USA home in one year. This is just a single 8-GPU server; it excludes the electricity required by the networking and storage hardware elsewhere in the data center, let alone the electricity required to run the facility’s climate control.

      xAI alone has deployed hundreds of thousands of H100 or newer GPUs. Let’s SWAG 160K GPUs = ~20K DGX servers = >200MW for compute alone.

      H100 is old. State of the art GB200 NVL72 is 120kW per rack.

      Musk is targeting not 160K, but literally one million GPUs deployed by the end of this year. He has built multiple new natural gas power plants which he is now operating without any environmental permits or controls, to the detriment of the locals in Memphis.

      This is just one company training one typical frontier model. There are many competitors operating at similar scale and sadly the vast majority of their new capacity is running on hydrocarbons because that’s what they can deploy at the scale they need today.

      • Grimy@lemmy.world
        link
        fedilink
        English
        arrow-up
        1
        ·
        1 day ago

        I should have specified it was an earlier llama model. They have scaled up to more then a flight or two. You are mostly right except for how much a house uses. It’s about 10,500 kW per year, you’re off by a thousand. It uses in an hour about 8 hours of house time, which is still a lot though, specially when you consider musks 1 million gpus.

        https://kaspergroesludvigsen.medium.com/facebook-disclose-the-carbon-footprint-of-their-new-llama-models-9629a3c5c28b

        Their first model took 2 600 000 kwh, a plane takes about 500 000. The actual napkin math was 5 flights. I had done the math like 2 years ago but yeah, I was mistaken and should have at least specified it was for their first model. Their more recent ones have been a lot more energy intensive I think.

        • boor@lemmy.world
          link
          fedilink
          English
          arrow-up
          2
          ·
          1 day ago

          Thanks for catching, you are right that the average USA home is 10.5MWh/year instead of kWh. I was mistaken. :)

          Regarding the remainder, my point is that the scale of modern frontier model training, and the total net-new electricity demand that AI is creating is not trivial. Worrying about other traditional sources of CO2 emissions like air travel and so forth is reasonable, but I disagree with the conclusion that AI infrastructure is not a major environmental and climate change concern. The latest projects are on the scale of 2-5GW per site, and the vast majority of that new electricity capacity will come from natural gas or other hydrocarbons.

    • Damage@feddit.it
      link
      fedilink
      English
      arrow-up
      2
      ·
      6 days ago

      If their energy consumption actually was so small, why are they seeking to use nuclear reactors to power data centres now?

      • null@lemmy.nullspace.lol
        link
        fedilink
        English
        arrow-up
        1
        ·
        edit-2
        6 days ago

        Because demand for data centers is rising, with AI as just one of many reasons.

        But that’s not as flashy as telling people it takes the energy of a small country to make a picture of a cat.

        Also interesting that we’re ignoring something here – big tech is chasing cheap sources of clean energy. Don’t we want cheap, clean energy?

        • boor@lemmy.world
          link
          fedilink
          English
          arrow-up
          1
          ·
          3 days ago

          AI is the driver of the parabolic spike in global data center buildouts. No other use case comes close in terms of driving new YoY growth in tech infra capex spend.

        • Leon@pawb.social
          link
          fedilink
          English
          arrow-up
          0
          ·
          5 days ago

          Sure we do. Do we want the big tech corporations to hold the reins of that though?

          • Valmond@lemmy.world
            link
            fedilink
            English
            arrow-up
            1
            arrow-down
            1
            ·
            4 days ago

            If cheap(er/better) energy is invented then that’s good, why would tech corpos be able to “hold the reins” of it exclusively?

            • Leon@pawb.social
              link
              fedilink
              English
              arrow-up
              1
              ·
              4 days ago

              Well, patents and what have you are a thing. I’m mostly thinking that I wouldn’t want e.g. Facebook to run any nuclear reactors or energy grids. That’s something I prefer the government does.

      • Imacat@lemmy.dbzer0.com
        link
        fedilink
        English
        arrow-up
        1
        ·
        6 days ago

        To be fair, nuclear power is cool as fuck and would reduce the carbon footprint of all sorts of bullshit.

      • finitebanjo@piefed.world
        link
        fedilink
        English
        arrow-up
        0
        ·
        edit-2
        6 days ago

        Because the training has diminishing returns, meaning the small improvements between (for example purposes) GPT 3 and 4 will need exponentially more power to have the same effect on GPT 5. In 2022 and 2023 OpenAI and DeepMind both predicted that reaching human accuracy could never be done, the latter concluding even with infinite power.

        So in order to get as close as possible then in the future they will need to get as much power as possible. Academic papers outline it as the one true bottleneck.

        • Valmond@lemmy.world
          link
          fedilink
          English
          arrow-up
          0
          arrow-down
          1
          ·
          4 days ago

          And academia will work on that problem. It reminds me of intel processors “projected” to use kilowatts of energy, then smart people made other types of chips and they don’t need 2000 watts.

          • finitebanjo@piefed.world
            link
            fedilink
            English
            arrow-up
            1
            arrow-down
            2
            ·
            edit-2
            4 days ago

            Academia literally got cut by more than a third and Microsoft is planning to revive breeder reactors.

            You might think academia will work on the problem but the people running these things absolutely do not.