• 5714@lemmy.dbzer0.com
    link
    fedilink
    English
    arrow-up
    3
    arrow-down
    9
    ·
    edit-2
    5 months ago

    https://en.wikipedia.org/wiki/Climate_model

    Cloud-resolving climate models are nowadays run on high intensity super-computers which have a high power consumption and thus cause CO2 emissions.[44] They require exascale computing […]. For example, the Frontier exascale supercomputer consumes 29 MW.[45] It can simulate a year’s worth of climate at cloud resolving scales in a day.[46]

    Techniques that could lead to energy savings, include for example: “reducing floating point precision computation; developing machine learning algorithms to avoid unnecessary computations; and creating a new generation of scalable numerical algorithms that would enable higher throughput in terms of simulated years per wall clock day.”[44]

    • Cethin@lemmy.zip
      link
      fedilink
      English
      arrow-up
      8
      arrow-down
      1
      ·
      5 months ago

      Let me point a critical part that you seemed to have skipped:

      … to avoid unnecessary computations…

      Using ML algorithms to add more computations that weren’t necessary doesn’t help. Using it to improve computations can, if it’s more efficient than not using it. ML can be a useful and good thing, but the extreme vast majority of what it’s currently being used for is trying to come up with more places to shove it where it doesn’t reduce computations and instead increases it.

      • 5714@lemmy.dbzer0.com
        link
        fedilink
        English
        arrow-up
        1
        ·
        5 months ago

        I just copied the Wikipedia part, because I thought it funny how AI in media is different from AI in science. I don’t have a stance on the power consumption of climate models because without the models we’d be very unequipped for the storm we brew.

        Sorry for creating the image of me criticising valuable science.