• technocrit@lemmy.dbzer0.com
    link
    fedilink
    arrow-up
    69
    ·
    edit-2
    3 months ago

    This is an extremely misleading headline.

    From the abstract:

    … applying the L-Mul operation in tensor processing hardware can potentially reduce 95% energy cost by element-wise floating point tensor multiplications and 80% energy cost of dot products.

    In other words… This method of computation could save 95% of the energy spent on floating point multiplication (and save 80% on dot products)… Not 95% of total energy.

    It’s an improvement (potentially), but I don’t see any analysis of how this would impact total energy.

    • IrritableOcelot@beehaw.org
      link
      fedilink
      arrow-up
      5
      ·
      3 months ago

      Good point. Though, the vast majority of ML training and use is tensor math on floating points, so largely dot and cross products, among other matrix operations.

    • Handles@leminal.space
      link
      fedilink
      English
      arrow-up
      22
      ·
      3 months ago

      Came here to say that a 95% reduction in energy consumption will only greenwash a corresponding or larger increase in usage — but yours is of course the correct response! 👏👏👏

    • skarn@discuss.tchncs.de
      link
      fedilink
      arrow-up
      8
      ·
      3 months ago

      But… Isn’t that kind of the point? Slashing computational cost so that we can deploy that stuff wherever it’s needed without a tenfold increase in the world’s energy bill?

      Whether we should do that at all is a very different question.

      • Handles@leminal.space
        link
        fedilink
        English
        arrow-up
        23
        ·
        3 months ago

        “Wherever it’s needed” is the operative term here, isn’t it? Looking at how it’s already being implemented, nobody seems to bother asking whether “AI” is really needed.

        • skarn@discuss.tchncs.de
          link
          fedilink
          arrow-up
          10
          ·
          3 months ago

          Personally, I find myself in a bizarre situation.

          I have some open source ““Ai”” solutions that I find really really nice and helpful e.g. the image search in Immich, or LanguageTool which bills itself as an AI spellchecker.

          At the same time I am horrified at the stupidity underlying 99% of big tech AI stuff that gets wall street hot.

          • Handles@leminal.space
            link
            fedilink
            English
            arrow-up
            7
            ·
            3 months ago

            That’s the difference, isn’t it? People can use “AI” to make simple little things easier. Corporations want it to replace and automate the jobs of swathes of the workforce. It’s the latter that is the “growth market”, and the one that eats the most power.

            • skarn@discuss.tchncs.de
              link
              fedilink
              arrow-up
              5
              ·
              3 months ago

              That wouldn’t be so bad per se… Many improvements in human conditions have been achieved by automating stuff and kicking people out. Think of the green revolution.

              The problem is that the use case here is to massify the production of literal removed, like clickbaity articles on social media content, or ever larger volumes of advertisement. Those jobs don’t need to be replaced, they just need to go away for good.

              Are we really going to use an AI to write motivation letters from a list of bullet points, to send it to an HR that will condense it into a list of bullet points using AI? Seriously?

      • Artyom@lemm.ee
        link
        fedilink
        arrow-up
        1
        ·
        3 months ago

        Improving the technology behind AI will only increase the return on investment per watt, so you’ll want to spend even more on it than before. This would more than likely increase the energy demands (assuming it doesn’t turn into vaporware).

  • kbal@fedia.io
    link
    fedilink
    arrow-up
    8
    ·
    3 months ago

    Isn’t this basically just the old trick of estimating (x * y) as (x + y - 1) when x and y are somewhat close to 1?

  • Jayjader@jlai.lu
    link
    fedilink
    arrow-up
    4
    ·
    3 months ago

    I wonder what other applications this might have outside of machine learning. I don’t know if, for example, intensive 3d games absolutely need 16bit floats (or larger), or if it would make sense to try using this “additive implementation” for their floating point multiplicative as well. Modern desktop gaming PCs can easily slurp up to 800W.

    • DdCno1@beehaw.org
      link
      fedilink
      arrow-up
      1
      ·
      3 months ago

      I have a vague memory of a 3D engine from the 1990s using an approach like this, but I’m not entirely sure.

      • IrritableOcelot@beehaw.org
        link
        fedilink
        arrow-up
        5
        ·
        3 months ago

        I think you’re thinking of the famous fast inverse square root algorithm from Quake.

        With respect to the top comment, the only reason 3d graphics are possible (even at 850W of power consumption) is due to taking a bunch of shortcuts and approximations like culling of polygons. If its a reasonable shortcut it either has or will be taken.