• xthexder@l.sw0.com
    link
    fedilink
    English
    arrow-up
    20
    arrow-down
    2
    ·
    edit-2
    5 months ago

    My job has been to run things on GPUs for almost 10 years now. The only thing anyone practical is doing on that many GPUs is AI training, massive scientific simulations, or crypto mining. 1 or 2 of them is enough to run something like ChatGPT.

    Real-time graphics it turns out don’t scale well across multiple GPUs. There’s a reason SLI has gone away for consumer GPUs. At the current ratio, each of those $3000+ GPUs is only driving 8000 pixels (assuming each led puck is being used as 1 pixel, given their size). It makes no sense other than bragging rights