• Septimaeus@infosec.pub
    link
    fedilink
    English
    arrow-up
    2
    ·
    17 days ago

    For local inference these days it’s basically all AS since it’s the only consumer hardware that can replace $20-80k server builds, but if you mean ML/AI developers specifically, they still prefer discrete GPUs for local training (on personal budgets: either recent consumer gaming/professional cards or older AI cards).

      • Septimaeus@infosec.pub
        link
        fedilink
        English
        arrow-up
        2
        ·
        16 days ago

        My bad, Apple Silicon, the private local ML “meta” as of ~Nov’25 (and last I saw) was RDMA clusters of M3 Ultras that could scale up to 3.6TB VRAM (a lot)