• Wildmimic@anarchist.nexus
    link
    fedilink
    English
    arrow-up
    13
    arrow-down
    1
    ·
    10 hours ago

    Since limited waver space (because HBM takes so much of it at the same capacity) is one of the main driving factors for the Ram-apocalyse, this might actually help. And the performance looks actually great (the comparison to the DDR4 i am running shows some parameters being 10x better, this is actually a meaningful upgrade), meaning that some applications that currently use HBM might be able to just use DDR6 instead, causing a greater allocation of wavers towards that.

    • brucethemoose@lemmy.world
      link
      fedilink
      English
      arrow-up
      2
      ·
      4 hours ago

      Generative ML inference isn’t so dramatically bandwidth limited anymore, either, so the “demand” for HBM isn’t as clear.

      • boonhet@sopuli.xyz
        link
        fedilink
        English
        arrow-up
        3
        ·
        3 hours ago

        Eh the really big models want a lot of bandwidth still, otherwise you just don’t make use of all the compute power you have.

        • deadbeef79000@lemmy.nz
          link
          fedilink
          English
          arrow-up
          1
          ·
          2 hours ago

          I generally edit my comments like this:

          Since limited waver wafer space

          So that is fixed for the next reader but the comment chain still makes sense.