• sp3ctr4l@lemmy.dbzer0.com
    link
    fedilink
    English
    arrow-up
    4
    arrow-down
    1
    ·
    edit-2
    3 days ago

    You can build a pc with a minisforum BD795i SE (also an integrated CPU, but not LPDDR RAM, its a standard desktop mobo but small, itx, and you can’t remove the 7945hx CPU)) and an AMD 9070, for under $1500, (thats including everything in the pc) and that’ll do basically everything you could possibly want at 1440p.

    I have a wish list on newegg that is just barely above $1500 (at least pretax), including also a decently good 32", curved 1440p monitor.

    These kinds of uh, high grade laptop cpu but on an other wise standard, but small pc mobo, that can be the size of a large lunch box… these are pretty common in China, but fairly few in the West seem to have caught on.

    PC part picker still doesnt even list these mobos in the right way, you can’t plan a build with them, because wtf its a mobo and a cpu at the ssme time? That breaks how our website works!

    • F04118F@feddit.nl
      link
      fedilink
      arrow-up
      2
      ·
      3 days ago

      For gaming, yes.

      But the Framework Desktop seems to be made for a different usecase:

      128GB of unified RAM

      Unified means both CPU and GPU have access to it. Why would you need so much RAM and why would you care if the GPU has direct access?

      Neural Networks. You can fit a pretty serious LLM into 128GB of RAM, and if GPU has direct access, still run inference at reasonable performance.

      I love my 9070XT, but you can’t run anything approaching what Claude or ChatGPT gives you in just 16GB of VRAM

      • Ulrich@feddit.org
        link
        fedilink
        English
        arrow-up
        4
        ·
        2 days ago

        the Framework Desktop seems to be made for a different usecase

        They specifically advertise it as a gaming device. The author of this article is using it as a gaming device.

        You can fit a pretty serious LLM

        LLMs are fucking stupid and no one should use them.

        • F04118F@feddit.nl
          link
          fedilink
          arrow-up
          4
          ·
          edit-2
          1 day ago

          You are absolutely right. Baffles me why they’d put 128GB of RAM in there and use an SoC arhitecture where RAM is shared with GPU, to the detriment of upgradability, if not for AI.

          Any gamer would prefer upgradable RAM and upgradable GPU, especially from Framework.

          How else would you explain this decision to compromise their brand values and overspend on RAM, if not AI?

      • sp3ctr4l@lemmy.dbzer0.com
        link
        fedilink
        English
        arrow-up
        3
        ·
        edit-2
        2 days ago

        … I mean… 128 gb of shared ram could be used that way…

        But… you really think this is a common enough use case to design a pc around?

        Like… how many people want to game, vs… run / train a local LLM?

        Is this what Framework is doing, making LLM development boxes?

        Not… trying to make an affordable, general use oriented, ‘buy once and then upgrade with parts/modules as you prefer’?