I have a Proxmox server running two Opteron 6272 CPUs on an Asus KGPE-D16 (chosen because it was the fastest computer that supported Libreboot, although I haven’t gotten around to installing it). Using normal BIOS settings, it’s drawing just under 100W at idle, measured via smart plug reported in Home Assistant. With aggressive efficiency settings (PowerCap to P-state 4 and disabling CPU 2 entirely) it idles at 70W. It’s a server, not a gaming PC, so it doesn’t appear to have any options for underclocking or adjusting voltage.

Anybody know of any other ways (maybe software-based) to get the power draw down further?

  • spaghettiwestern@sh.itjust.works
    link
    fedilink
    English
    arrow-up
    2
    ·
    1 day ago

    Tangential question: What kind of server apps require that kind of processing power? I run a server on an Intel N200 laptop with multiple apps and services and it rarely uses more than 12% CPU and 15 watts. I’m wondering if I’m going to eventually run into something that needs a more powerful platform.

      • spaghettiwestern@sh.itjust.works
        link
        fedilink
        English
        arrow-up
        6
        ·
        13 hours ago

        A single Opteron 6272 is somewhat faster than the N200, but the Opteron’s TDP is 115 watts while the N200’s is only 6 watts. OP’s server with 2 processors is more than 2x as fast as my single processor laptop, but can require nearly 40x the electricity. For a home server it’s major overkill.

        • MangoPenguin@lemmy.blahaj.zone
          link
          fedilink
          English
          arrow-up
          1
          ·
          10 hours ago

          Newer CPUs can also just be better optimized and have more faster cache and that sort of thing, so might be faster at running a process even if they’re the same on paper.

    • grue@lemmy.worldOP
      link
      fedilink
      English
      arrow-up
      1
      ·
      1 day ago

      Nothin’ I’m running, that’s for sure!

      It’s not really that there are services that require that much processing power for a single request; it’s that it’s designed to handle normal requests for hundreds or thousands of users at once.

      I suppose that supporting 0.5TB of RAM means it could deal with quite a big LLM, but any sort of halfway-modern GPU would absolutely run circles around it in terms of tokens per second, on any model that fit in their VRAM.

      • spaghettiwestern@sh.itjust.works
        link
        fedilink
        English
        arrow-up
        1
        ·
        edit-2
        1 day ago

        Sounds like my laptop will be plenty fast for some time to come.

        This platform doesn’t use much power to begin with, but I do run TLP using a battery profile despite the fact it’s always plugged in. My intent is to lower the power consumption a bit further and extend battery run time if the AC fails. There’s no noticeable impact on application performance. If you’re running Linux maybe it will work on your hardware.