Framework just announced their Desktop computer: an AI powerhorse?

Recently I’ve seen a couple of people online trying to use Mac Studio (or clusters of Mac Studio) to run big AI models since their GPU can directly access the RAM. To me it seemed an interesting idea, but the price of a Mac studio make it just a fun experiment rather than a viable option I would ever try.

Now, Framework just announced their Desktop compurer with the Ryzen Max+ 395 and up to 128GB of shared RAM (of which up to 110GB can be used by the iGPU on Linux), and it can be bought for something slightly below €3k which is far less than the over €4k of the Mac Studio for apparently similar specs (and a better OS for AI tasks)

What do you think about it?

  • @[email protected]
    link
    fedilink
    English
    72 days ago

    Memory bandwidth is 256GB/sec, much less than M4 Max (526GB/s) or M2 Ultra (800GB/s). Expect performance to reflect that.

    • @SuperIce
      link
      English
      62 days ago

      It’s comparable to the M4 Pro in memory bandwidth but has way more RAM for the price.

      • @[email protected]
        link
        fedilink
        English
        41 day ago

        Good point. You can’t even get an M* Pro with 128GB. Only the Max and Ultra lines go that high, and then you’ll end up spending at least twice as much.

  • The Hobbyist
    link
    fedilink
    English
    102 days ago

    I think it has potential but I would like to see benchmarks to determine how much. The fact that they have 5Gbps Ethernet and TB4 (or was it 5?) is also interesting for clusters.

  • @[email protected]
    link
    fedilink
    English
    6
    edit-2
    2 days ago

    It’s absolutely positioned to be a cheap AI PC. Mac Studio started gaining popularity due to its shared RAM, Nvidia responded with their home server thing, and now AMD responds with this.

    It being way cheaper and potentially faster is huge. the bad news is that it’s probably going to be scalped and out of stock for some time.

    • @[email protected]
      link
      fedilink
      English
      31 day ago

      Perhaps I’m mistaken but I read this as “Nvidia takes 3k just for the chip” and framework “2,5k for the whole system”?

      Either way exciting news for self hoster in the next years!