Comment on Datacenters in space are a terrible, horrible, no good idea.

• brucethemoose@lemmy.world ⁨7⁩ ⁨hours⁩ ago

  100kW? Nvidia BGX 200 servers are 14kW each, not counting the interconnect, or anything else. According to nuggets I’ve read online, we’re talking 200 megawatts for an Earth-based AI datacenter these days, without something exotic like underclocked Cerebras WSEs

  Plugging 200 megawatts into this:

  www.calctool.org/…/stefan-boltzmann-law

  I get about 0.46 square kilometers, depending on the coolant temperature and ultimate efficiency of the system.

  I have no clue what the construction of such a monstrosity would look like, but if it was a simple 0.5 inch aluminum sheet, it would weigh like 15,000 metric tons.

  source

  • NotMyOldRedditName@lemmy.world ⁨7⁩ ⁨hours⁩ ago

    They aren’t making a datacenter like on earth. They’re putting up a ton of satellites that will each generate about 100kW.

    Everyone keeps thinking they’re putting these massive things up there, they are not doing that.

    source

    • brucethemoose@lemmy.world ⁨7⁩ ⁨hours⁩ ago

      That’s interesting, but what’s the point? If it’s like 2 DGX boxes in each satellite, spaced out, the interconnect between them is going to be very slow, and the individual computational power of each satellite will not be that impressive.

      And if you connect them all in one constructed mesh and wire them together, well, you’ve made a 200MW datacenter! The economies remain the same.

      If hardware gets more power efficient, well… Then why do you need to go to space anymore?

      source

      • NotMyOldRedditName@lemmy.world ⁨7⁩ ⁨hours⁩ ago

        Ya, the economies of how much total space / material for the global network is similar, although lets say higher due to losses in efficiency in distributing it over so many dishes, but in terms of how big any individual radiator is and how much space each one is going to take, the smaller sizes make it easier to manage. Trying to figure out a 150-200m^2^ solar panel is a lot easier than trying to figure out a 1km^2^

        The individual power of each satellite having to use a mesh network to train might not be fast enough, maybe they’ll still use land based ones for training, but no single person needs more compute than what a satellite can provide. So from the inference / customer computation side of things, it isn’t a problem.

        source