3 ms·

> You seem to not even understand the very basics of heat transfer. Basic physics: The moon is very cold in surface shadows and below the surface. It is an en

> You seem to not even understand the very basics of heat transfer. Basic physics: The moon is very cold in surface shadows and below the surface. It is an enormous pre-chilled heat sink. The surface is also the support structure for any scale of radiative cooling with the same heat physics as orbit, but much better for larger and enhanced radiative engineering. For example, heat pumps can centralize waste heat energy. Higher heat density vastly increases radiative efficiency. • Permanent shadow: 40-60 ˚K, -230 to 210 ˚C • In polar shadow: 25-30 ˚K, -250 to -245 ˚C • Under 1 meter of surface, equatorial: 250 ˚K, -23 ˚C • Under 1 meter of surface, polar: 200-220 ˚K, -75 to -50 ˚C Many advantages beyond unlimited heat sink/radiative area: all compute in one place, i.e no size limit, so low inter-center latencies, no orbit safety negotiations or periodic orbit re-lifts required, able to update entire data center in a single trip, easier maintenance and stability in gravity on a surface, solar panels can be distributed over distance limiting total space debris risk, different component lifetimes don't result in wasted components, ... Only downsides are a higher Earth-Datacenter latency, lunar dust resistant design, and a need to be at a pole for all-month solar power. Nuclear power, or nuclear + solar, would allow any site. Note that shade can be created anywhere on the surface via reflective shielding, and power can be used to heat, in order to stabilize temperatures in a desired band. Buried installations can use insulation for even greater temperature control.

godelski 9mo ago
Great, now do the math and let's talk
- Nevermark 9mo ago
  Things scale so differently, we don’t need a parts list to make a general tradeoff relationship. Of the moon and orbital, orbit is much closer and will be cheaper to start with. But a lunar site would scale to much greater mean density and unlimited total capacities. And be much cheaper for reasons I gave, at some threshold scale. Neither is easy, and it’s not at all clear that either is actually better than down here. Especially with nuclear efforts and funding rising quickly.
  - godelski 9mo ago
    I'm not asking for a parts list, I'm asking for math
    - Nevermark 9mo ago
      if you want some math beyond identifying different scaling issues, go for it.
      - godelski 9mo ago
        I have. That's why I'm certain you're wrong. But if you're too lazy to do it yourself there's plenty of papers written on this stuff
        Nevermark 9mo ago
        Well then, use your math to address a point. Just one. In a vacuum, radiative heat loss per time hyper scales with temperature to the 4th power. In orbit large and complex heat transfer systems are not going to be practical. On a surface, specialized heat pumps can localize heat energy to very high intensity. With critical reliability advantages of stability, vibration control, complete sun shading, weaker size constraints, etc. That is a tremendous advantage that will overwhelm most other details and tradeoffs, because the two main constraints, and operating costs, are energy production and heat dispersion. The latter imposing a limit on the former. (You have no knowledge of how effectively I use my time. If you have a valid point, make it, instead of - whatever you are doing. Claiming you know things without sharing your reasoning and aspersive language are for the posers. Just communicate why you think, what you think.)
        godelski 9mo ago
        > In a vacuum, radiative heat loss per time hyper scales with temperature to the 4th power. Correct, but you do know that the Stefan–Boltzmann Constant is 5.67e-8 W/(m^2K^4), right? And that emissivity <1 in all real world applications? It is only 1 when a blackbody is radiating in a vacuum. This is how I know you don't understand the math. Because you didn't take the time to understand it. Plug in dummy numbers. I'll make it easy, 100^4=1e8. You didn't think about how T works and the domain we're operating in. T's power isn't a huge factor when we're trying to dump lower levels of heat. Let's say we're trying to discharge our power draw of 300W at 100C, that's still going to take a 0.5m^2 black body radiator sitting in perfect darkness PER CPU!!! A data center has hundreds of thousands! You realize how much fucking surface area that is? And this is before we consider all the other heat generated from the datacenter and the fact that you're dumping heat back to the moon's surface which will radiate it right back at your radiator making it much less efficient. Add the sun and you're fucked. > You have no knowledge of how effectively I use my time. Of course I do. You made a statement so preposterous I know you don't use it to do math or physics. Maybe you watch some math YouTube but that's not the same. I don't have to know everything you do do to know what you don't do. You know how I know this stuff? It's because I've put things into space. Yet you were even too arrogant to check NASA's website
habinero 9mo ago
> The moon is very cold in surface shadows and below the surface. It is an enormous pre-chilled heat sink Technically true, but not really. "Radiative cooling" is heat loss through thermal radiation and it's really ineffective. We use air cooling / water cooling for a reason. Satellites and spacecraft are engineered to make sure they can shed enough heat and they use a fraction of the power a datacenter would. All that energy eventually gets turned into heat, and it has to go somewhere. It's a ridiculous idea that's never going to make even a tiny bit of economic sense.
direwolf20 9mo ago
Is the moon in space? I guess us-east-1 is also a space data center, if you think about it.
panick21_ 9mo ago
> Only downsides are .... Yeah the only downsides are those you listed, and about 1000 others. Anybody that is serious about data centers on the moon should have their brain examined.
- ben_w 9mo ago
  IMO, they're bad, but not so bad as "have their brain examined". While they absolutely do have huge problems at current costs, and I don't trust Musk's estimates for future costs. It's not implausible that collectively humanity (well, China: it's not like the ESA appears to value cheap launches yet) is going to get launch costs down a lot further, something that makes the question of "how cheap is cheap enough?" worth asking. Then you can take a look at the existing constellations and their combined power throughput, look at whatever fraction of that power budget is not radiated by RF/laser output for comms, and trivially that's the power budget with minimal redesign for compute. IMO all of space is still not good enough to be worth caring about: the moon is about twice the difficulty of LEO, and LEO now getting to the point that we're seriously asking about Kessler cascades; but also in space the waste heat is currently only a problem with no currently-useful side-effects, whereas down here on Earth we have possibilities for using the waste heat as an industrial input, e.g. using DCs as the heat source for district heating, or combining with ocean water to become evaporative desalination (which is otherwise pointlessly energy-intensive). That, and the arguments about space-based power is as yet still marginal given how hostile an environment space itself is to PV. And PV on the moon doesn't even get the advantages (launch cost or ~24h light) of PV in a sun-synchronous orbit around Earth.* But it's close enough to not be insane to do a real engineering analysis. Even if the answer turns out to still be 10x more expensive than the ground, which is what I'm expecting it to be. * Side note: for a while I've noticed that China has production and money to afford to build a global power grid on Earth with 1 Ω resistance the long way around. This would allow 24h PV everywhere from deserts on the other side of the world including across seasons. Less material would be required to do this on the Moon because it's smaller, and also you don't really need to go across the equator so it can be much shorter, but also this would need someone to put an aluminium plant onto the moon that has negligible consumables and IIRC we don't have one of those yet. Still, if moon-base design were up to me, I'd suggest sending up 1000 km of HVDC cable on some early missions and put a ring around one of the poles, with some PV every 60° or so. This is still not a sensible design for moon-based compute.

godelski 9mo ago

Great, now do the math and let's talk

Nevermark 9mo ago
Things scale so differently, we don’t need a parts list to make a general tradeoff relationship. Of the moon and orbital, orbit is much closer and will be cheaper to start with. But a lunar site would scale to much greater mean density and unlimited total capacities. And be much cheaper for reasons I gave, at some threshold scale. Neither is easy, and it’s not at all clear that either is actually better than down here. Especially with nuclear efforts and funding rising quickly.
- godelski 9mo ago
  I'm not asking for a parts list, I'm asking for math
  - Nevermark 9mo ago
    if you want some math beyond identifying different scaling issues, go for it.
    - godelski 9mo ago
      I have. That's why I'm certain you're wrong. But if you're too lazy to do it yourself there's plenty of papers written on this stuff
      - Nevermark 9mo ago
        Well then, use your math to address a point. Just one. In a vacuum, radiative heat loss per time hyper scales with temperature to the 4th power. In orbit large and complex heat transfer systems are not going to be practical. On a surface, specialized heat pumps can localize heat energy to very high intensity. With critical reliability advantages of stability, vibration control, complete sun shading, weaker size constraints, etc. That is a tremendous advantage that will overwhelm most other details and tradeoffs, because the two main constraints, and operating costs, are energy production and heat dispersion. The latter imposing a limit on the former. (You have no knowledge of how effectively I use my time. If you have a valid point, make it, instead of - whatever you are doing. Claiming you know things without sharing your reasoning and aspersive language are for the posers. Just communicate why you think, what you think.)
        godelski 9mo ago
        > In a vacuum, radiative heat loss per time hyper scales with temperature to the 4th power. Correct, but you do know that the Stefan–Boltzmann Constant is 5.67e-8 W/(m^2K^4), right? And that emissivity <1 in all real world applications? It is only 1 when a blackbody is radiating in a vacuum. This is how I know you don't understand the math. Because you didn't take the time to understand it. Plug in dummy numbers. I'll make it easy, 100^4=1e8. You didn't think about how T works and the domain we're operating in. T's power isn't a huge factor when we're trying to dump lower levels of heat. Let's say we're trying to discharge our power draw of 300W at 100C, that's still going to take a 0.5m^2 black body radiator sitting in perfect darkness PER CPU!!! A data center has hundreds of thousands! You realize how much fucking surface area that is? And this is before we consider all the other heat generated from the datacenter and the fact that you're dumping heat back to the moon's surface which will radiate it right back at your radiator making it much less efficient. Add the sun and you're fucked. > You have no knowledge of how effectively I use my time. Of course I do. You made a statement so preposterous I know you don't use it to do math or physics. Maybe you watch some math YouTube but that's not the same. I don't have to know everything you do do to know what you don't do. You know how I know this stuff? It's because I've put things into space. Yet you were even too arrogant to check NASA's website

habinero 9mo ago

> The moon is very cold in surface shadows and below the surface. It is an enormous pre-chilled heat sink Technically true, but not really. "Radiative cooling" is heat loss through thermal radiation and it's really ineffective. We use air cooling / water cooling for a reason. Satellites and spacecraft are engineered to make sure they can shed enough heat and they use a fraction of the power a datacenter would. All that energy eventually gets turned into heat, and it has to go somewhere. It's a ridiculous idea that's never going to make even a tiny bit of economic sense.

direwolf20 9mo ago

Is the moon in space? I guess us-east-1 is also a space data center, if you think about it.

panick21_ 9mo ago

> Only downsides are .... Yeah the only downsides are those you listed, and about 1000 others. Anybody that is serious about data centers on the moon should have their brain examined.

ben_w 9mo ago
IMO, they're bad, but not so bad as "have their brain examined". While they absolutely do have huge problems at current costs, and I don't trust Musk's estimates for future costs. It's not implausible that collectively humanity (well, China: it's not like the ESA appears to value cheap launches yet) is going to get launch costs down a lot further, something that makes the question of "how cheap is cheap enough?" worth asking. Then you can take a look at the existing constellations and their combined power throughput, look at whatever fraction of that power budget is not radiated by RF/laser output for comms, and trivially that's the power budget with minimal redesign for compute. IMO all of space is still not good enough to be worth caring about: the moon is about twice the difficulty of LEO, and LEO now getting to the point that we're seriously asking about Kessler cascades; but also in space the waste heat is currently only a problem with no currently-useful side-effects, whereas down here on Earth we have possibilities for using the waste heat as an industrial input, e.g. using DCs as the heat source for district heating, or combining with ocean water to become evaporative desalination (which is otherwise pointlessly energy-intensive). That, and the arguments about space-based power is as yet still marginal given how hostile an environment space itself is to PV. And PV on the moon doesn't even get the advantages (launch cost or ~24h light) of PV in a sun-synchronous orbit around Earth.* But it's close enough to not be insane to do a real engineering analysis. Even if the answer turns out to still be 10x more expensive than the ground, which is what I'm expecting it to be. * Side note: for a while I've noticed that China has production and money to afford to build a global power grid on Earth with 1 Ω resistance the long way around. This would allow 24h PV everywhere from deserts on the other side of the world including across seasons. Less material would be required to do this on the Moon because it's smaller, and also you don't really need to go across the equator so it can be much shorter, but also this would need someone to put an aluminium plant onto the moon that has negligible consumables and IIRC we don't have one of those yet. Still, if moon-base design were up to me, I'd suggest sending up 1000 km of HVDC cable on some early missions and put a ring around one of the poles, with some PV every 60° or so. This is still not a sensible design for moon-based compute.