Next thing people are saying you can't do inference or training because satellite to satellite latency is too big

Also mostly wrong: https://x.com/i/grok/share/34b7a26aa6954d5ab02d8bbdf63fd2b2

@levelsio Personally, I am worried that after earthss data centers use up all the water, space data centers will use up all the vacuum in space

Nobody is talking about this yet but its a huge risk

You CAN put massive radiators, it's all technically feasible.

BUT the costs are the issue. Sending these massive radiators in space and the whole project itself is not economically sensible.

It would be wayyy cheaper to just build datacenters on earth, with micro reactors. We have ample space on earth for this and maintenant + costs will be wayyy cheaper than putting that stuff in space.

From my research a 1kg gpu would involve sending a 100kg radiator in space. It does not make economic sense.

Elon is just trying to pump the IPO, spaceX went from launching satelites + colonizing mars to an AI company in a matter of months because there's hype.

@edon_d Nothing is easy

@levelsio What's the efficiency for radiative heating like that?

@Marco777Polo Yes so that's why you make the panels gigantic

@complex_maths Very bad but you make them absolutely gigantic so it doesn't matter

@levelsio I can't believe you think space datacenters are easy to do!

@levelsio I don't think they are doing it because it is cheaper or more efficient. But because in space, there is no government to regulate AI, they can run whatever model they want on those satellites, and nobody can stop them.

@levelsio They are block heads rage bating, my 16 year old daughter could explain this to me. Assuming you did not attend a single class of A-Level or High-School Physics surely you can @grok or ask ChatGpt. I mute people like this.

@pbertrand_dev @levelsio broski, the water system in data center is a closed loop. How in the world they are using watter ffs

@pbertrand_dev @levelsio Oh no! Does that mean our vacuums here will stop working?! Wake up sheeple!

@levelsio i thought the main issue would be maintenance no? also it cant be one big data center...have to be a swarm of smaller satellites interconnected.

@complex_maths @WissamKadamani @levelsio Yes, of course, because they work via capillary action. satellites have been using radiators for half a century. There’s an incredible amount of literature on this.

@levelsio The sun is dying. It will leave us with nothing but darkness & cold as we hurtle into the void

@levelsio The physics of radiating heat to space are well understood but actually executing that in space are not. Heat has to be moved from the chips to the radiator. Such a system could not use water or any liquid that an Earth-bound system

@levelsio As if the existing Starlink constellation isn't already producing ~100 MW already. Jumping to GWs is not that big of a leap...

@ibobnotnot @levelsio @complex_maths Negligible compared to the opportunity cost of waiting years for permits and ramping up of the necessary amount of electricity production here at earth.

You can’t just plug into the grid, but in space you can cause the grid is the rays from the sun

@levelsio Have you seen the trouble of good and reliable ISS cooling? It's disturbingly loud inside nonetheless the PhD papers optimising it.

And they are quite low on energy density compared to a chip, which has a higher energy density than a cooking stove.

@complex_maths @levelsio The difference in chip die temperature to liquid temperature is between 5C and 20C. With good thermal design, no more than a 5-10C difference. Maybe another 10-15C for liquid to radiator average temperature, so ~20C difference overall. GPUs have a 93-110C max die temp. It closes.