FRESH

Hacker News

Home

Nvidia-backed Starcloud trains first AI model in space, orbital data centers

15 points by walterbell

by leothetechguy

5 subcomments

I remember another hacker news commentator describing these orbital data centers as a obviously bad idea to the point where any investments into that technology are incomprehensible. I share that sentiment, is there something I'm missing?

by freefaler

0 subcomment

The great Scott Manley did a video about that:
https://www.youtube.com/watch?v=d-YcVLq98Ew
basically it doesn't make sense with current technologiesa and even with Starship's proposed specs/price it won't be profitable.
Check it out...

by m_rpn

0 subcomment

They've run out of terrestrial snake oil to sell so they now need interstellar snake oil.

by palmotea

1 subcomments

> Washington-based Starcloud launched a satellite with an Nvidia H100 graphics processing unit in early November
An. That's not a datacenter, that's a server.
> Starcloud CEO Philip Johnston told CNBC that the company’s orbital data centers will have 10 times lower energy costs than terrestrial data centers.
> “Anything you can do in a terrestrial data center, I’m expecting to be able to be done in space. And the reason we would do it is purely because of the constraints we’re facing on energy terrestrially,” Johnston said in an interview.
Does that include lifting acres of solar panels into orbit?
If this takes off (no pun intended), maybe we'll get to deal with fun future problems like AI data centers blotting out the sun.

by moi2388

2 subcomments

“ Anything you can do in a terrestrial data center, I’m expecting to be able to be done in space,” Starcloud CEO Philip Johnston told CNBC.”
In theory, yes. But this cannot possibly be economical.
Any idea how much solar panels you’d need to power an entire data centre from space?
And how insanely much space you need for radiating away heat? There is no conduction or convection, so I’d love to see them try, and make this economically viable.

by xt00

5 subcomments

So using Stefan-Boltzmann equation if you have a 1m^2 surface at 100C you can radiate about 1kW from that surface -- assuming both sides radiate that, then lets assume it is double. Assume each blackwell chip + support electronics etc needs about 2kW of power to run. So each 1sq meter of say a copper plate is needed to cool 1 blackwell chip. So if you have some way to make some massive radiators that are basically giant plates spanning thousands of square meters, then you should be good. the Stefan-Boltzmann equation is proportional to the 4th power of T (in kelvin), so if you can somehow manage to use a heat pump for the heat from the GPU's into your heat sink such that you could run your radiators at a much hotter temperature, then the blackbody radiation that they put out dramatically goes up. So cooling is quite challenging but not impossible. (I also neglected importantly that you would need to use the giant solar panels as a sun shade for these radiators otherwise they would be pulling in heat from the sun)
For power, you need to somehow manage to generate all of the power that you would need to cool. So the most logical would be some huge solar panels -- assuming you could use similar tech to the space station, you can get aroudnd 100kW from those solar panels -- assume you can do say 10X better somehow, then now you have 1MW of power.
Unclear what the goal here is -- if the idea was doing this for cost, it sounds super unlikely to pan out -- if they want to put a datacenter in space such that nobody can tell somebody what to do, it would seem just as easy to go hide a datacenter in some random far flung corner of the world in a bunker. Seems just like a great way to light some money on fire.

by butvacuum

0 subcomment

Anatoly Cherdenko would be terrified.

by grim_io

0 subcomment

So, AI is now officially rocket science.
Anyways... This is dumb.
Radiation shielding, power, cooling, maintenance. All unnecessarily made more complex.
What for?