[PowerfulWizard]

To remove heat by radiation there's a big benefit to running the GPUs hotter as the radiation will be proportional to the fourth power of temperature. This resource is using 85 deg C versus 60 deg C for OP which will improve cooling performance.

If Nvidia/SpaceX can make the chips run at a higher temperature that would help a lot although I assume it would have already been done if it were possible. Another option is to add a heat pump to raise radiator temperature if the smaller radiator mass can pay for the heat pump mass.

[cornholio]

If radiator temperature is critical you can just run a heat pump to boost it. Modern heat pumps can get pretty close to the Carnot limit for the temperature interval, for example a pump cooling to 300K and dumping out heat at 400K (125°C) will have a theoretical COP limit of 4 and a practical limit of 2.5-3.

That means that for every 3 units of heat your chips emit, you will use a 4th unit to spin the pump. If your panels generate 150kW, you will only have 113kW available for compute and the rest is cooling. Radiators will more than halve vs 340K operation, so it's net beneficial.

It's all a giant techno-economic optimization problem: the extra mass of solar panels you need vs saved radiators vs mass of the pump, the die temperatures you achieve and corresponding performance, durability and chip price point etc.

[cucumber3732842]

Could they run the chips at "normal" temperatures and then heat pump that all into some other mass that lives at "aerospace alloy" temperatures and radiates it away more efficiently?

[PowerfulWizard]

I think it's possible, I'm not familiar with what's really been tried but I see that there are some publications on it, for example there's an ESA project under way: https://connectivity.esa.int/archives/projects/heat-pump-cen...

[mc32]

Sorry but wouldn’t space based DCs need radiation hardened parts which are typically a few generations behind SoTA? Do the centers deploy massive shields to protect the electronics?

[PowerfulWizard]

I haven't heard of that being a problem for Starlink, whatever they've done there seems to be working. The Orbital AI would be at a higher orbit but still within LEO which reduces the radiation. I wonder if occasional bit flips even matter for typical AI calculations. Quantization seems to suggest a lot of the lower bits aren't all that important. Either way SpaceX has already had success building systems with commercial-off-the-shelf products in space.

To me the biggest problem of the system is the dependence on launch cost. The more launch cost can be lowered the more viable the system is and the more options there are for solving every problem. But IMO the reason Musk is building this system is to lower launch cost. He needs a project that can fund a high volume of launches and create economies of scale to lower launch cost, the way that Starlink did for Falcon 9.

[bigbadfeline]

> I haven't heard of that being a problem for Starlink, whatever they've done there seems to be working.

Starlink doesn't need GPUs, they use hardened chips. The first question is, can whatever works as a 2nm GPU work where Starlink is?

The second question is even funnier: Starlink satellites have a limited lifespan and regularly burn up in the atmosphere when they fall back to Earth - that's due to the need to fly them at low altitudes where, coincidentally, the radiation is lower.

Where then will the flying data centers be - high with higher radiation or low with a higher fall rate? And I doubt those heavy birds are going to burn up fully before hitting somebody's head.