[OneDonOne]

If the "physics" tells you that your satellite cannot radiate heat away from your nVidia GPU cluster because each H100 needs 1.1 meter square of radiator, then opinions do not matter. The same applies to power supply and bandwidth.

[rayiner]

The radiator is much more efficient than the solar panels (for obvious reasons), so that’s not the limiting factor.

As to power, Caltech is already at 2-3 pounds per square meter: https://magazine.caltech.edu/post/sspp-space-solar-power-pro...

> Another way to think about it: An SSPP spacecraft with a 60-meter-by-60-meter surface area made using today’s space PV-cell technology would cost $36 million and weigh nearly 9,000 pounds, or almost as much as a Ford F-450 truck. With the ultra-lightweight PV-cell technology Atwater envisions, it would cost just $450,000 and weigh about 300 pounds, or about as much as an IKEA three-seat sofa

That’s megawatt-level solar power under 5 tons using today’s leading edge technology. Starship super heavy can launch 100 tons into LEO.

As to bandwidth, Starlink V3 backhaul capacity is 1 terabit. Microwave radio frequencies have an insane amount of bandwidth.

[OneDonOne]

The Caltech Concept is just that — a concept. No prototype, no tests, no manufacturing, no results. When they achieve this order of magnitude improvement on a prototype scale, that's when we should take them seriously.

[rayiner]

I'm referring to the description of "today’s space PV-cell technology" (I believe the lightweight panels used to upgrade the ISS in 2021).

SSPP has already sent prototypes to space: https://www.caltech.edu/about/news/space-solar-power-project....

There's a ton of work being done on lightweight solar panels for space. E.g. https://ascentsolar.com/asti-technology-and-unique-advantage...

[selimthegrim]

I have a lot of respect for Atwater but I'm pretty sure a lot of people at Caltech think this is donor driven research.

[sfifs]

ISS today generates and radiates away about 120 KW of energy with its old tech 3250 sq m of panels panels and it's current radiators. That's what 3 H100 racks need

There may be an economic challenge - which seems to be the sort of problem mass manufacturing can solve very well.

There may be a compute model & latency problem, how do you organize model training when racks are much further apart than in traditional data centres (although speed of light is 50% faster in vacuum than glass fibre). But that's algorithms.

Relative to everything else in orbit, powering a rack of compute and some comms per satellite seems not really to be a physics problem.

[pfdietz]

What are the assumptions behind that 1.1 m^2 figure?

[alephnerd]

OP is quoting Mikhail Klassen at Planet Labs [0].

The cost of replacement is exorbitant for commercial usecases, but is acceptable for defense usecases.

The issue is too many people are looking at the commercial usecase while ignoring the defense usecase that is what is actually driving the conversation and dealflow in this segment.

[0] - https://www.mikhailklassen.com/posts/orbital-data-centers/or...

[pfdietz]

Ok, there are at least two bad assumptions there.

First, it assumes the radiator is at the same temperature as the GPU. But radiators become dramatically more effective as temperature increases, with radiated power increasing as the fourth power of absolute temperature. So a heat pump that drives the radiator at higher temperature could make the radiator far smaller. More power would be required (and the radiator would have to radiate this energy too) but the radiator could become much smaller.

The other problem is assuming the radiator is intercepting sunlight. But it can be shaded by reflective films or kept edge-on to the Sun.

[OneDonOne]

> Ok, there are at least two bad assumptions there.

> First, it assumes the radiator is at the same temperature as the GPU. But radiators become dramatically more effective as temperature increases, with radiated power increasing as the fourth power of absolute temperature. So a heat pump that drives the radiator at higher temperature could make the radiator far smaller.

Carnot's theorem and refridgeration cycles rear their head. Your heat pump still needs power to pump heat uphill., negating any savings from a smaller radiator. And what happens when you shrink a radiator? It becomes a high temperature radiator, meaning the GPU must operate outside its operational tmperature of about 70 degrees C. So small radiator = big pump + extra solar panels and batteries + dead GPU

Also, how do you deal with the added cost and complexity for the solar tracking mechanisms for the "edge-on-to-the-Sun" radiator?

[laughing_man]

>Also, how do you deal with the added cost and complexity for the solar tracking mechanisms for the "edge-on-to-the-Sun" radiator?

Almost every satellite needs attitude control, so this isn't something out of the ordinary. I'm not sure solar tracking mechanisms really fall into the "expensive" or "complex" categories in 2026.

[pantalaimon]

Why do you need batteries? Why do you think the solar panels to run the heat pump require as much space as you save on the radiator?

And GPUs operate just fine at 90°C

[pfdietz]

It's not just space, it's mass. A solar panel can be made very low in mass. The physical limit comes from the absorption of light in a thin layer of semiconductor. For CdTe, this would be about 1 micron. PV in space could be gossamer thin sheets of thin film semiconductors, with tremendous power/mass.

[pfdietz]

> Carnot's theorem and refridgeration cycles rear their head. Your heat pump still needs power to pump heat uphill., negating any savings from a smaller radiator. And what happens when you shrink a radiator? It becomes a high temperature radiator, meaning the GPU must operate outside its operational tmperature of about 70 degrees C.

Sorry, that's all nonsense. Yes, the extra power needs to be radiated. But the advantage of operating at high temperature is so extreme that the more effective radiation will overwhelm that unless the heat pump is extraordinarily inefficient. If the heat pump would be perfect, operating at the Carnot limit, then if it doubled the absolute radiator temperature it would double the amount of energy to be radiated, but the area of the radiator would decrease by a factor of (2^4)/(2) = 8.

As for the second point, no, this does not require the GPU to operate at higher temperature. What made you conclude it would?

> Also, how do you deal with the added cost and complexity for the solar tracking mechanisms for the "edge-on-to-the-Sun" radiator?

I don't need to; I'm just debunking a bad argument. What you are doing there is called "moving the goalposts". But satellites normally have means of orienting PV toward the Sun. So, maybe have the radiator perpendicular to those? Those claiming the idea violates the laws of physics and using solar absoption on the radiator as part of the argument need to show no such scheme can work, even in principle.

[OneDonOne]

Again,

Smaller radiator = bigger heat pump + more batteries and solar panels

Satellites have a mass and power budgets. Your scheme only looks at temperature. If you want to build infinitely large AI Data satellites, go ahead.

[foobarian]

The other neat thing is that in space, the background temperatures are a lot lower than Earth due to lack of atmosphere, and available potentially more often than 1/2 the day cycle depending how high you're prepared to orbit.

[pfdietz]

A reflector could let the radiator see dark sky all around, even in low orbit. There are drag concerns.