web page content

The high cost of rad-hard chips is basically due to amortization over very small volumes. It’s not like you need isotopically pure silicon.

Cheers

Phil Hobbs

A data center in space would ideally never be eclipsed by Earth, and would be geostationary, and be in a low orbit, and never be above clouds.

The whole AI-in-space idea is silly for many reasons.

A fiber bundle the size of a pencil will carrya million times more data than the entire usable RF spectrum DC to UV. Maybe a billion times.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

Geostationary and in low orbit is a contradiction in terms.

And your objections are terminally silly. If you objected in less specific terms you might have a faint chance of being taken seriously.

Do show your calculations. We need more comedy here.

Silicon has three stable isotopes

Agreed. Larger volume will probably reduce costs. Whether a 1300x price drop is possible seems unlikely.

I don't know much about space hardened technology, so I did some random online reading. I was expecting to find fairly modern versions of common Intel and AMD CPU's. Nope. Common CPU's seem to based on

For RF, an orbital data center might need resurrect tunnel diodes, which are quite resistant to ionizing radiation and therefore quite reliable.

China might have the right idea: "China Turns on the World’s First Underwater Data Center"

The (un)popular press keeps making stupid claims like

"According to a report published this week by the United Nations University Institute for Water, Environment and Health, the water consumption of data centers could reach 9.3 trillion liters by 2030, which is the equivalent to the water needs of all of sub-Saharan Africa."

Data centers don't use water, they just dump heat into it. And salty or dirty water works too.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

As usual, I beg to differ. Most data centers do consume large amounts of cooling water. Some details:

"Myths vs. Reality: Data Centers And Water Usage"

"Evaporative cooling (open-loop): A majority of large, modern data centers use water-based cooling for better energy efficiency. This often involves cooling towers or evaporative chillers: warm water absorbs heat from servers and is then cooled by evaporation in a tower. As water evaporates into the air, it carries away heat - dramatically cutting the electrical power needed for cooling. The trade-off is high water consumption. Most big data centers today use some form of evaporative cooling because it’s energy-efficient, especially in hot climates, but it directly uses water (often drawn from municipal supply)."

"Among the water-cooled facilities, the vast majority of large-scale data centers use open-loop evaporative cooling meaning they evaporate water as part of the cooling process. This has been the standard because it’s effective and energy-saving, but it does consume water."

"Typically, 70-80% of the water in evaporative cooling is lost as evaporation into the air. The remaining 20-30% is discharged as liquid wastewater (which goes to a sewer or treatment plant)."

"Loudoun County, Virginia - the world’s largest data center hub - supplied around +/-1 billion gallons of water to data centers in 2023, mostly relying on treated potable water because reclaimed water capacity was insufficient."

Cooling towers only make sense if there is lots of clean water available. I there isn't, don't do it.

Google AI tells me that if one were to dump a gigawatt of heat into the Mississippi river, the water temp would go up 20 microkelvins.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

My apologies. I missed some relevant projects that appeared in a video that I watched last night:

"Big Tech Wants To Build Data Centers In Space: Does This Make Sense?"

1. The video mentions Starcloud: "Starcloud launched its first test satellite, designated Starcloud-1, equipped with a Nvidia H100 GPU which Nvidia claimed was 100x more powerful GPU compute than had been in orbit before. The company described the mission as the first deployment of "data-center-class GPU compute" in orbit."

For high speed data between data centers, they propose using Space-X Starlink Mini Lasers: "Starcloud orders Starlink lasers for orbital data center network"

1. Google has Project Suncatcher: "Exploring a space-based, scalable AI infrastructure system design" <https://research.google/blog/exploring-a-space-based-scalable-ai-infrastructure-system-design/>
   formatting link
2. Axiom: "Axiom Space Partners with Kepler Space and Skyloom to Operationalize the World’s 1st Orbital Data Center"
   formatting link

The author also mentions that doing AI training in an orbital data center won't work because of the currently limited communications bandwidth.

I thought you wanted to use reclaimed water? Reclaimed water is hardly "clean" as it's usually full of dissolved minerals. If the cooling tower vents water to improve heat extraction, there's going to be some messy maintenance removing limescale and biofilm. Perhaps de-ionized or distilled water? If it's a closed system, that might work.

Your model assumes uniform hot water dispersion. It takes a while for the heated water to circulate and settle to a uniform temperature. Fast moving water is probably best.

Dumping gigawatts of heat into a river ecosystem is likely to produce dead fish and attract protesters, media attention and attorneys. Looking at drone photos of data centers, very few are near rivers.

Maybe the recently minted data center billionaires could be convinced to provide the city with free hot water? Just turn the tap, and you have hot water. No need for a water heater or solar panels. It could also be used for hydronic winter wall and floor heating.

I guess 20 uK terrifies some people.

Moscow has central hot water. I helped them install the flowmeters on a big hotel.

Most is unmetered. When people get too hot, they open a window.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

I guess not everyone has a Mississippi to dump it into.

By the way, the Mississippi river has an average flow rate of 340 m^3/s. Dumping 1 GW into that should raise the temperature by 700 mK, not 20 uK.

Jeroen Belleman

That’s an absurd underestimate. According to

Cheers

Phil Hobbs

At New Orleans, it's a mile wide and runs FAST.

Google AI gave me the 20 uK value.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

That's popular in Europe. It's called district heating.

It's 20uK after the heat has gotten distributed across the whole river. It's the local hot spots in the distribution process that create the problems.

The Russians didn't invest a lot in their public services. They hadn't got a lot of money to invest.

Am 16.06.26 um 02:37 schrieb john larkin:

That the answer google gave to me:

--------------------------------------------------

The Mississippi River discharges an annual average of about 593,000 cubic feet per second (16,800 m³/s) into the Gulf of Mexico. However, the flow rate varies drastically depending on the location along its

--------------------------------------------------

But then, at N.Orl., you could just as well use the Gulf of Mexico, if it still does exist. I heard the Mississippi was burning already in the 60s.

Gerhard

I got this from some AI that DuckDuckGo puts up. Granted, I should have double-checked that. Wikipedia says the yearly average varies from 6000 to 20000 m^3/s. Dumping 1 GW into a river with a flow rate of 6000 m^3/s begets a temperature rise of 40 mK.

It's always possible to choose the numbers depending on the point one wants to make.

Jeroen Belleman

Especially when you pick them 200 times too low. ;)

Cheers

Phil Hobbs