@Npars01 @noellemitchell Sorry, that does not track with what the hyperscalars are actually doing. Training runs need high continuous base load and even coal doesn’t do that well.

In reality, GenAI requirements are turbocharging the drive to carbon-free nuclear

datacenterfrontier.com/cloud/a

@dragonsidedd @Npars01 @noellemitchell nuclear is slow to build. Gas is what’s being built, because their return on investment works on the scales of political cycles a industry can support.

And your argument for continuous loads is partially false. Training runs for the mega-models do last for months, but between runs there is great variability. And that’s where gas shines.

powermag.com/industry-exec-dat

@gimulnautti @dragonsidedd @Npars01 @noellemitchell nuclear is not only slow to build and mired in regulatory gates, but slow to roi. Profit does not manifest for sixteen *years* after the plants construction, fueling, and activation to to grid.

Overhead costs are enormous as well. Armed guards, safety crews, training etc...all add up.

Solar, wind and hydro offer turnkey profit in less than a year. Very attractive.
Follow

@Nimbius666 @gimulnautti @noellemitchell @Npars01 @dragonsidedd

"Too cheap to meter" nuclear is just a few years away... and has been for 60 years now.

@Npars01 @pieist @gimulnautti @noellemitchell @dragonsidedd the US also has a major waste issue. None of it is recycled, no state is willing to accept it, and most plants run a surplus of waste onsite.

To date there is no international consensus on how properly to label and warn a generation of humans 1000 years into the future of our waste pit hubris.

@BashStKid Hang on, the ability to create virtual realities entirely in my head in which I have superpowers isn't itself a superpower? Ffffff....

@Npars01 @pieist @Nimbius666 @gimulnautti @noellemitchell @dragonsidedd Nuclear waste isn’t as difficult a problem as people think. It can be buried deep enough that only a modern nation state could get near it, and if they do, by nearly astronomical odds, stumble on it, the tomb would be so conspicuous only fools wouldn’t check for all forms of safety.

The waste problem is entirely political.

@Npars01 @pieist @Nimbius666 @gimulnautti @noellemitchell @dragonsidedd That doesn’t solve the other issues. Our current nuclear energy strategy is an aging disaster. Solar, wind, and tidal are where we need to go right now. I don’t see the political will to rebuild our nuclear energy infrastructure coming any time soon.

@Extra_Special_Carbon @Npars01 @Nimbius666 @gimulnautti @noellemitchell @dragonsidedd

Solar power satellites are where it's at, baby. Can also be repurposed as a Starlink-zappers, for planetary demusking.

@pieist @Npars01 @Nimbius666 @gimulnautti @noellemitchell @dragonsidedd How do we get the power from there to here? We can’t beam 1.21 gigawatts from space to earth without some serious atmospheric side effects.

@Extra_Special_Carbon @Npars01 @Nimbius666 @gimulnautti @noellemitchell @dragonsidedd

Great Scott! (Whoa, this is heavy.)

There's a lot of debate on the topic, but purportedly at geostationary distances the maser beam spread would be sufficient enough that harms would be pretty small and mitigable ... perhaps even negligible compared to those of legacy energy sources. I'm not going to try to link a lot of resources, I'd just be googling for them same as you.

@pieist @Extra_Special_Carbon @Npars01 @gimulnautti @noellemitchell @dragonsidedd soviets tried this with solar illumination at night using a mirror system and it infuriated most of the science community 😆
@pieist @Extra_Special_Carbon @Npars01 @gimulnautti @noellemitchell @dragonsidedd correct me if I'm wrong but musks constellations are mostly hanging around for a month or less before decay error and deorbit.

Not a great product.

@Npars01 @pieist @Nimbius666 @gimulnautti @noellemitchell @dragonsidedd I’m neither pro nir anti nuclear. I think it could be done safely. It isn’t being done safely. It won’t be done safely any time soon, so we should move on until we get our act together.

@Extra_Special_Carbon @Npars01 @pieist @Nimbius666 @gimulnautti @noellemitchell
FWIW I am super bullish on high temperature superconductors for spherical tokamaks.

And a viable machine is coming online in <2 years

psfc.mit.edu/sparc

@dragonsidedd @Extra_Special_Carbon @Npars01 @pieist @Nimbius666 @gimulnautti @noellemitchell Can we wait until a technology exists before we count on it?

Also, non-real-time computing does not require a fixed-power energy source. Things like nuclear is totally unnecessary for training tasks. Train when you have power. Pause the training when you don’t.

And nuclear is not a stable source of energy. France had to turn off nuclear because of too-hot rivers, and emergency shutdowns are common.

@ahltorp @Extra_Special_Carbon @Npars01 @pieist @Nimbius666 @gimulnautti @noellemitchell

> Can we wait until a technology exists before we count on it?

You should learn about HTS. It is decades-old technology that has been gaining incrementally.

> nuclear is totally unnecessary for training tasks. Train when you have power. Pause the training when you don’t.

This would take orders of magnitude *more* power due to the save and read data required for suspend-and-restore.

@dragonsidedd @Extra_Special_Carbon @Npars01 @pieist @Nimbius666 @gimulnautti @noellemitchell I was not referring to HTS, I was referring to fusion reactors. Once they a) work, and b) have been tested a while, then they are relevant. Before then it’s just wishful thinking, which is no basis for a system of government.

As for pausing, I recommend basic power management literature. This is not the forum for a detailed discussion. I would write a longer text, but: sealioning.

@ahltorp @dragonsidedd @Extra_Special_Carbon @Npars01 @Nimbius666 @gimulnautti @noellemitchell

Speaking of which, if crocodiles are flourishing in French rivers thanks to nuclear power, just imagine how the sea lions must feel.

@ahltorp @Extra_Special_Carbon @Npars01 @pieist @Nimbius666 @gimulnautti @noellemitchell

Honest question:

What is your explanation for why the professionals who specialize in data canter design are putting their careers and companies' investments on the line with this specific solution?

And why does "baseline power load" always factor so highly in the relevant proposals?

@dragonsidedd @Extra_Special_Carbon @Npars01 @pieist @Nimbius666 @gimulnautti @noellemitchell Easy. Because they want the hardware to work constantly to increase ROI, and then they externalise the power network costs.

That should, however, not be society’s problem to solve for them.

@ahltorp @Extra_Special_Carbon @Npars01 @pieist @Nimbius666 @gimulnautti @noellemitchell
you must deal with some strange IT providers.

at the company I work for, increasing ROI at a datacenter involves *minimizing* its energy usage, while still being able to serve the customers' data

@Npars01 @ahltorp @Extra_Special_Carbon @pieist @Nimbius666 @gimulnautti @noellemitchell

I've been involved with data center planning in a large cloud company. Would you believe, Saudi Arabia's requirements have never come up, but base load considerations have.

Also if you want to "fry the planet", non-CO2-generating fission is counter to the end goal.

Fun fact: if your wastewater stream is hot, you are losing money and the executives probably want that to stop

Sign in to participate in the conversation
Qoto Mastodon

QOTO: Question Others to Teach Ourselves
An inclusive, Academic Freedom, instance
All cultures welcome.
Hate speech and harassment strictly forbidden.