Recent submissions

Comment · 1 month ago
Comment · 3 months ago
I used to work for a guy who said at that company we made stuff that "looked like the soviets designed it in the 70s"
Comment · 3 months ago
Comment · 4 months ago
>I bought my son a bigger 3D printer and told him to stop playing with that boy.

This is the way

Comment · 4 months ago
This is a common problem I see. Trying to solve these problems in a vacuum. He makes a joke about that in the video.

This talk will help put it in perspective: https://californiaconsultants.org/event/energy-and-thermal-m...

In order to talk about these things we need to think about the power from the grid, all the way to the cooling water coming out. Not just talk about how we can calculate or optimize each little piece. Of course we can do that, it's the overall system we are talking about and need to understand.

Forest for the trees.

What do I mean by that here?

He does the calculations for 20 kW. Which sounds like a lot. It's only enough to cool one node, not one rack, one node, in a modern data center. The starlink satellite he referenced is viewed as the absolute most modern thermal design in a satellite.

Most data centers have 2000-5000 nodes and the hyperscalers have 100,000 or more.

So to replace a single data center we need 2000-5000 of these things up there, at a minimum, or one thing that is 2000-5000 times bigger.

And maintenance.

And the hardware gets obsoleted every few years.

Or you could just put it in the desert in Nevada. But we don't need rockets to get there.

Live from the official Hacker News API

A modern reader for Hacker News · SvelteKit + Skeleton