Gasper Zejn

Human. Born into 344ppm. Most interested in the energy stuff. (Physics, not some made up things.)

Blog: https://zejn.net/b/

Last Notes

2025-10-09 11:29:02 GMT

- reply

AI datacenter buildout is driving up investment into grid, which drives up network costs,, which are (in most parts of the world) socialized over all the grid users.
Problem with network upgrades is that each user get his bills increased, even if they do not use more energy.

2025-10-09 10:54:59 GMT

- reply

Ember is also a good source of stats and analyses, but they're not on mastodon.
Here's their press release that Guardian is citing, I think, where they show that renewables have covered more than the increase in demand.
https://ember-energy.org/latest-updates/solar-and-wind-met-all-electricity-demand-growth-in-h1-2025/

2025-10-09 10:38:03 GMT

- reply

Then again, AI is additional load on grid, which was not expected and is indeed making a dent in the wrong way in the progress towards electrification and decarbonisation and this is something generally agreed upon.
5/5

2025-10-09 10:35:57 GMT

- reply

With ongoing electrification, for an european family home of 4, you can project an increase in electricity usage to about 300% current (+100% for heat pump and +100% for EV).
This alone will increase the peak grid power, and will need additional power plants built, to cover worst case scenarios.
A lot of this will be offset by batteries and solar, not because they're awesome (which they are), but because we can't build out the grid fast enough leaving ☀️ and 🔋 the only option.
4/...

2025-10-09 10:29:52 GMT

- reply

Also, different power plants produce different amounts of generation in relation to installed power.
Wind plants can't work with 100% whole year, they typically get between 25% and 40% of their nominal power (averaged over year). Hydro is between 30% and 40%, nuclear are 80+%, solar depends on geographical latitude, gas peaker plants are as low as 5%, and coal is all over the place.
It's this number that is dropping for coal power plans, as they can't compete on market.
3/...

2025-10-09 10:24:17 GMT

- reply

2. Security of supply is having enough dispatchable power to cover peak demand, in case other sources are not available. This means having fossil plants and fuel on standby, but most of the time not burning anything.
Even though there's more available power, the fossil plants are getting used less.
2/...

2025-10-09 10:20:21 GMT

- reply

Yeah, two very obvious counter arguments to your post are
1. you are comparing power with generation, and
2. you are ignoring security of supply.
1. Guardian talks about generation, which is what matters more, because that's the amount of energy used. That this has now surpassed fossil sources is the first milestone before the fossil use starts decreasing.
1/...

2025-10-09 09:58:17 GMT

- reply

Oh, man, oversimplified. I can't know if for illustrative purposes or just not knowing better?

2025-07-07 14:53:43 GMT

- reply

Yeah, but you don't have the extension or subprocess loophole. :)

2025-07-07 14:14:52 GMT

- reply

Or you could use network namespaces:
sudo ip netns add myns
sudo ip netns exec myns su -l anze -s /bin/bash -l
and then run the test in the given bash (or whichever shell you are using).
You can check ip addr, it should show only loopback device.

2024-08-29 06:45:31 GMT

- reply

I tend to make a cut based on the engineering vs. research.
If it's something you want to maintain - eg. a web app - you probably want to have it engineered.
If it's something from a researcher whose priority is not maintenance in production environment, but a proof of concept, he will not be an expert in python packaging.
Thus, if it's an engineered project, it should be installable. If it's a research project, a researcher will have a bunch of scripts in a folder.

2024-08-27 10:43:28 GMT

- reply

I think Django templates are known as not being the fastest due to some design limitations. This is basically the reason why Jinja exists.