cross-posted from: https://piefed.world/c/twitterblueskymastodon/p/1124740/fancy-label-for-old-infrastructure
why did they call them “ai datacenters” when they could have called them “slopping malls”
You must log in or # to comment.
- Waterburners
- fooling stations
- halleucinariums
While I agree with the sentiment, I do not think that they are deserving of a cute name. It kinda detracts from the harm they cause.
“Ponzi scheme” or “fraud” - as many are owned by skeezy shell corps.
Example: One proposed in rural Colorado is owned by a fake corp where one guy is the entire board of directors, runs in the red, and receives millions of dollar “loans” with no loan terms from anonymous donors.
Oligarch propaganda centers
“Money vortex” is too vague.
What about “Hype Factories”?
Because “AI Datacenter” is more palatable than “Surveillance Datacenter”.
The goal here is a reimagining of the financial system and surveillance infrastructure while calling it AI for marketing purposes.
what’s the difference between a datacentre and an AI datacentre
One stores useful information.
Do “AI datacentres” actually exclusively run AI workloads, or are they generic datacentres where a bunch of the servers have GPUs?
From what I know yes. A regular data center has data storage. AI data centers are just GPU and CPU with minimal memory capacity. Though, I could be mistaken.
Memory capacity depends on the application, but all the AI servers I’ve seen are full of memory. Shared storage isn’t always too big, but again that’s application dependent.
AI datacenters are those that are being built with the trillions of fake dollars that the AI companies are “spending” and “investing” in each other
Power density. Traditional data centers used to have cabinets that could handle 5kW to 10kW each. A.I. data centers have cabinets that are pushing 120kW each.
You need specific hardware for AI datacenters, like the latest NVIDIA GPUs
Specifically you need hardware with tons of high bandwidth memory that can be accessed by tons of very parallellised compute (so GPUs, though they’re only GPUs in architecture, they’re not actually that great for graphics), where single thread speed doesn’t matter at all.
For normal workloads this hardware wouldn’t be very efficient, you don’t need usually as much memory bandwidth usually and since not every problem is infinitely parallellisable, you’d want more single thread speed too. You’d get much better performance out of regular server CPUs for most tasks.
