pendell

Current Hyperfixation: Wizard of Oz

  • He/Him

I use outdated technology just for fun, listen to crappy music, and watch a lot of horror movies. Expect posts about These Things. I talk a lot.

Check tags like Star Trek Archive and Media Piracy to find things I share for others.



NireBryce
@NireBryce

open-loop water cooling of data centers really shouldn't be a thing. If you can't shed it another way, that should just be your DC cap, sorry. Water use has been a problem for years but I didn't realize so many were OPEN LOOP. (That is, evaporative cooling, but not in the heatpipe sense, instead in the "we evaporate the water into the atmosphere" sense.)

this isn't a defense of GPT/OpenAI: this is an indictment of most tech companies that do the same, but especially OpenAI

At least steal two stones with one bird: have treated water from sewage treatment plants cool them, closed loop, since they're going to empty into an open body of water anyway


pendell
@pendell

Datacenters should just be built with like a big underground swimming pool full of coolant that they have to reuse until it's reduced into nothing before they can get more and that should be a law or something I think


spiralingvoid
@spiralingvoid
This page's posts are visible only to users who are logged in.


You must log in to comment.

in reply to @NireBryce's post:

it's widespread unfortunately :C

but it's more the compute-per-query that's the worry, though I'm sure google might have similar these days, last time I tried to get numbers I couldn't find anything but that was '22

those are numbers you can't even look up as an engineer, they are business and accounting secrets. attributing the necessity of all the massive grinding mill of ever-running background jobs to each marginal usage of a thousand different end user products is a fraught proposition at best. so basically in order to create a new estimate you have to have executive approval for access to that data and i have a suspicion those guys don't want those numbers getting out

i mean ffs google's ability to expand their resources has often been limited by the availability of raw materials, like the metals hard drives are made of. it's a huge enterprise and a massive slice of the expenditure is prospective rather than necessary, i don't foresee them being willing to disclose their impacts any time soon

though I do wonder if the IR data mentioned in the preprint can be drone-automated.

argh now i'm imagining them escalating the arms race to using geothermal to have waste heat to vent to throw off IR business intelligence along the lines of "yeah lets just run 500,000[,000] jobs of known compute and see how much the datacenter vents".

unfortunately that would require them to have a reason to give a fuck, which currently they do not

edit: also if they're going at that they would just do a better thing to begin with and dump heat into inactive geo tubes. currently they use open loop evaporation because it's cheaper than drilling enough tubes to sink that heat into the ground.

idk maybe there are other scaling concerns with a DC sized power load but i know that those systems are basically inexhaustible thermal sinks when it comes to heat pump systems

yeah, though at least it's not... the worst form of open cycle, which radio towers have, where it just sprays radiators with water and points a fan at it. but that's emergency, i think.

but yeah, so much of all of these companies is just about finding subsidized resources and then burning them at 400x the rate the subsidy was meant for

right, and specifically i don't think there is any reason for businesses to consider researchers' revelations of their waste to be a threat right now. i'm not aware of anyone actually enforcing that in any way except vaguely, at the market level