Mastodawn

Coach Pāṇini ®Jul 30, 2023

@clive https://mastodon.world/@paninid/110804712051678298

Sampath de Pāṇini ® ✅ (@paninid@mastodon.world)

Attached: 2 images I can’t resist. This is a fun game now. Turtles all the way down. #AI #satire #creativity

Mastodon

@paninid

😂

DanaBlankenhorn Jul 30, 2023

@clive Many hyperscale data centers recycle the heat generated in their operation. In Georgia, Google uses it for water purification.

@DanaBlankenhorn

Cogeneration is super cool!

Rolling it out a lot more would help for sure

DanaBlankenhorn Jul 30, 2023

@clive Cloud was all about saving money, in any way possible. Cogeneration was one way.

Mike Fraser

Jul 30, 2023

@clive Are most data centres not closed loop chilling? And you're using an evap cooling tower are you not just sending fresh water back into the atmosphere? Are we looking for a problem that's not there?

@mike

Nah, read the paper ... closed loop is part of the heat exchange getting the heat to the cooling towers; the cooling towers evaporate, and thus consume, water

Eric R. Scott Jul 30, 2023

@mike @clive this is like saying that watering lawns or flushing the toilet doesn't consume water.

Mike Fraser

Jul 30, 2023

@LeafyEricScott @clive This isn't a hill I need to die on, but flushing your toilet is dumping fresh water into a puluted sewer. Watering your lawn is dumping water directly on to the ground at a much higher rate than evaporation, and possibly into a polluted water table. I get what everyone's saying but the analogies seem weak.

Michael Downey 🧢Jul 30, 2023

@clive

Thank you for this work and helping raise the alarm.

Google’s water use is soaring in The Dalles, records show, with two more data centers to come

The company's data centers used 355 million gallons of the small city's water last year.

oregonlive

@downey

Yes! Those stats -- Google's use of water in The Dalles tripling over five years -- are pretty intense

What's more, the fact that the newspaper had to *sue* the *city* to compel the city to release those figures illustrates one of the big problems in grappling with the water use of data centers ...

If cities seriously priced their scarce freshwater at what it's worth, it'd give tech firms a reason to be more efficient

But cities don't

Michael Downey 🧢Jul 30, 2023

@clive

"but the jobz"....

https://iv.melmac.space/watch?v=taJ4MFCxiuo&local=true

The Simpsons - Monorail Town Hall Meeting & Song

Taken from Season 4 Episode 12 Episode Name: Marge Vs. The Monorail Visit or subscribe to my YouTube Channel for other high quality classic Simpsons clips.

BBB Q | Invidious

@downey

ugh

Eddie D Jul 31, 2023

@downey @clive your local government isn't working for u if they spend over a year keeping data secret.

@ed_donofrio @downey

Very true

Erik Jonker Jul 31, 2023

@clive …but compared to the water footprint of a hamburger it’s not that bad ?

@ErikJonker

That’s one way of looking at it, yes!

Or — speaking as someone who eats hamburgers — hamburgers are even more unsustainable uses of freshwater, as plenty of folks have been warming us for years

Riley S. Faelan Jul 31, 2023

@clive: Er, what's your source on Microsoft using fresh water up for cooling?

Most types of industrial water cooling circulate their transfer liquid instead of passing it through. The main exception is a Google data center in Finland which is specifically built to take advantage of a convenient source of sea water (it's the Baltic Sea, so not very salty, but distinctly not fresh, either) flow nearby.

Peter Kraus Jul 31, 2023

@clive how much water would asking the same number of questions of a "normal" search engine require? 500 ml sounds like much, but without a reference value it's a bit hard to judge.

@pkraus

Good question! I didn’t see that figure in the paper — either they didn’t include it or I missed it

My *suspicion*, for which I have no evidence, so take it with a grain of salt, is that inferencing with a model is much more computationally intensive than a traditional search, if only because the latter benefits from two decades of engineering and computer science that has chased efficiencies at scale

Samuel Leuenberger Jul 31, 2023

@clive I intuitively found this figure overblown, so I dug in your source and I see that the authors are basing all their calculations (3.8L/kWh) on a figure in a WSJ article!
Is that serious ? I wouldn't trust a paper whose authors do not even bother providing a citation to a proper study, especially when their core statement is based on this source.
I do hope that the WSJ article has a citation itself and would appreciate if someone with an access could provide it.

Dźwiedziu Jul 31, 2023

@Leuenberg
AFAIU they can't get better figures, because there is no publicly available data.

Still the numbers can differ, but the problem remains.

@clive

@clive That estimate says your queries are using .3 kWh of electricity or so, so it is about 2 cents of power and the water that gets boiled by that (since effectively all energy consumed by computers ends up as heat).

But I don't believe it, assuming 10s of runtime, that means your query used 180 servers for that 10 seconds. LLMs are hungry but not THAT hungry.

jwatt Jul 31, 2023

@ncweaver @clive @clive This is good work. Another issue, besides power usage, is water usage, and I'm not sure if it's with the article or the underlying research: seems a little overly simplistic to assume all water is fresh water, doesn't it? There's a lot of room for optimization there, and it seems like they'd have incentives to optimize.

@jwatt @clive
It depends. Data centers on a river or ocean will direct heat-exchange. Other than that you use evaporation to dump heat, all AC type systems on any scale do ("cooling towers").

However, the water use is in the noise compared to say agricultural use for alfalfa. A single acre of alfalfa will use 4 acre-feet/year, so 5 MILLION liters of water to grow cattle feed (that is commonly shipped to Saudi Arabia where growing it is banned).

@jwatt @clive
To get a sense of scale, the Imperial Valley in CA (which is absolute desert) has 130,000 acres of alfalfa planted! If you care about water use, care about agriculture, and if you care about agricultural use, alfalfa is a god-damned ecological crime.

@ncweaver @jwatt

Yep — and almonds are right up there too

jwatt Jul 31, 2023

@ncweaver @clive guessing alfalfa is a lot denser or cheaper than grass? I know I have to pay a lot more for grass-fed beef and milk. Do those have better water footprints, too?

@jwatt
Grass fed beef means you just let the cows lose on open land. Alfalfa is a hay/forrage crop, so used to feed cattle when you have too many/too dense. Mostly dairy cattle, rather than beef cattle.

jwatt Jul 31, 2023

@ncweaver you seem knowledgeable about this stuff: If we cut out alfalfa or did something else, how much would it effect the beef availability? I’d say maybe we’d just lose fast food and lower quality meat, but a lot of people depend on those as inexpensive and healthy sources of fats and proteins. In the rural US, buying a cow from a farmer and getting it processed can feed a family for a long long time—beef seems like it could be a really efficient source from that perspective.

@jwatt Beef is absolutely a nightmare from a food conversion standpoint. You use beef where you have rangeland, grass that you aren't growing organized crops, as it can convert cellulose to protein.

Vegetable proteins (beans etc) are an order of magnitude cheaper & more efficient. For animal protein, fish. Lots and lots and lots of farmed salmon is incredibly efficient food conversion and really high quality protein.

Dźwiedziu Jul 31, 2023

@clive
Oh great, *another* way that AI is killing us.

But it's not the AI that is facilitating the killing…

/me turns dramatically to the camera.