Joe Brockmeier

RE: https://mastodon.green/@gerrymcgovern/116334729524665446

I'm just a touch skeptical about the claims here. I can believe that large data centers may raise temperatures around them, but the claim about raising nearby temperatures up to 16.4F and that the changes could be felt up to 3.3 miles away.

To be clear, I'm not questioning the concept that data centers may raise temperatures -- I'm just a bit skeptical about the, no pun intended, degree of the changes.

However, I do think we should be hard applying the brakes on building new data centers, etc., to study their environmental impacts before just blindly building out more.

Apr 2 at 3:03pmWeb
Show threadalpha1beta1d ago

@jzb Anyone someone uses "Up to" in a statement like that, I assume that the max case is one instance.

In this case, they don't even say they found it, they just "calculated it"

Show threadChuckles1d ago
@jzb @gerrymcgovern heat pollution does not distribute evenly: it follows prevailing winds and currents, resulting in literal temperature hotspots.
Show threadtschenkel1d ago

@jzb

I think the confusion with these numbers comes from assuming that it will be only the heat output of the data centre that causes the heating, but there are more effects at play than just the heat from compute and cooling systems:

1. the electrical energy input ia directly converted to heat that is dumped into the environment (both from compute and auxiliary systems). Total P_electric = Qdot_heat. This is worse if the data centre uses gas turbines to cover peak demand, as some do!

2. the cooling system will likely be evaporation cooling, i.e. it will introduce water vapor into the air, which adds additional thermal capacity to the air.

3. sealed (concrete) ground and buildings cause a heat island effect (also happens in towns, green field shopping malls, etc.) because which is a well understood effect and has been shown that it can be double digit degrees temperature difference between built up and surrounding green belt areas.

So, overall, the quoted numbers are very realistic from a thermodynamics analysis (and - to me, who teaches thermodynamics - not surprising at all).

Show threadJoe Brockmeier1d ago

@tschenkel Thanks so much for the reply - very much *not* my area of expertise.

Which is not to say I actually have any areas in which I am an expert, possibly excepting music trivia and cats, but this one can be ruled out entirely...

Show threadNina Kalinina1d ago
@jzb this is indeed an interesting question. Consider this: Russian city of Ekaterinburg (pop 1.5M, 1100 sq km, harsh winters) uses city-wide heating with the capacity of about 6800 Gcal/h if I am to believe the Wikipedia. That is roughly 7.9 GW. The largest DC in Europe back in 2024 was 1.2 GW (all of which goes into heat). I'm not sure if the heat from a DC is indeed enough to raise the temperatures around the DC by a large factor, but it surely is enough to keep tens of thousands of European homes warm, so the claim from the article/the paper doesn't sound too outlandish to me.
Show threadNina Kalinina1d ago
@jzb there is the paper available that seemingly only evaluates existing data, and it looks very alarming: https://www.researchgate.net/publication/403073048_The_data_heat_island_effect_quantifying_the_impact_of_AI_data_centers_in_a_warming_world
Show threadJoe Brockmeier1d ago

@nina_kali_nina Thanks. I will take a look at that paper.

I hope I conveyed well enough that I'm not trying to refute the study, etc., but I'm just raising an eyebrow about the magnitude. A little bit.

This is just one more objection to the rampaging, blind adoption of LLMs...

Show threadWulfy—Speaker to the machines1d ago

@jzb

Every time I actually fact checked these "WAAAH DATACENTRES" actual numbers, and I've done this three times already, this is what I found;

1. The numbers include estimates of future loads

2. The numbers purposefully conflate the #Ai #datacenters with the existing Datacentres, "the cloud" your #AWS , Facebook, Dropbox, Netflix, iCloud etc and yes your #Mastodon servers. Which at the last check is about 50/50 but it's hard to get real data.

3. The water consumption is very poorly presented often conflating water use with water destruction

4. The numbers often pick a comparison that sounds terrible but in the global scale is fuck all.
Like "AI uses as much water as the city of New York!!!!!!!!!!"

Having said that the trend of #broligarchs just slapping down methane emergency generators on the perimeter running 24/7 is just criminal.

TLDR; The "WAAAH DATACENTRES" "journalism" really depends on the personal bias of the author