huh, the cloud really did eat everything
-
huh, the cloud really did eat everything
-
huh, the cloud really did eat everything
"adiabatic assist" is a term for a type of cooling equipment that evaporates water to improve efficiency
"water-cooled chiller" being a type of cooling equipment that evaporates water to reject heat -
"adiabatic assist" is a term for a type of cooling equipment that evaporates water to improve efficiency
"water-cooled chiller" being a type of cooling equipment that evaporates water to reject heatafter reading the report to the best of my ability i can say that i don't plan to take anybody saying that the water use is obviously a problem at their word
-
after reading the report to the best of my ability i can say that i don't plan to take anybody saying that the water use is obviously a problem at their word
@whitequark Would be interested in your key take-aways here -- not sure I can make it through the whole report...
-
@whitequark Would be interested in your key take-aways here -- not sure I can make it through the whole report...
@chandlerc basically:
- to the extent AI datacenters come to exist, they are not clearly worse than existing cloud datacenters
- to the extent AI power use happens, it is not clearly worse to cool it with evaporation than with other methods and in fact it seems that it may be significantly better in places where you are not water-constrained
- even the most optimistic projections of AI expansion involve water use figures that are quite small compared to general industrial or ag water use -
@chandlerc basically:
- to the extent AI datacenters come to exist, they are not clearly worse than existing cloud datacenters
- to the extent AI power use happens, it is not clearly worse to cool it with evaporation than with other methods and in fact it seems that it may be significantly better in places where you are not water-constrained
- even the most optimistic projections of AI expansion involve water use figures that are quite small compared to general industrial or ag water use@chandlerc I think it is unquestionable and immediately obvious that power use is already bad and will get worse if only by asking any grid operator whether they are ready for this
the same does not appear to be the case for water use, and I think it strains credibility to keep using it with no further clarification as a catchy anti-AI slogan
-
@chandlerc I think it is unquestionable and immediately obvious that power use is already bad and will get worse if only by asking any grid operator whether they are ready for this
the same does not appear to be the case for water use, and I think it strains credibility to keep using it with no further clarification as a catchy anti-AI slogan
@whitequark thanks, that matches my take aways from reading other sources, and nice to that from another reader+source perspective. The power / grid side is definitely where it has seemed like the real problems lie here...
-
@whitequark thanks, that matches my take aways from reading other sources, and nice to that from another reader+source perspective. The power / grid side is definitely where it has seemed like the real problems lie here...
@chandlerc @whitequark the only thing with water is that it's availability is wildly variable. So it's possible to see large datacenters having negative impacts in certain locales if people were stupid enough to build them in places without abundant water. Which, well, it's a dice roll.
-
@chandlerc @whitequark the only thing with water is that it's availability is wildly variable. So it's possible to see large datacenters having negative impacts in certain locales if people were stupid enough to build them in places without abundant water. Which, well, it's a dice roll.
@dotstdy @whitequark Low-availability of water in locations seems at least reasonably likely to be priced in effectively, and one of the only things that seems likely to be a _consistent_ pressure on DC construction is operational economics.
Definitely will be some cases of super short-sighted sweetheart deals w/ local government combined with unethical DC operators, but even that seems _much_ more likely to occur in harmful ways due to electricity than water...
-
@dotstdy @whitequark Low-availability of water in locations seems at least reasonably likely to be priced in effectively, and one of the only things that seems likely to be a _consistent_ pressure on DC construction is operational economics.
Definitely will be some cases of super short-sighted sweetheart deals w/ local government combined with unethical DC operators, but even that seems _much_ more likely to occur in harmful ways due to electricity than water...
@chandlerc @whitequark well if you're doing a datacenter you can always roll in your own electrical supply. You know the "I'll just run portable gas generators permanently" musk bondoogle. Pricing is never effective for things as scarse as water though, as I think the history of ag irrigation shows, especially in such a bubble you can just pay any amount for the water and price everyone else out. But we'll see, I certainly agree it's not a universal concern.
-
undefined oblomov@sociale.network shared this topic on