Liquid Cooling Exhibits(chipsandcheese.com)
chipsandcheese.com
Liquid Cooling Exhibits
https://chipsandcheese.com/p/liquid-cooling-exhibits-at-hot-chips
8 comments
These are some very innovative ideas and they totally and completely side step the need of AI for fresh water, leakages and alternative coolants.
They definitely don't. The water consumption of AI is not happening in the cooling blocks in the racks, it's happening outside in cooling towers. The heat a data center produces must be removed from the data center, and that is typically done via evaporation. The hot water interfaces with evaporative cooling towers and moves back inside to cool the chips again.
It would be neat if we could harvest that heat and use it for some purpose, but today we largely use it to make steam.
It would be neat if we could harvest that heat and use it for some purpose, but today we largely use it to make steam.
The issue with harvesting the waste heat is the temperature of it (low.) I commented about this recently: https://news.ycombinator.com/item?id=45007358
Totally. I'm imagining something like heating for residential or greenhouses. Or a preheated input to some other heating process. One could presumably also spend energy to concentrate it a bit, but it's never going to turn turbines.
Still, the mind wonders -- here's a pocket of relatively warm, surely we could do something with that waste?
Still, the mind wonders -- here's a pocket of relatively warm, surely we could do something with that waste?
It can be done without water it's just marginally more energy intensive. Traditional dry coolers are perfect for these temps.
No they don’t, if you remove evaporative cooling and go to a closed loop system, the coefficient of performance is almost cut in half. The people building data centers will almost invariably choose evaporative cooling because opex is substantially cheaper.
Why is this downvoted? It's a simple fact of physics that's been fully understood for literal centuries, taught in high school [1]. For a given heat exchanger size, evaporative cooling will always win (well, unless it's raining), with drastically lower thermal resistance.
[1] https://en.wikipedia.org/wiki/Specific_heat_capacity
[1] https://en.wikipedia.org/wiki/Specific_heat_capacity
I assume people get upset when they’re told they are wrong, I don’t take it personally.
Thanks for seconding me :)
Thanks for seconding me :)