Synthetic intelligence (AI) is sizzling proper now. Additionally sizzling: the info facilities that energy the know-how. And maintaining these facilities cool requires an amazing quantity of power.
The issue is barely going to develop as high-powered AI-based computer systems and gadgets grow to be commonplace. That is why College of Missouri researcher Chanwoo Park is devising a brand new kind of cooling system that guarantees to dramatically scale back power calls for.
The work is printed within the journal Utilized Thermal Engineering.
“Cooling and chip manufacturing go hand-in-hand,” mentioned Park, a professor of mechanical and aerospace engineering within the Mizzou School of Engineering. “Without proper cooling, components overheat and fail. Energy-efficient data centers will be key to the future of AI computing.”
Fixing tomorrow’s drawback
Knowledge facilities are massive services filled with servers that include pc chips that retailer and course of information. They’re mainly large pc hubs that home the servers that run web sites, cell purposes and information from the cloud.
They’re additionally power-hungry. In 2022, information facilities used greater than 4% of all electrical energy within the U.S., with 40% of that power being spent to maintain gear cool. As demand on information facilities will increase, much more power shall be required.
Presently, information facilities are cooled with both air-moving followers or liquid that strikes warmth away from pc racks.
Park and his group are growing a two-phase cooling system designed to effectively dissipate warmth from server chips by way of section change, similar to boiling a liquid into vapor in a skinny, porous layer. The system can function passively with out consuming any power when much less cooling is required. Even in lively mode, the place a pump is used, it consumes solely a negligible quantity of power.
“The liquid goes in different directions and evaporates on a thin metal surface,” Park mentioned. “Using this boiling surface, we’re able to achieve very efficient heat transfer with low thermal resistance.”
The system additionally features a mechanical pump that’s activated to soak up extra warmth solely when wanted.
Early assessments present that two-phase cooling strategies drastically scale back the quantity of power wanted to maintain gear cool.
The group is now fabricating the cooling system—designed to simply join and disconnect inside server racks. Park hopes they will be in use throughout the coming decade simply as AI-powered computer systems grow to be mainstream.
“Eventually there will be limitations under current cooling systems, and that’s a problem,” Park mentioned. “We’re trying to get ahead of the curve and have something ready and available for the future of AI computing. This is a futuristic cooling system.”
Park’s work aligns with the objectives of the Heart for Power Innovation, a constructing being constructed on campus to permit interdisciplinary researchers to resolve challenges introduced by rising power issues and fast development in AI. The thought is to leverage superior know-how to optimize power manufacturing, storage and effectivity.
“The center will allow us to explore additional ideas and innovations around energy-efficient processes,” Park mentioned. “These are complex problems that require different areas of expertise. I look forward to future collaborations.”
Extra data:
Rohan Kokate et al, Experimental evaluation of subcooled circulation boiling in a microchannel evaporator of a pumped two-phase loop, Utilized Thermal Engineering (2024). DOI: 10.1016/j.applthermaleng.2024.123154
College of Missouri
Quotation:
A cool answer: Designing a next-generation cooling system to assist information facilities grow to be extra power environment friendly (2024, July 25)
retrieved 25 July 2024
from https://techxplore.com/information/2024-07-cool-solution-generation-cooling-centers.html
This doc is topic to copyright. Aside from any honest dealing for the aim of personal examine or analysis, no
half could also be reproduced with out the written permission. The content material is supplied for data functions solely.