liquid cooling
In plain wordsCooling computer chips by running liquid past them instead of blowing air. AI chips now run so hot and are packed so tightly that air alone can no longer keep them cool.
Removal of heat from computing hardware by a circulating liquid, which carries heat far more effectively than air. In direct-to-chip (cold-plate) cooling, the dominant form for AI, coolant flows through metal plates mounted on processors and memory and returns to a coolant distribution unit, which passes the heat to facility water and then to cooling towers or dry coolers; rear-door heat exchangers instead cool a rack's exhaust air, and immersion cooling submerges whole servers in a dielectric fluid, either single-phase or two-phase (boiling). With accelerators drawing from about 700 W to well over 1,000 W each and racks exceeding 100 kW, systems such as Nvidia's GB200 NVL72 are designed for liquid cooling only. Benefits include higher density, less fan power and better power usage effectivenesspower usage effectiveness, n.: A score showing how much of a data center's electricity actually reaches its computers: total power drawn divided by the computers' share. A perfect 1.0 would mean no waste; the extra goes mainly to cooling.; costs include plumbing, leak risk and the expense of retrofitting older facilities. Depending on design, liquid cooling can raise or lower a site's water consumption.
The old hall tops out at 15 kW a rack on air, so the Blackwell racks are going into the new liquid-cooled building.
Origin
Water-cooled mainframes such as IBM's 3081 (1980) and the Cray-2 supercomputer (1985), whose circuits were immersed in Fluorinert, preceded decades of air-cooled data centers. AI rack densities brought liquid back from about 2023, and Nvidia's GB200 NVL72 (2024) made it standard for frontier AI hardware.
Related
- AI data centerAI data center, n.: A building or campus packed with AI chips, power equipment and cooling, built to train and run AI models. The biggest use as much electricity as a city.
- rack-scale systemrack-scale system, n.: An AI computer designed as one whole cabinet of chips, cooling and wiring, rather than separate servers, so all the chips inside can work together like one big chip.
- power usage effectivenesspower usage effectiveness, n.: A score showing how much of a data center's electricity actually reaches its computers: total power drawn divided by the computers' share. A perfect 1.0 would mean no waste; the extra goes mainly to cooling.
- environmental impact of AIenvironmental impact of AI, n.: The effects of AI on the planet: the electricity, water, materials and land used to make chips, build data centers and train and run models, and the pollution that results.
- GPUGPU, abbr.: A computer chip first designed to draw video-game graphics that turned out to be ideal for AI, because it can do thousands of simple calculations at the same time.
Mentioned in
AI data centerAI data center, n.: A building or campus packed with AI chips, power equipment and cooling, built to train and run AI models. The biggest use as much electricity as a city., orbital data centerorbital data center, n.: A proposed computer center circling the Earth, powered by almost constant sunlight, that would run AI without using land, water or power grids on the ground., power usage effectivenesspower usage effectiveness, n.: A score showing how much of a data center's electricity actually reaches its computers: total power drawn divided by the computers' share. A perfect 1.0 would mean no waste; the extra goes mainly to cooling., rack-scale systemrack-scale system, n.: An AI computer designed as one whole cabinet of chips, cooling and wiring, rather than separate servers, so all the chips inside can work together like one big chip.