NVIDIA shifts AI server cooling to hotter liquid loops

NVIDIA’s Rubin generation of AI infrastructure is built for 100% liquid cooling, using closed-loop liquid systems for every chip and networking component instead of fans. The design allows cooling liquid to run at up to 45 degrees Celsius, or 113 degrees Fahrenheit, which helps facilities reject heat more efficiently and reduces dependence on chilled air.

Cooling has historically accounted for up to 40% of a data center’s electricity consumption, and industry estimates suggest that raising chiller plant temperatures by just one degree can cut cooling energy costs by about 4%. NVIDIA’s approach is designed to support dry-cooler-based operation in favorable climates, cutting facility cooling water consumption from roughly 2.6 million gallons per megawatt per year for conventional cooling-tower-based systems to near zero, or up to a 100% reduction in water use.

The Rubin architecture also changes the physical design of AI server rooms. Coolant made of 75% water and 25% propylene glycol flows through cold plates directly on processors, reducing the need for cold aisles, perforated server fronts and fan noise that can reach 85 decibels. Fully liquid-cooled servers can also increase rack density, with a system that previously occupied six rack units now fitting in two.