Nvidia AI Servers Reduce Data Center Water Use

Nvidia announced last week that its newest AI servers, built around its Vera Rubin platform, can significantly reduce on-site water use. This development addresses a growing concern about the substantial water consumption required to cool the massive computing facilities needed for artificial intelligence.

A large data center can consume up to 5 million gallons of water per day, a volume comparable to the daily usage of a town with tens of thousands of residents, according to the Environmental and Energy Study Institute (EESI). This water usage, alongside rising electricity demand and pressure on local grids, has fueled a backlash against AI infrastructure.

Nvidia's solution involves cooling systems that can operate at higher temperatures than conventional setups. The servers utilize a circulating fluid, a mixture of water and propylene glycol, which enters the system at 45 degrees Celsius (113 degrees Fahrenheit). This fluid circulates through cooling plates atop computing chips, including Nvidia's Rubin GPUs and Vera CPUs, absorbing excess heat. The coolant can reach up to 55 degrees Celsius (131 degrees Fahrenheit) before being cooled outdoors by dry coolers, which function as large radiator coils to dissipate heat outside the data center. This closed-loop system aims to minimize the need for evaporative cooling, a process that consumes large amounts of water.

The innovation lies not just in the adoption of liquid cooling but in the design that allows for higher operating temperatures. This approach is a step towards mitigating the environmental impact of AI infrastructure, particularly concerning water resources, as the demand for AI computing power continues to escalate globally.