Nvidia's AI Data Centers: Running Hotter to Slash Water Consumption

Nvidia asserts that transitioning to a fully liquid-cooled infrastructure and operating servers at elevated temperatures can reduce water consumption to "near zero."

Amid increasing public scrutiny regarding data centers' substantial water and energy demands, Nvidia is prominently showcasing its Rubin generation reference design. This blueprint for a fully liquid-cooled data center, according to the company, has "eliminated massive amounts of power usage and pretty much all water usage." However, this solution does not fully address broader concerns surrounding AI data centers, including their environmental impact during construction or the extensive power generation requirements of these colossal facilities. Furthermore, as highlighted by Gizmodo, Nvidia's accompanying blog post refrains from detailing the cost implications of building such advanced data centers compared to those employing less efficient air cooling, despite the company's claim that "every cloud provider and data center operator building for [Rubin] is making the transition."

These efficiency gains are partly attributed to operating AI servers at higher temperatures, reaching up to 113 degrees Fahrenheit (45 degrees Celsius). This strategy aligns with Amazon's recent emphasis on increased heat tolerances to enhance the efficiency of its predominantly air-cooled data centers.

Nvidia's system is designed to capture heat directly at the chip, transporting it through liquid loops that operate at significantly higher temperatures. This innovative approach enables outdoor dry coolers to efficiently reject heat for much of the year, providing considerably greater flexibility irrespective of ambient air temperatures.

According to Josh Parker, Nvidia’s head of sustainability, this reference design slashes water usage "from roughly 2.6 million gallons per megawatt per year for conventional cooling-tower-based systems to near zero — up to a 100 percent reduction."