Nvidia says its AI data center design runs hotter to use less water

Public pressure against data centers has underscored their importance Water and energy consumptionAnd now Nvidia Highlight Its claim is that Jill Rubin’s reference design for an all-liquid-cooled data center has “eliminated massive amounts of energy use and almost all water use.” However, it does not address all concerns related to AI data centers, including during their construction Power generation requirements Huge facilities. Also, as Gizmodo Nvidia’s blog post notes the cost of building this style of data center versus one that uses less efficient air cooling, but claims that “every cloud provider and data center operator that builds for (Rubin) is in the process of switching.”

The efficiency gains are due in part to the AI ​​servers running hotter, up to 113°F (45°C). In a recent report, Amazon likewise Described Higher heat tolerance as part of making air-cooled data centers more efficient.

With Nvidia’s system, “heat is captured directly on the chip and transferred via liquid loops that operate at much higher temperatures, allowing external dry coolers to reject heat efficiently throughout most of the year,” with greater flexibility when it comes to ambient air temperature.

According to Nvidia’s head of sustainability, Josh Parker, the reference design takes water use “from about 2.6 million gallons per megawatt per year for traditional cooling tower-based systems to nearly zero — up to a 100 percent reduction.”