As the artificial intelligence (AI) server market approaches an expected $49 billion by 2027, keeping AI infrastructure from overheating will continue to pose a significant challenge. IT teams are tasked with simultaneously addressing the perfect heat storm of data center thermal management – high thermal design power thermal design performance (TDP), high-powered servers, high rack densities and a need for low latency.

In recognition of these industry-wide issues, liquid cooling vendor ZutaCore introduced support for Nvidia H100 and H200 GPUs via its HyperCool two-phase direct-to-chip (DTC) liquid-cooling technology. ZutaCore claims its solution is designed to triple server processing capacity and requires 50% less energy and physical space compared to conventional cooling systems.

Two-phase DTC liquid-cooling technologies have been growing popular across CPU chip manufacturers in recent months, and “this announcement from ZutaCore supporting next-generation GPU designs is a significant milestone in the industry,” IDC Research VP Peter Rutten said.

Hyperscaler cloud providers and data center operators are privy to waterless DTC liquid cooling specifically for the AI and other high performance computing (HPC)  (HPC) workloads their customers are deploying. “With direct-to-chip waterless two-phase liquid-cooling technology, data centers can harness the power of the latest generation of processors and servers while meeting their sustainability goals,” ZutaCore CEO Erez Freibach told SDxCentral.

Each Nvidia GPU can consume up to 700 watts of power, which will “challenge data centers that are already struggling to control their heat, energy consumption and [carbon] footprint,” Freibach said. “HyperCool was designed specifically to solve these problems and has already been proven to cool the most powerful processors of 1,500 watts or more and currently for 100 kW per rack of computing power,” he said.

Aside from impacting the amount of available water in local communities, data centers don’t want water in their facilities because of leakage risks. When a data center houses hundreds of thousands of dollars worth of AI servers, leaking water “can be catastrophic,” damaging and corroding the infrastructure, Freibach said.

HyperCool removes the need for water and instead employs a two-phase boiling and condensation process that moves large amounts of heat off of the processors and away from the servers. By applying coolants directly to the chips to remove and disperse heat, those water-related threats can be avoided.

The liquid cooling solution also supports heat reuse in the data center, which can lead to a better data center power use effectiveness (PUE) score in any climate, according to the vendor.

Freibach claims that ZutaCore offers the only two-phase waterless DTC liquid cooling solution currently in production deployments. The “inherent advantage” to two-phase solutions like ZutaCore’s is that each piece of the chip receives the cooling it needs automatically. The technology’s dielectric liquid cooling is self-activated based on the core temperature of the processor, “only when and where needed, while coolant levels are automatically maintained,” he said.

HyperCool has been qualified by processor manufacturers like Intel and AMD, and the solution is currently deployed with major server manufacturers like Dell, Supermicro, ASUS and Pegatron.

“HyperCool has already proven its ability to cool Pegatron servers running Intel’s 4th generation Xeon processors,” said Andy Lin, Pegatron server product marketing manager. “By adding support for the Nvidia GPU, ZutaCore is now paving the way toward a more sustainable AI future where upcoming AI servers can be deployed in an energy-efficient, cost-effective and reliable way.”