The HyperCool direct-to-chip system from ZutaCore is designed to cool up to 120kW of rack power without requiring a facilities modification. Credit: Nataliya Sdobnikova ZutaCore has launched a liquid cooling system called HyperCool that can handle the forthcoming Nvidia GB200 Grace Blackwell superchip without modifications to the data center or the rack – a common requirement of liquid cooling. Grace Blackwell is a gigantic chip with 400 billion transistors and five components: two Blackwell GPUs, a Grace CPU, and two I/O chips. Its power draw is believed to be over 1kW, making air cooling an extreme challenge. ZutaCore claims to have the industry’s first dielectric cold plates to support this new platform, featuring the ability to cool 120kW of rack power with little or no modifications to current real estate, power, or cooling systems. (Read more about dielectric cold plates and other immersion cooling methods.) For contrast, air cooling taps out at about 30 kW and is no longer viable for cooling the system. And liquid cooling usually means some kind of piping has to be installed in the facilities, but ZutaCore claims that isn’t necessary with its system. ZutaCore’s technology can cool the NVIDIA GB200 Superchip with a single cold plate, “delivering an industry-first waterless direct-to-chip liquid cooling solution,” said Erez Freibach, co-founder and CEO at ZutaCore, in a statement. “This breakthrough frees AI data center owners and operators from the risks associated with water-based cooling, empowering them to deploy the processing power needed for compute-intensive workloads like generative AI without the fear of leakages.” ZutaCore says HyperCool can be implemented in new or existing data centers to deliver 10 times more computing power, a 50% reduction in total cost of ownership, 100% heat reuse, and reduced CO2 emissions. There is a small but growing ecosystem of servers certified to work with HyperCool, including models from Dell Technologies, ASUS, Pegatron, and SuperMicro. In addition to its offering for the GB200 Grace Blackwell Superchip, ZutaCore previously unveiled HyperCool support for older Nvidia H100 and H200 GPUs. Separately, ZutaCore announced a strategic OEM agreement with UNICOM Engineering, a Dell Titanium Partner, to deliver warrantied HyperCooled AI servers for the H100 and H200 GPUs. ZutaCore also announced that a global AI-as-a-service provider has already chosen its HyperCool direct-to-chip liquid cooling for AI servers with deployments starting in Q3 2024. Related content news AMD holds steady against Intel in Q1 x86 processor shipments finally realigned with typical seasonal trends for client and server processors, according to Mercury Research. By Andy Patrizio May 22, 2024 4 mins CPUs and Processors Data Center news Broadcom launches 400G Ethernet adapters The highly scalable, low-power 400G PCIe Gen 5.0 Ethernet adapters are designed for AI in the data center. By Andy Patrizio May 21, 2024 3 mins CPUs and Processors Networking news HPE updates block storage services The company adds new storage controller support as well as AWS. By Andy Patrizio May 20, 2024 3 mins Enterprise Storage Data Center news Nvidia to build supercomputer for federal AI research Federal agencies including the IRS and Pentagon will have access to the Nvidia DGX SuperPOD system through MITRE, a nonprofit organization that operates federally funded R&D centers. By Andy Patrizio May 15, 2024 2 mins Supercomputers Data Center PODCASTS VIDEOS RESOURCES EVENTS NEWSLETTERS Newsletter Promo Module Test Description for newsletter promo module. Please enter a valid email address Subscribe