Liquid cooling solutions are gaining momentum as AI workloads turn up the heat in data centers
Artificial intelligence is creating unprecedented demand in data centers as the need to process large amounts of data continues to increase.
As tech giants rush to expand their infrastructure to accommodate AI workloads, they are faced with the growing challenge of how to power these activities sustainably and affordably – and this has even led companies like Oracle and Microsoft to adopt nuclear power as explored a potential solution.
Another critical issue is controlling the heat generated by powerful AI hardware. Liquid cooling has proven to be a promising way to maintain optimal system performance while meeting increasing energy demands. In October 2024 alone, several technology companies announced liquid-cooled solutions, highlighting a clear industry shift in that direction.
Liquid-cooled SuperClusters
At the recent Lenovo Tech World event, the company showcased its next-gen Neptune liquid cooling solution for servers.
The sixth generation Neptune, which uses open-loop, direct hot water cooling, is now deployed across the company’s partner ecosystem, allowing organizations to build and run accelerated computing for generative AI while reducing data center energy consumption by up to 40% reduced. , the company says.
At OCP Global Summit 2024, Giga Computing, a subsidiary of Gigabyte, presented a direct liquid cooling (DLC) server designed for Nvidia HGX H200 systems. In addition to the DLC server, Giga also unveiled the G593-SD1, which features a dedicated air cooling chamber for the Nvidia H200 Tensor Core GPU, aimed at data centers that are not yet ready to fully embrace liquid cooling.
Dell’s new Integrated Rack 7000 (IR7000) is a scalable system designed specifically with liquid cooling in mind. It is capable of managing future applications up to 480 kW, while capturing almost 100% of the heat generated.
“Today’s data centers cannot keep up with the demands of AI and require high-density computing and liquid cooling innovations and modular, flexible and efficient designs,” said Arthur Lewis, president of Dell’s Infrastructure Solutions Group. “These new systems deliver the performance organizations need to stay competitive in the rapidly evolving AI landscape.”
Supermicro also unveiled liquid-cooled SuperClusters designed for AI workloads powered by the Nvidia Blackwell platform. Supermicro’s liquid cooling solutions, powered by the Nvidia GB200 NVL72 platform for exascale computing, have begun testing samples for select customers, with full production expected in late Q4.
“We are driving the future of sustainable AI computing, and our liquid-cooled AI solutions are rapidly being adopted by some of the most ambitious AI infrastructure projects in the world, with more than 2,000 liquid-cooled racks shipped since June 2024,” said Charles Liang, president and CEO of Supermicro.
The liquid-cooled SuperClusters feature advanced in-rack or in-row coolant distribution units (CDUs) and custom cold plates to house two Nvidia GB200 Grace Blackwell Superchips in a 1U form factor.
It seems clear that liquid cooling will be at the core of data center operations as workloads continue to grow. This technology will be critical to managing the heat and energy demands of the next generation of AI computing, and I think we are only just beginning to see the potential impact it will have on efficiency, scalability and and sustainability.