AI and liquid cooling: the most frequently asked questions and answers

What is liquid cooling, and why is the technology suddenly so relevant to data centers?

This contribution was submitted by Schneider Electric. Because of the relevant perspective and topicality of the topic, the editors are publishing it in full on ITdaily.

Data centers have always relied on air-cooled servers for computing power, but the acceleration of AI deployments is now causing a trend toward liquid cooling. Training AI requires a significant increase in computing power, requiring chip developers to increase thermal design power. This is the maximum amount of heat generated by a processor that must be dissipated by a cooling system to prevent overheating.

As chips give off more heat, air cooling is ultimately not enough, making liquid cooling the better option for AI servers. Liquid cooling is not new; after all, it has been used in High Performance Computing (HPC) for years. Still, traditional data center environments will have to adapt for a switch to liquid cooling to meet the demand for AI.

Here are the most frequently asked questions and answers about AI and liquid cooling:

How urgent is the need for liquid-cooled servers as AI becomes more prevalent?

Companies’ growing interest in implementing AI applications is calling for more liquid-cooled servers. Fluids are better at dissipating heat than air, and many new chips used for AI solutions cannot be adequately cooled with air. Therefore, data center managers must be quick to build infrastructures to support liquid-cooled servers.

What forms of liquid cooling can you employ?

Most popular is attaching a heatsink to a processor or memory, for example. Water with a special composition is passed through the plate, absorbs the heat and then transports it through a Technology Cooling System (TCS) loop to a coolant distribution unit (CDU) with a heat exchanger that dissipates the heat. Although a water-based fluid is most common, a specially designed fluid can also be used that boils in the heatsink and then condenses in the CDU. Although heatsinks dissipate most of the server heat via liquid, air cooling is still required.

read also

Hot water, cool servers: Nvidia’s Blackwell chips as pool heaters

Another method is immersion cooling, in which a server is placed in a chassis or tank filled with a dielectric fluid to cool the entire server. This dielectric fluid is usually a synthetic oil that dissipates heat. Although this method is not currently as popular as the heatsink, there are advantages. Namely, immersion cooling does not require airflow through the server, which ensures high thermal stability throughout the server.

How can existing data centers support the implementation of high-density AI servers with liquid cooling?

It is best to connect a CDU (coolant distribution unit) to an existing water system (loop) of your data center, building or facility, allowing heat from the Technology Cooling System (TCS) loop to go directly to the facility’s cooling system. If a connection to the existing water system is not available, a data center operator can deploy a CDU that removes liquid to air. When new data centers are built, the water systems can be designed to be fully capable of liquid cooling right from the start of operation.

What is a CDU and what is its function?

A CDU, or coolant distribution unit, controls the temperature, chemical composition and flow of the fluid to the server being cooled. The function of a CDU is similar to that of a transformer that regulates voltage. In this case, the CDU uses a heat exchanger and pumps to regulate the flow and temperature of the fluid being sent to the equipment for cooling. The CDU also isolates the Technology Cooling System (TCS) loop from the systems of data center, facility or building via an integrated “liquid-liquid” or “liquid-air” heat exchanger.

What is a good ratio of air- to liquid-cooled servers for a new data center?

Data centers will not be 100% liquid-cooled, although liquid cooling is increasingly becoming a must in high-density environments. The challenge for designers is to ensure that their cooling systems are flexible. Soon this will be easier as cooling systems come to market where it is possible to switch between air and liquid cooling. This not only provides the necessary flexibility but also increases efficiency, which supports data center sustainability strategies.

Is it possible for liquid-cooled servers to use 100% free cooling?

Yes, but it depends on the requirements of the server and the climate at the data center location. Free cooling uses outside air to lower the temperature of the fluid used for cooling. Using outside air is a more economical cooling method and suitable for lower temperature locations. However, water used to cool chips does not have to be cold; temperatures above 37°C are still acceptable, depending on the processor. This means that free cooling is possible even in warmer climates, helping to control data center cooling costs and promote sustainability.


This contribution was submitted by Schneider Electric. Because of the relevant perspective and topicality of the topic, the editors are publishing it in full on ITdaily. Schneider Electric also published a white paper on the topic, Navigating Liquid Cooling Architectures for Data Centers with AI Workloads.

newsletter

Subscribe to ITdaily for free!

  • This field is for validation purposes and should be left unchanged.