Subscribe to the Data Center Knowledge Newsletter
Get analysis and expert insight on the latest in data center business and technology delivered to your inbox daily.
Insight and analysis on the data center space from industry thought leaders.
Making the Case for Liquid Cooling in High-Density Data CentersMaking the Case for Liquid Cooling in High-Density Data Centers
As AI-driven high-density data centers expand, data center liquid cooling becomes essential to manage GPU heat – yet making the business case to the board remains crucial.
4 Min Read
Liquid cooling systems are key to managing the heat from AI-driven, high-density data centersImage: Alamy
High-density data centers with high-performing GPUs are changing how the world works by unleashing the power of AI. They’re also unleashing something else – tremendous amounts of heat. The only way to cool the power-hungry GPU racks in these data centers is with liquid. But deploying liquid cooling systems may first require addressing concerns that CFOs and board members have about the technology.
The liquid cooling systems needed for high-density data centers have higher upfront costs than the air-cooling systems used in traditional data centers. There are also worries about water usage, with the largest data centers capable of consuming millions of gallons of water per day. And for some, the idea of bringing liquid into the white space can be counterintuitive and risky after trying to keep it out for decades.
Facility operations teams can put these concerns to rest by making a business case for liquid cooling that communicates the necessity of liquid cooling and incorporates the value of design best practices.
Establishing the Need for Liquid Cooling
Some leaders assume that the higher-performing GPU racks used for AI workloads can simply be dropped into a traditional data center infrastructure. But the air cooling systems used in traditional data centers struggle to keep up when power density exceeds 15-20 kW per rack. Liquid, on the other hand, has 15-25 times the thermal conductivity of air.
Put simply, for a high-density data center to function reliably, it requires an equally robust liquid cooling solution. That solution can be closed loop to minimize water loss using one of three design approaches:
Direct-to-chip cooling puts water on a heat sink or cooling plate within the IT equipment to remove heat.
Rear-door heat exchangers capture hot air from IT equipment on a coil – before that hot air enters the white space – and then transfers the heat to water.
Immersion cooling submerges IT systems in a non-conductive liquid coolant to absorb their heat and then uses a heat exchanger to transfer away that heat.
Whatever design approach is used, it should be engineered in conjunction with the GPU infrastructure to help make sure it’s built to deliver the right cooling performance. Just as important as knowing how much power a GPU is going to consume, for instance, is to know how much cooling it’s going to consume. Advanced planning can also help uncover critical issues early, like if a building has the existing chilled water capacity to support a liquid cooling system.
The True Cost of Data Center Liquid Cooling
If leaders only consider the CapEx impact of a liquid cooling system, they’re missing the bigger picture. Liquid cooling can deliver OpEx savings that can potentially recoup the CapEx investment in less than two years.
For instance, the power used by a pair of cooling distribution units to circulate water is a fraction of how much energy a fan wall would consume to cool the space. Liquid cooling systems also operate within a more targeted footprint, cooling only the areas of the IT infrastructure where heat is generated rather than the entire room. And for every one-degree increase in room temperature, data centers can realize a 1-2% energy savings.
If liquid cooling systems need to be deployed at multiple data centers, a vendor-neutral basis of design can also provide savings. This involves creating a standardized design that delivers similar performance characteristics across a variety of locations while minimizing rework for each deployment. It also matches the right vendor technologies to a data center’s environmental and operating needs rather than forcing specific vendor technologies onto the data center.
Uncovering Opportunities to Simplify
Liquid cooling systems don’t need to be a burden on an organization. Across the life of these systems, there are opportunities to ease their deployment, use, and maintenance – all of which minimize demands on staff.
For starters, a partner who is experienced in deploying liquid cooling technologies can help companies with everything from assessing their facility infrastructure to planning and designing a tailored liquid cooling system to sourcing and supplying technologies. That same partner can also take on key tasks to simplify the deployment, like holding pre-purchased components so they don’t sit idle at a company’s data center, where they could be damaged or stolen.
Data from a liquid cooling system can also be integrated in a way that best serves the people responsible for monitoring and maintaining it. If that’s the facility operations team, for example, the data can be integrated into their building management or automation system. This can allow the team to easily keep an eye the liquid cooling system like they do any other building system.
Warming Up to Liquid Cooling
By establishing the need for liquid cooling and showcasing the savings and efficiencies it can deliver, data center teams can create a compelling case for liquid cooling and protect their high-value, energy-intensive GPU racks for years to come with the best available technology.
About the Authors
You May Also Like
