Maintaining precise temperature and humidity control in data centers has long been a top concern for operators, but recent trends have greatly changed cooling requirements. As artificial intelligence (AI) and digital transformation adoption rates continue to rise, more high-density racks are being deployed in data centers and traditional air cooling methods may not be able to meet the demands of these racks. Data center operators experiencing rising rack densities or planning to deploy high density racks should consider alternative approaches to thermal management.
Like high-performance computing (HPC) deployments, AI applications require the ability to process massive amounts of data with extremely low latency. But unlike HPC, that data isn’t just in the form of text and numbers. AI applications often process data from heterogenous sources in multiple forms, including large image, audio, and video files. Entering 2020, 73% of AI applications were working with image, video, audio, or sensor data.
Feeling the Heat
To accommodate the demand for multi-format AI applications, chip manufacturers have responded with more powerful chips. Thermal power densities for leading central processing units (CPUs) and graphics processing units (GPUs) rose sharply in the last two years, after relatively modest growth in the previous five years.
With more of these high-powered CPUs and GPUs being packed into 1U servers and equipment racks being packed with these 1U servers to drive down latency, a growing number of facilities are now requiring support for racks with densities of 30 kW or more.
As rack densities approach and exceed 30 kW, data center operators reach the upper limits of their air-cooling systems. Air just does not have the thermal transfer capacity required to cool high-density racks, no matter how the system is optimized. Avoiding disasters from heat-related equipment failure may require the adoption of liquid cooling.
Benefits of Liquid Cooling in the Data Center
Introducing liquid cooling technology to the data center doesn’t just protect against equipment failure from overheated racks. It can provide numerous benefits that help organization’s meet their performance and sustainability goals while supporting these new, data-intensive applications.
Liquid cooling leverages the higher thermal transfer properties of water or other fluids to support efficient and cost-effective cooling of high-density racks. It can be up to 3,000 times more effective than using air, allowing the CPUs and GPUs in densely packed racks to operate continuously at their maximum voltage and clock frequency without overheating.
Liquid cooling also enables improved energy efficiency, provides better space utilization, reduces data center noise levels, and offers a low total cost of ownership. Like air cooling, liquid cooling is available in a variety of configurations that use different technologies, including rear door heat exchangers, direct-to-chip cooling, and immersion cooling.
In its paper, Data Center Thermal Management Report 2020, technology analyst firm Omdia predicts strong growth in liquid cooling and expects the trend to double between 2020 and 2024. Some of the factors fueling this growth in liquid cooling adoptions include increased chip and server power consumption as discussed previously, edge growth and changing energy efficiency and sustainability requirements.
Mitigating Risks Associated with Liquid Cooling
Some data center operators are leery of liquid cooling because of the perceived risks of bringing liquid to the rack. Yet, current-generation liquid cooling technologies can be deployed to minimize both the risks of leaks and the potential consequences of any leaks that do occur. This key is integrating risk mitigation into every step of the system design process.
Specific risk mitigation strategies will depend on the fluids selected as the various fluids used in liquid cooling systemshave different chemical compositions and service and maintenance requirements. When water is used as the fluid, it is often mixed with glycol to protect against corrosion and freezing. Dielectric fluids can be used in direct-to-chip and immersion cooling, which eliminates the risk of IT equipment damage from exposure to fluids. However, these fluids are expensive and may have environmental, health and safety considerations so similar risk mitigation strategies should be employed as when using water.
Once the fluid has been selected, the fluid distribution system can be designed to ensure wetted material combability of all materials exposed to the fluid. Special attention should also be paid to fitting design and location as poorly designed fittings can represent a weak spot in the distribution system and quick disconnect fittings may be required for system maintenance . Properly designed liquid cooling infrastructure also supports effective risk mitigation. Dedicated cooling distribution units (CDUs) should be employed to create a secondary cooling loop separate from the facility water system to provide more precise control of fluids being delivered to racks.
Even in the most carefully designed systems, leak detection technologies should be employed across the distribution to ensure that if any leaks do occur, they can be quickly detected.
When operators and their partners take the time to perform a detailed risk assessment at the front end of a project and incorporate risk mitigation strategies into every phase of the design, the risks associated with liquid cooling are far outweighed by the benefits this system provides. And with today’s technology, liquid cooling comes with far less risk than supporting high-density racks with insufficient cooling systems.
The Costs of an Overheated Data Center
AI and machine learning will continue to increase in popularity as businesses use these technologies to create competitive advantage. According to a McKinsey’s Global Survey, “The State of AI in 2020,” 50% of the respondents said they have adopted AI in at least one of their business functions. In a recent Gartner poll, 47% of respondents said that AI investments were unchanged from the start of the pandemic, and 30% said they planned to increase their investments in AI technology.
AI is now being deployed in data centers supporting cloud based HPC, finance, online gaming, healthcare, film editing, animation and media streaming among other industries. As a result, high-density equipment racks are becoming more widely deployed, forcing thermal management systems to evolve to meet new requirements.
Organizations that ignore the limitations of air cooling will, at best, experience increased energy costs and reduced performance as CPUs and GPUs throttle back their clock speeds to prevent overheating. Those who aren’t so lucky could experience costly equipment failure or shutdowns.
If an organization anticipates deploying racks of 30 kW or higher to meet the demands of new business-critical applications, it’s important to face the reality that using traditional air-cooling methods spells disaster from both a performance and sustainability standpoint. Finding partners with experience in managing the changes required to integrate liquid cooling into the data center can help organizations safely and effectively use liquid cooling technology.
Fred Rebarber is the senior technical director of thermal solutions at Vertiv. To learn more, read our white paper, “Understanding Data Center Liquid Cooling Options and Infrastructure Requirements.”
