An introduction to how data centers are cooled

Key Highlights: 

• Data centers are crucial for processing data, however this generates significant heat that demands efficient cooling strategies to maintain optimal performance.

• Effective cooling is more critical than ever as data centers are projected to process 463 exabytes daily by 2025.

• Advanced cooling methods, like Precision Liquid Cooling, offer superior sustainability, efficiency, and compatibility with existing infrastructure, reducing energy consumption and carbon emissions.

Data centers are essential digital infrastructure facilities that house vast arrays of servers responsible for processing and storing data. The UK Government recently classified data centers as critical national infrastructure (CNI), marking their first new CNI designation in nearly a decade. This move underscores the growing importance of data infrastructure in our increasingly digital world. These data centers are critical to underpinning the functioning of the internet and the services that rely on the internet such as banking, national security, email, online messaging apps and video conferencing platforms. As AI technology becomes more widely deployed and important for businesses globally, the critical role data centers play will only increase. 

The computing activities within these centers generate significant heat, and it’s estimated that data centers globally will process 463 exabytes of data each day by 2025. This would generate somewhere in the region of 20 gigawatts of heat, assuming that a single mid-range server can process around 1Gb of data per second. This means there is a significant need for effective cooling that ensures optimal performance. But, what are the various cooling methods? And what, from a data center’s financial and environmental perspective, is considered to be the best one? As we explore this in greater detail, you might have some questions, which we hope to answer in our FAQ section at the bottom of this article.

Traditional Cooling Methods

Traditional cooling methods are flawed and come with significant inefficiencies and limitations. These methods have been the standard for many years but simply won’t be able to meet the demands of the workloads required for AI.  They tend to consume a lot of energy and can be less efficient in densely packed environments.

Air Cooling

Air cooling is the most conventional method employed in data centers, despite the inherent inefficiency of air as a cooling medium. It involves circulating cool air through the server racks to dissipate heat. This method typically uses Computer Room Air Conditioners (CRAC) or Computer Room Air Handlers (CRAH). These units draw warm air from the data center, cool it down, and recirculate it. While air cooling is well-established and has lower initial setup costs, it tends to consume a lot of energy and can be less efficient in densely packed data centers. As the power density of new processors increases, air cooling struggles to cope.

Raised Floor Cooling

Another traditional method is raised floor cooling. This design uses a raised floor system to distribute cool air through perforated floor tiles directly to the server racks. The cool air is supplied by CRAC or CRAH units located below the raised floor. Raised floor cooling improves airflow management and can be effective in moderately dense setups.

However, raised floor cooling has several drawbacks. One major issue is the complexity and cost of installation and maintenance. Raised floor systems require a significant initial investment in infrastructure, including the construction of the raised floor and the integration of CRAC or CRAH units. Additionally, these systems can be difficult to modify or upgrade. Airflow management is another critical challenge, as improper sealing or placement of perforated tiles can lead to uneven cooling and hotspots. The position of servers within a rack can also impact cooling efficiency, with poorly arranged servers contributing to thermal imbalances. Furthermore, raised floor cooling can be less effective in handling the higher thermal power densities of modern high-performance computing racks, making it increasingly less viable for today's data center requirements. 

Advanced Data Center Cooling Techniques

As data center demands continue to grow, traditional cooling methods often fall short of providing the necessary efficiency and capacity. This has led to the development of advanced cooling techniques designed to meet the high thermal power densities and energy efficiency requirements of modern data centers. These innovative methods offer superior cooling performance, enabling higher server densities and more sustainable operations.

Liquid Immersion Cooling

Liquid immersion cooling is an advanced method where servers are submerged, in large tanks,  in a thermally conductive but electrically non-conductive liquid. The liquid absorbs the heat generated by the servers and transfers it away from the hardware. This method provides superior heat dissipation and allows for higher server density. The tanks are designed to scale outwards rather than upwards, allowing for more efficient space utilization and easier access to the hardware for maintenance.

However, liquid immersion cooling has several drawbacks. The initial setup and maintenance can be complex and costly. The immersion tanks require careful management to prevent fluid contamination, ensure consistent performance, and avoid IT hotspots. Additionally, not all server hardware is compatible with immersion cooling, limiting its applicability in diverse data center environments.

Chilled Water Systems

Chilled water systems use a network of high pressure pipes to circulate cold water to heat exchangers or cooling coils within the data center and the servers themselves. The heat from the servers is transferred to the water, which is then chilled by external cooling units. Chilled water systems offer high cooling capacity and design flexibility, but they come with significant downsides.

The installation and operational costs are high due to the extensive plumbing and cooling infrastructure required. Regular maintenance is necessary to prevent leaks and ensure system efficiency. Furthermore, chilled water systems can be less responsive to rapid changes in heat load, potentially leading to thermal inefficiencies during peak computational activities. Using chilled water can also mean that evaporation is required to release the captured heat into the atmosphere

Direct To Chip

Direct to chip cooling works by circulating a cooling liquid to plates which are in contact with a computer’s hottest components (for example CPUs and GPUs). Depending on how the system is designed, fans may still be needed to remove heat from the components inside the server that are not cooled by the cold plates. . Other direct to chip deployments use evaporation.

Direct to chip systems will typically be able to remove between 70-75% of the heat in the system, still leaving ~30% that needs to be removed by other cooling methods such as fans.  

Precision Liquid Cooling

Precision Liquid Cooling represents a significant advancement in data center cooling technologies. Precision liquid cooling combines the benefits of tank immersion and direct-to-chip cooling methods. Similar to tank immersion, it promotes sustainability by minimizing water use and reducing carbon emissions by up to 40%. This technology captures almost all generated heat, enabling efficient reuse and supporting environmentally friendly data center initiatives.

Precision Liquid Cooling also achieves the high performance typical of direct-to-chip cooling. It targets the hottest components, in parallel, with a small amount of dielectric fluid, maintaining optimal temperatures and preventing performance-degrading hotspots.

In Precision Liquid Cooling systems, a closed loop of dielectric coolant circulates within each server’s sealed chassis, and is delivered to the heat-generating components. The coolant absorbs the heat and transfers it via  heat exchangers to a warm water-based circuit, to be either reused, or  dissipated outside the data center. Crucially, Precision Liquid Cooling isn’t reliant on fans or evaporation, instead using dry coolers to remove heat.

Additionally, Precision Liquid Cooling simplifies integration by utilizing the same rack-based architecture as traditional air-cooled systems, ensuring compatibility with existing infrastructure. This blend of sustainability, performance, and ease of integration establishes Precision Liquid Cooling as the most  advanced and versatile cooling solution for modern data centers.

Advantages of Precision Liquid Cooling

Precision Liquid Cooling provides several distinct advantages that make it an attractive option for modern data centers. One of the primary benefits is enhanced cooling efficiency. By providing targeted cooling, this method ensures that heat is efficiently removed from critical components, resulting in improved performance and reliability of the servers. Additionally, Precision Liquid Cooling allows for increased server density. Traditional air-cooling methods require ample space for airflow, limiting the number of servers that can be housed in a rack. With Precision Liquid Cooling, more servers can be installed in the same space, maximizing data center capacity.

Another significant advantage is reduced energy consumption. Liquid cooling systems are more energy-efficient compared to air cooling systems. The reduced need for air conditioning and the efficient heat transfer properties of liquids results in lower energy usage, leading to substantial cost savings. 

While the initial setup costs for Precision Liquid Cooling systems may be higher than traditional air cooling systems, the long-term operational savings and reduction in cooling complexity are substantial. The efficiency of liquid cooling reduces energy bills and maintenance costs. Additionally, the extended lifespan and reliability of hardware due to better cooling further contribute to cost savings. Understanding the cost dynamics of cooling is crucial for data center operators.

FAQ:

Q: How does data center cooling work?

A: Data center cooling involves various methods to manage and dissipate the heat generated by IT equipment. Efficient cooling is crucial for maintaining optimal operating temperatures, ensuring equipment reliability, and improving energy efficiency in data centers. Below are the common cooling options used:

• Air Cooling: Air conditioning units or computer room air handlers (CRAHs) circulate chilled air through raised floors or overhead ducts to cool IT equipment. Power-hungry fans are used to distribute the cool air around the data hall and through the servers to remove their heat.

• Chillers: Chillers generate cold water, which is then circulated through pipes to absorb heat from servers and other hardware.

• Direct to Chip Cooling: This method uses cold water generated by chillers, circulated directly to chips or components via pipes to absorb heat efficiently.

• Liquid Immersion Cooling: Servers are submerged in large tanks filled with dielectric fluid (a thermally conductive but electrically non-conductive liquid) to absorb and transfer heat away from the hardware.

• Precision Liquid Cooling: Dielectric coolant is applied directly to components, offering efficient heat transfer and reducing energy consumption.

Each method is designed to capture the heat generated by the IT equipment and move it away from the data hall, rejecting it to the outside environment.

Q: How much does it cost to air-cool data centers?

A: Cooling data centers using air represents a significant portion of their operational expenses due to the high energy requirements involved. The cost to cool data centers can vary widely depending on factors such as the size of the facility, the efficiency of cooling systems, geographical location, and the climate in which the data center is situated.

Typically, air-cooling accounts for around 40% of a data center's total energy consumption. For larger facilities or those located in warmer climates, these costs can be even higher. The expense includes not only the electricity used by cooling equipment such as air conditioners, chillers, and pumps but also maintenance and infrastructure costs associated with these systems.

In recent years, advancements such as the adoption of more efficient cooling technologies like Precision Liquid Cooling and innovative cooling designs have aimed to reduce these costs. 

These technologies can lower energy consumption and operational expenses over the long term, making them an attractive investment for data center operators looking to optimize efficiency and reduce overall operating costs.

Q: How to calculate cooling requirements for data centers?

A: Calculating the cooling requirements for a data center involves several factors. The first step is heat load calculation, determining the total heat load generated by all the servers and equipment, usually measured in kilowatts (kW) or British Thermal Units (BTU). Based on the heat load, the cooling capacity needed is calculated, considering the efficiency of the cooling system and the desired temperature range.

Incorporating redundancy is essential to ensure continuous operation even if one cooling unit fails. Safety margins account for unexpected increases in heat load or inefficiencies. Metrics like Energy Efficiency Ratio (EER) or Coefficient of Performance (COP) are crucial in evaluating the efficiency of the cooling system, with higher values indicating better efficiency.