Conquering with Cooling: The ‘Cold Power’ Revolution

Conquering with Cooling: The 'Cold Power' Revolution

This image captures the inside of a data centre, where the enormous electricity costs from continuously running extensive ICT equipment and other facilities are evident. In fact, electricity alone accounts for over 60% of data centre operational expenses, with CRAC (computer room air conditioning) units being the primary contributor. As information technology advances, data centres—crucial for storing and processing data—face a growing challenge in energy-efficient cooling. Effective cooling has emerged as one of the key challenges for data centres. Today, we delve into the array of techniques used to tackle this challenge.

 

Air Cooling

Air cooling is a widely used cooling method in data centres, suitable for various scales and requirements. It works by circulating air to dissipate the heat generated by equipment, ensuring safe and efficient operation. Air cooling systems, such as fans and air conditioners, are generally cost-effective, technologically mature, and quick to install, offering a broad selection in the market.

However, with the advent of highly integrated row-level data centres and the emergence of high-density servers combined with low-density hybrid models, traditional cooling methods are struggling to meet the demands due to uneven heat distribution caused by server density variations. To address common issues in conventional data centres—particularly in environments with limited space and where rapid deployment and high

security are required—rack pool technology has been developed. This design integrates multiple critical functions into a single system, including cabinets, UPS, power distribution, cooling, monitoring, and cabling, while achieving efficient cooling through the isolation of hot and cold airflows.

（Figure 1-1: Rack Pool）

（Figure 1-2: Rack Container）

By integrating cabinets, power distribution, UPS systems, cooling, monitoring, fire protection, and structured cabling into a single container, traditional data centre projects are simplified into factory-produced, customised units. This enables rapid deployment, plug-and-play setup, and easy replication and expansion of data centres.

 

Liquid Cooling

Compared to air cooling, liquid cooling offers notable advantages, including faster heat dissipation, higher efficiency, and lower overall energy consumption. It significantly lowers the PUE (Power Usage Effectiveness) value, provides more precise temperature control, and enhances equipment performance and lifespan. These benefits have made liquid cooling a prominent topic in current discussions.

（Figure 2-1: Indirect and Cold Plate Liquid Cooling）

Liquid cooling uses a coolant to transfer heat from equipment via a heat exchanger, moving the heat to the coolant and then to external cooling facilities (such as cooling towers or chillers) for dissipation. This method avoids direct contact between the liquid and internal electronic components, reducing the risk of failure.

Cold plate liquid cooling is widely accepted in the market for its outstanding efficiency, especially in high-performance computing and data centre applications. In contrast, while immersion cooling performs better in some extreme scenarios, its higher cost and technical complexity lead to lower usage rates.

For the latest NVIDIA GB200 GPU, cold plate liquid cooling is particularly crucial. This GPU generates significant heat during intense computational tasks, making traditional air cooling systems inadequate. Cold plate liquid cooling directly contacts the GPU with the coolant, rapidly transferring heat away and ensuring stable operation under high loads.

（Figure 2-2: Direct and Immersion Liquid Cooling）

The previously mentioned immersion liquid cooling involves fully submerging the entire server or compute node in a liquid coolant. The liquid covers and directly contacts all electronic components, using its high thermal conductivity to transfer heat away from the equipment. This method not only effectively lowers the operating temperature of the devices but also reduces the impact of dust and other airborne contaminants on the equipment.

                                (Figure 2-3: Cold Plate)                          

(Figure 2-4: Liquid-Cooled Rack)

In addition to the “immersion” method, liquid cooling technology is applied to various components and equipment within data centres by designing more efficient heat exchange systems and cooling networks to distribute and manage heat. This approach allows for customised designs based on the specific needs and equipment layout of the data centre, improving overall cooling efficiency. This is known as a distributed liquid cooling solution.

 

Hybrid Liquid Cooling

Hybrid liquid cooling uses a combination of liquid and air cooling methods to optimize cooling based on the specific needs and heat output of the equipment. This approach ensures effective cooling while reducing system energy consumption and maintenance costs. As a result, hybrid cooling is one of the solutions within liquid cooling technology, often involving the simultaneous use of air cooling and cold plate liquid cooling. A common practice in hybrid cooling is to use liquid cooling for high-power, high-heat density components, while air cooling is used for low-power components. For instance, in IT equipment cooling, liquid cold plates are typically installed on CPUs, GPUs, or memory modules, while fans provide forced air cooling for other components.

Additionally, some designs integrate cold plates, pumps, and heat exchangers within the computing system. Despite hybrid cooling, air conditioning is still necessary for non-liquid-cooled components. To further reduce air conditioning power consumption, liquid-cooled rear door heat exchangers can be installed on cabinet doors for initial cooling of hot air. This approach can be used in high-temperature room designs or even eliminate traditional air conditioning, resulting in a fully liquid-cooled system.

In data centres, liquid cooling technology not only enhances cooling and energy efficiency but also improves environmental conditions, increases equipment reliability, and supports the future demands for high-density and high-performance computing. Each liquid cooling technology has its own applicable scenarios, advantages, and drawbacks. Choosing the right liquid cooling solution requires considering factors such as the type of equipment, density, operating environment, and budget.

 

As the saying goes, “It’s not how you start, but how you finish.” The “cold” power of data centres is an indispensable part of building a new, green, energy-efficient ecosystem.