WGI Insights

Cooling Down AI & Data Centers: Introduction to Liquid Cooling Systems

Written by WGI | Jul 23, 2024, 8:44:40 PM

Liquid and Immersion Cooling Technologies

The advent of liquid cooling technologies is rapidly progressing, utilizing water, refrigerants, and various synthetic phase-change materials as liquid dielectric coolants. These advancements are designed to enhance energy efficiency, reduce equipment size, and improve the economic feasibility of system upgrades for facilities. Currently, about 20% of data centers are already utilizing liquid cooling technologies. Meanwhile, around 60-65% are actively considering upgrading their existing air-cooled systems to liquid cooling, immersive cooling, or a transitional hybrid solution. These upgrades aim to bridge the gap between current air-cooled systems and advanced liquid cooling, ensuring optimal investments and minimal downtime. Many new data center facilities are specifically designed for AI workloads and incorporate liquid cooling from the start. However, existing facilities face challenges in transitioning smoothly and effectively. As data centers continue to evolve, their thermal design power (TDP) and rack densities are increasing, driven by advancements in IT technologies. AI-capable graphics processing unit (GPU) chipsets now surpass TDPs of 300 W, 800 W, and are rapidly approaching 1,000 W or more.
These chipsets are utilized in cloud and enterprise applications such as deep learning, natural language processing, AI chat generation, imaging, and training. As a result, the TDP of typical AI servers is now approaching 6 kW to 10 kW per server. Server designers predict that near-future server power requirements will increase from the current 35-50 kW per cabinet to as high as 100 kW per cabinet. Major IT industry players and Multi-Tenant Data Centers (MTDC) owners and operators are keen to introduce upgrades to improve efficiency and better serve their clients. They must decide how much space to allocate for new TDP workloads to meet current demand and future growth over the next one to three years. Some will convert a few racks at a time, while others will dedicate entire rooms to these workloads and support the addition of liquid cooling systems. Understanding the available technologies and the specific demands of your data center will be critical to its future success.

CDU – Cooling Distribution Units

  • Cooling distribution units provide separation between the IT equipment in the racks and the outdoor heat rejection equipment, like a cooling tower or dry cooler.
  • The heat exchanger in the CDU keeps the two water systems separated so they never mix, allowing the liquid circulating in the racks to be unaffected by the water circulated outdoors.
  • CDUs can be very energy efficient because they avoid the use of refrigeration equipment like chillers and DX coils using compressors.
  • These systems are usually cost-effective compared to most other systems.

RDHx - Rear Door Heat Exchangers

  • RDHx systems are finned coil doors attached to the back of server racks. They perform heat exchange with coolant gas or water and facilitate hot air removal by use of RDHx fans.
  • An efficient method to remove heat close to the source, eliminating the need for hot or cold aisles.
  • RDHx can significantly reduce the energy needed for cooling. In certain conditions, the RDHx managed to cut cooling energy by 70-80% at the rack level, and 40-50% overall in the data center.
  • RDHx is typically accepted as a preferred initial step in migrating a data center facility toward a liquid cooling system through a staged, modular approach. It facilitates the development of a hybrid cooling system by integrating with existing air-cooling technology without necessitating significant structural changes to the white space.

Immersion Cooling

With the increase in AI hardware and other high-density applications, the need for a more efficient cooling solution has become more apparent.

Single-Phase Immersion Cooling

Single-phase immersion cooling involves installing servers vertically in a coolant bath containing a circulating hydrocarbon dielectric fluid, similar to mineral oil. The coolant is pumped to a heat exchanger where heat is transferred to a cooler water circuit. This technique uses “open baths,” as there’s little to no risk of the coolant evaporating. The fluid makes direct contact with the server components, facilitating effective heat transfer.
  • Advantages:
    • Simplicity
    • Safety
    • Uniform heat distribution
    • Excellent energy efficiency
    • Low likelihood of fluid loss (does not require liquid replenishment)
    • Lower maintenance
  • Likely to be applied in:
    • High-density environments where traditional air cooling is not a match for the load density
    • Situations where space is limited

Two-phase immersion cooling

Two-phase immersion cooling involves submerging servers or other heat-generating IT components in a thermally conductive dielectric liquid or coolant. As heat is transferred from the submerged servers, the surrounding liquid boils, changing phase to vapor. This vapor is then condensed back into liquid form by water-cooled condenser coils integrated at the top of the sealed racks.
  • Advantages:
    • Precision
    • Versatility
    • High cooling capacity (leveraging the latent heat of vaporization)
    • More efficient space utilization
    • Capability to handle higher heat loads.
  • Likely to be applied in:
    • Scenarios with exceptionally high heat loads, such as AI or HPC clusters.
    • When energy efficiency is a top priority, two-phase systems offer the advantage of minimizing power consumption while maximizing cooling efficiency.
    • Situations where space is limited.

Enhancing Data Center Efficiency with WGI

At WGI, our MEP experts are at the forefront of advancing data center performance with both traditional and cutting-edge cooling solutions. By integrating innovative technologies with a focus on energy efficiency and scalability, we ensure that your data center excels in thermal management and is well-prepared for future advancements. This is Part Two of our exclusive three-part series on Cooling Down AI & Data Centers. Be sure to explore Part One: Intro to Cooling Systems, for an in-depth look at the most common data center cooling techniques and how they are being utilized. In the final installment of this series, Part Three, we’ll delve into the world of direct-to-chip cooling, uncovering how this next-generation solution is reshaping data center efficiency - So, stay tuned!

Contact Us

Ready to transform your data center's cooling efficiency? Contact us today and discover how our team of experts can help you stay ahead of the curve with innovative and sustainable solutions. Let's turn your cooling challenges into a breeze!
View full post