Chilling out with liquid cooling in data centres

Change your location to view content tailored for you.

7 June 2024 | 5 min | Featured article

By , and

The need for increased computational power alongside reduced carbon emissions is leading data centre operators to explore innovative technologies to meet both their operational needs and their clients’ carbon reduction targets. One of these technologies is liquid cooling.

Aurecon’s Philip Motteram, Ross McCarthy and David Wong describe how liquid cooling for data centres is gaining significant attention across Asia Pacific.

Key takeaways

Data centres must decarbonise to cut their own emissions and respond to their clients’ Scope 3 reduction targets
Liquid cooling is an option for current and future data centres, allowing for more condensed racks into smaller spaces, and higher PUE (power usage effectiveness)
Adoption of liquid cooling requires careful planning, evaluation, and consideration of technical, economic and environmental factors

The rising demand for power density from AI

Liquid cooling technology in data centres represents a leap forward in addressing the escalating demands for power density, from advancements like artificial intelligence, while navigating the imperatives of sustainability and energy efficiency. As the industry grapples with these challenges, successful adoption of liquid cooling requires careful consideration of numerous factors including performance optimisation and cost-effectiveness.

With constant access to information, and increasing computational demands, server and GPU (general processing unit) technologies are raising power densities. Traditional data centres typically have a power density of 8 to 15 kW per rack, but GPU-based servers for AI model training require 45 to 55 kW per rack or more^[1].

Efforts to decrease energy usage and carbon emissions have seen an increase in the use of renewable energy and also necessitates investigating new cooling technologies.

Traditional air cooling vs liquid cooling

'Traditional air cooling' is reaching its limits for two reasons:

The technology has been optimised to its maximum efficiency, with fans now contributing up to 25 per cent of total PUE energy
Air cooling is unable to cool high-density racks of high TDP (thermal design power) GPUs efficiently and sustainably

'Liquid cooling' offers a promising solution with high density cooling and energy efficiency benefits. Liquid cooling uses fluid, rather than air, to cool the data centre servers. This allows the cooling medium to be brought closer to the heat source, thus requiring less power for fans.

Moreover, liquid cooling solutions can enable increased PUE (Power Usage Effectiveness). What this also does is provide data centre owners with the option to potentially fit more racks into the same size halls as traditional air-cooled facilities, or build more condensed data centres.

Liquid cooling offers a promising solution with high density cooling and energy efficiency benefits.

The adoption of liquid cooling in data centres across Asia Pacific will vary due to infrastructure readiness, regulations, and economic factors). In land constrained areas like Singapore and Hong Kong, the appeal lies in building more compact data centres. In Australia, where energy is costly, liquid cooling’s potential to reduce PUE and create more energy-efficient data centres aligns with new government funding programmes for green data centres^[2].

The benefits of liquid cooling technologies and reusing waste heat

Choosing between direct and integrated liquid cooling systems impacts the effectiveness and efficiency of data centre cooling. It’s crucial to understand each system’s advantages and limitations, especially given the rapid evolution of GPU servers. For instance, while companies are awaiting the delivery and installation of Nvidia’s H100 GPUs, its B200 version with double the memory is already available.

Below we compare different liquid cooling technologies:

Technology 1: Direct-to-chip

Involves circulating coolant directly through server components, providing efficient heat transfer and precise temperature control.
Designed to support high TDP (thermal design power) GPUs, maximising their heat dissipation efficiency while handling intense applications such as generative AI.
Provides superior thermal resistance for removal of heat.

Technology 2: Integrated heat exchange liquid cooling

Utilises a separate cooling loop to transfer heat away from servers.
These systems can be used in conjunction with air-cooled systems to cool environments with mixed rack densities.

Technology 3: Immersion cooling

Submerging hardware (servers, GPUs, CPUs) in special dielectric liquids (such as mineral oil) that conduct heat but not electricity.
A core benefit of this technology is reducing the volume of heat generated from the cooling process.

When evaluating different liquid cooling technologies, data centre operators must consider specific needs and constraints, such as whether the facility is new or refurbished, available land space, and optimum layout. Compatibility with existing infrastructure, future scalability, and flexibility considerations are also important.

Additionally, operators should evaluate the impact on maintenance, as liquid cooling systems require specialised resources to maintain their efficiency and these skills can be difficult to source in some remote or regional locations.

Reusing waste heat depends on the technology

The fleets of 'hot and always-on servers' within a data centre produce significant waste heat, and with pressure to decarbonise, the industry must make better use of this heat.

'Air-cooled data centres' generate waste heat but it isn’t hot enough for direct reuse and needs additional heating to be useful.

'Liquid cooling' produces slightly hotter waste heat, which still has limited uses but may be utilised for heating nearby swimming pools or district heating systems (common in Europe). For instance, Germany’s new climate protection act mandates data centres donate 10 per cent of their surplus heat to district heating systems by 2025, and 20 per cent by 2028^[3].

To reuse waste heat effectively, partnerships between data centre operators, energy retailers and wholesalers, local governments, and communities, are essential. The feasibility of reusing waste heat depends on the specific liquid cooling system and its operating constraints. Existing cooling towers can sometimes be repurposed but require modifications.

Adopting liquid cooling technologies for data centres

There are three main pathways for adopting liquid cooling:

Building new liquid-only data centres
Future-proofing newly built air-cooled facilities with the required infrastructure to support liquid-cooled racks in the future
Integrating liquid cooling into current air-cooled facilities

New data centres are often built with provisions for liquid cooling (2), meeting initial demands and allowing for future transitions. Existing data centres need to consider retrofitting for liquid cooling (3) to meet sustainability demands.

Changing existing facilities requires significant adjustments to rack and cabling infrastructure. A key challenge is balancing capacity: determining how much heat load each system will handle and how liquid cooling affects air-cooling systems.

Here today, not on the horizon

Liquid cooling technology is a viable and effective solution for data centre operators to improve efficiency, lower energy consumption, and reduce environmental impact. The benefit for data centre clients is in the reduction of their Scope 3 emissions to meet sustainability targets.