Liquid immersion cooling is clear winner in hotter summers

Direct to chip cooling is a stopgap, and air cooling can’t cope with AI, dense compute, and hotter summers. By Pascal Holt, Director of Marketing, Iceotope.

This year’s record-breaking July temperatures in the UK have demonstrated that even in a northern European temperate climate, previously held ideas on the viability of certain ITE cooling solutions have not matched the reality. Although, today’s temperatures do not guarantee a similar summer for 2023, a swathe of experts believe that hotter summers in northern Europe are inevitable.

The ambient temperature around a data center effects the amount of energy required to cool the technology hall within. Air cooling systems including ‘free’ cooling and mechanical cooling solutions will require increased energy to optimize the temperature around the server racks. While the dryer, dustier environment will have impacts on water use and air contamination. ‘Many companies are now facing extra costs to cool equipment in heatwaves’, states GlobalData in a statement on 21 July 2022. It also concluded that, more frequent extreme weather events will mean data center outages will become more common. With climate experts contemplating future summer temperatures reaching around 40oC, this will have a major impact on air cooling capabilities in the UK data center industry.

Liquid immersion cooling has raised its profile significantly over the last 10-years, and today’s solutions are highly effective in reducing energy consumption, carbon footprint, solutions cost, as well as saving floor space and offering downstream heat recycling capability. Technologies such as high-performance computing are driving user interest in liquid cooling. Research company MarketandMarkets, recently stated that HPC is expected to grow from $36 billion in 2022 to $49.9 billion in 2027, a CAGR of 6.7%. This market utilizes CPUs and GPUs to process AI and ML data sets that generate greatly increased heat which cannot effectively be removed by ambient or chilled air-cooling processes.

These benefits have allowed developers and the technology supply chain to introduce a variety of liquid cooling solutions on to the market. Some liquid cooling technologies have moved the game on, whilst others indulge the industry with no more than stop-gap measures. According to a recent whitepaper by Schneider Electric, the market for liquid cooling is split into two-basic families, direct to chip (cold plate or conductive), and immersive. From these two families there are five main liquid cooling methods.

Direct to chip, single phase, where the liquid coolant is delivered directly to the hotter components using a cold plate directly on the chip within the server. Most variants only cool the processors and possibly memory devices. Server fans are still required to cool other components. This method is well known by the gaming fraternity.

Direct to chip, two phase liquid cooling, similar to single phase cooling, except that the liquid utilized is two-phase, changes from liquid (cool) to gas (hot) to remove the heat from the main targeted devices.

Chassis-level precision immersion cooling, single phase, precision delivers the coolant to the hottest components first as it immerses the server board in a single-phase fluid. This method ensures all sources of heat are removed. All server fans are removed, and the coolant provides an environment which is inherently slow to react to external changes in temperature and renders boards and components immune to the ingress of humidity and pollutants. This method also enables near silent operation. Chassis-level immersion cooled servers are compatible with current rack cabinet designs.

Tub – open bath, single-phase cooling, completely submerges the ITE in fluid and usually servers are stacked vertically rather than horizontally in a normal rack. Heat absorbed by the fluid is circulated to a water loop via heat exchanger.

Tub – open bath, two phase cooling, as with the single-phase tub method, the IT is again completely submerged in the fluid. The solution uses two-phase dielectric coolant which changes state as it removes heat from the around the servers.

The increasing utilisation of artificial intelligence (AI) across markets as diverse as finance, security, medical, manufacturing, and construction continues to drive up the processing requirements and consequently the power draw of servers being installed in locations from data centres to stand-alone environments. Whilst energy and compute dense 100kW+ racks can reduce the space required for big data manipulation; they create a challenge for standard air-cooling techniques which cannot effectively remove the amount of heat generated from high performance IT equipment.

Recently increasing numbers of data center operators are investing in hybrid approaches to facility cooling that will provide a path forward for new and legacy sites. These combine the use of both air and liquid cooling techniques in technology halls. Targeting the application to the cooling solution according to the power requirements for specific loads. This would allow data centres to provide customers solutions for HPC capabilities as well as less performance orientated needs of mass storage and lower performance chips. Matching energy efficiency in HPC environments, while ensuring lower energy load to remain competitive and increasing overall site energy effectiveness.

For legacy data centres, large and small, it raises the question of why adapt the physical infrastructure simply to supply liquid cooling CPU/ GPU chipsets, whilst still requiring all the infrastructure for much less efficient air cooling for the rest of the server components – including fans, CRAHs and other CRAC equipment to provide sufficient airflow?

Precision immersion cooling reduces complexity and increases effectiveness

A consistent technology approach, such as chassis-level precision immersion cooling eliminates the heavy investment in air-cooling mechanical systems or allows data centres to reduce or stop using legacy systems altogether. Reclaiming energy previously used to air cool ITE helps maximise the power available for the compute load, and in turn, reduces PUE.

Chassis-level precision immersion cooling for AI applications and scaling compute is always far more effective at heat removal, including during greatly increased data throughput and processing, benefitting both the data centre operator and the customer. It is also less effected by extreme heat variations, ensure a consistent environment for the ITE.

Along with efficiency there is resilience; precision immersion cooling protects all the server components extending life cycles, and in the unlikely event of an outage, immersion cooling extends the period the components are protected from a heat build-up, allowing a structured shut-down process. If a hybrid cooling system was in place, the processors may be protected by the direct-to-chip liquid, however, other server components maybe damaged and the server warranty invalidated by an extended period outside manufacturer’s parameters.

Chassis-level precision immersion cooling at the edge

The numbers of data centres and the development of Edge computing networks are increasing and require massive capital expenditure. Growing numbers of investment firms expect climate commitments in order to release funds. More new funds are reliant on green projects and sustainable investments where the operator is expected to demonstrate responsibility in the consumption of resources and raw materials, guarantee lower PUE values and meet other ESG goals.

Precision immersion cooling solutions can demonstrate data center operator or IT owner’s commitment to sustainability through energy efficiency, extended component lifecycles, higher density compute in space saving environments, as well as reductions in ITE noise output, in built-up locations.

Two further benefits of precision immersion cooling, which are very current at this time are the greatly reduced volume of water required by this solution. Also, liquid cooling’s highly efficient heat transfer opportunity is becoming a consideration in large installations where local heat reuse infrastructure is included in local planning. This is a position that will certainly increase as authorities require more integration in local developments.

Whether the climate is heating up or not, most data center operators and IT equipment developers realise that the systems and infrastructure being planned must provide more energy efficiency and sustainability benefits and contribute to the corporate bottom line. Reduction in non-IT costs can save the industry hundreds of millions per year and being more energy efficient with our power grid requirements will allow expansion in other area of the communities where data centers are located. Immersion cooling is a highly effective contributing factor to various positive steps in the data center journey to a more sustainable future.

By John Kreyling, Managing Director, Centiel UK.
By Stewart Laing, CEO, Asanti Data Centres.
By Jamie Cameron, Associate Director, Cundall.
By Michael Crook, Data Center Market Development Manager, Corning Optical Communications.
In the next five years, Nigeria's data centre industry is set to explode, doubling its capacity...