Immersion cooling
Immersion cooling: A deep dive into a cool technology
Data is growing at an unprecedented pace. The advent of blockchain and AI solutions means the world generated and consumed over 120 zettabytes (120 billion terabytes!) in 2023 — three times the amount we generated just five years ago.
This presents opportunities for innovation, but comes with some big challanges. How do businesses cope with the vast amount of information they contend with? How can they protect all of their IT equipment? How can they power their AI operations sustainably? How can they ensure they are compliant with local regulations?
The answer lies in the data centre, but even these high-performance facilities are havig to overcome major challanges. The need for computing power is surging. The average rack density has increased 10X from 2-5kW. As such, the conventional air-cooled systems data centres use are struggling to keep up with growing demands.
For that reason, many data centres now opt for advanced liquid cooling systems like liquid immersion cooling or direct-to-chip systems. These solutions have the potential for increased efficiency, higher heat transfer capacity, and increased density.
But is it all as simple as it sounds? Or are there still some challenges to overcome? Let’s find out. But first. How does immersion cooling work in practice?
How does liquid cooling work?
Liquid cooling technologies are designed to manage heat in computing systems more effectively than air cooling. Depending on cooling requirements, each method offers distinct advantages.
The three main methods are:
- Immersion cooling: Servers are submerged in a non-conductive, dielectric liquid that absorbs heat without damaging components.
- Direct-to-chip cooling: Coolant is circulated through tubes in direct contact with key components like the CPU and GPU to transfer heat away quickly.
- Heat exchange cooling: Heat is transferred between two fluids, one cooling the servers and another removing the heat from the system.
Single-phase vs. two-phase immersion cooling
You now have a rough idea of how immersion cooling works. But as ever, it gets a little bit more complex.
There are two main types of immersion cooling: Single-phase and two-phase. Let’s look at each.
1. Single-phase immersion cooling
In a single-phase liquid cooling setup, the server racks are submerged in a dielectric coolant, which remains entirely liquid during the cooling process.
When the immersed servers produce heat, that heat is transferred directly to the fluid, as it has direct contact with the components. The coolant then transfers this heat to a cooling tower via a cooling distribution unit. Once the fluid cools, it can be circulated back into the tank.
As you may have guessed, this process is more straightforward than a two-phase system. It’s easier to maintain and cheaper to install. However, it’s often less efficient and requires more energy to run.
2. Two-phase immersion cooling
With a two-phase solution, the servers are placed in a sealed tank filled with a liquid with a low boiling point. Once the computer servers produce heat, the fluid begins to boil.
Unlike the single-phase system, which remains entirely liquid throughout, the two-phase immersion cooling solution lets the boiling liquid convert into vapour, removing the heat from the IT components.
The vapour then passes through a cooling system positioned above the rack servers. It transforms back into a liquid, which can then drip back into the tank.
Two-phase cooling methods are considerably more expensive to implement than single-phase solutions. But they are more efficient and require comparably less space.
Components of an immersion cooling system
Although the single-phase and two-phase immersion cooling systems have differences, each solution shares some essential similarities, particularly in the core components it uses.
- Cooling tanks: Naturally, you need a container to hold the fluid and IT equipment. These tanks must be secure and maintainable. As such, many data centres build solutions from scratch to facilitate this technology.
- Dielectric fluid: Dielectric liquids are a must. They need to conduct heat, but not electricity. The exact fluid chosen (hydrocarbon versus fluorocarbon) depends on the type of immersion cooling system.
- Heat exchangers: Once the servers produce heat, technology must be in place to prevent the liquid from overheating. Heat exchangers make this possible and are easy to integrate into immersion tanks.
- Pumps and filtration systems: Lastly, there’s the pump system. This facilitates the transfer of fluid between the immersion tank and the heat exchangers.
Of course, this is a rather straightforward overview of what an immersion cooling system includes. Different configurations are optimal depending on whether the system is single-phase or two-phase. A data centre may also need to adapt its setup if it wants to use a full immersion or partial immersion system.
Safety considerations for immersion cooling
A quick note on safety. Data centres have a responsibility to keep their immersion cooling systems secure and safe for the people who operate them. A failure could be catastrophic.
Choosing non-flammable liquids and managing toxicity are core concerns. In the case of dangerous liquids like fluorocarbon, effective safety measures and policies to eliminate health risks are a must.
In terms of design? Fluid leaks are still a worry. Every care needs to be taken to ensure pipes and immersion cooling tanks are leak-proof. This extends to maintenance, too. Regular inspections, even when there are no obvious issues, are vital.
Lastly, there are emergencies. If a disaster does occur, the data centre should have a clearly outlined response plan. Comprehensive safety data sheets go a long way.
Immersion cooling vs. air cooling
Immersion cooling is more expensive to install than air cooling. It’s more complex to manage. And it requires more dedication to maintenance. So why is it worth the time and money? Why aren’t businesses just sticking with the traditional air-cooled systems?
Well, there are a handful of compelling reasons for this. Let’s examine them.
Energy efficiency
Earlier, we mentioned that immersion cooling fluids have better thermal conductivity than air. This gives them a higher heat transfer capacity and superior power usage effectiveness (PUE), meaning they can help achieve up to a 96% improvement in computing efficiency.
Aside from the energy-saving potential, this also reduces the risk of overheating during high-performance computing — especially important given the intensive processes required to facilitate enormous AI and machine learning workloads.
Space
The need for large tanks, heat exchangers, and pipes might imply that immersion cooling takes up more space than an air-cooling system, but this isn’t the case.
The thermal conductivity and resulting efficiency of immersion cooling allow for higher rack densities and, by extension, more compact server designs. That means data centres can use space more efficiently and pack servers densely without worrying about overheating.
Environmental impact
Reduced energy consumption equals reduced emissions and a lower carbon footprint — a vital concern when 76% of consumers would stop buying from a firm that neglected ESG wellbeing.
Sadly, many data centres have a reputation for being particularly unsustainable, so much so that groups are trying to shut down new builds to save on water consumption. Immersion cooling is more sustainable as it has a greater cooling capacity, and much of the waste heat generated can be recycled. As an aside, it is also considerably quieter than the high-speed fans used for air cooling.
Cost considerations
In some cases, liquid cooling may be more expensive than traditional air cooling systems. However, running costs can be lower, which may lead to savings over time depending on the specific setup and usage.
Immersion cooling and the future of data centres
The need for immersion cooling is born out of necessity. AI technology is evolving fast, and edge computing models are commonplace. As such, we’ve seen the dawn of the AI data centre: facilities that can cope with enormous data processing requirements for AI and ML workloads.
Powering these facilities means data centres must think differently about efficiency and space. Immersion cooling is just one essential piece of this puzzle.
This paradigm shift is well underway. In February this year, SK Telecom, Iceotope Technologies, and SK Enmove (three global leaders in liquid cooling technology) signed an MOU committing to collaborating on the development of liquid cooling solutions to optimise energy efficiency for AI data centres.
Similarly, seven Dutch firms recently joined together to work toward efficient Edge computing data centres. Part of this consortium is data centre liquid cooling specialist Asperitas, who will help to provide immersion systems for the facilities.
You can see how immersion cooling infrastructure will be increasingly integral to the success of the modern data centre that can cope with AI workloads. Without it, there would be few opportunities to keep up with the enormous requirements these processes demand.
Summing up
The benefits of immersion cooling techniques are clear: potential cost savings, a more environmentally sustainable alternative, and, above all, the chance to keep pace with the world’s growing demand for data in the age of artificial intelligence.
Of course, the higher initial setup cost is still a concern. There’s also a learning curve when training staff and management to adapt to the technology. But these barriers will fade with time. Overall, the future of immersion cooling and, by extension, the future of efficient, sustainable data centre designs looks promising.
Macquarie Data Centres has been at the forefront of liquid immersion cooling technologies since 2022. Our flexible AI data centres incorporate the latest liquid cooling solutions to support modern businesses’ need for hybrid cooling for high-density workloads.
Learn more about our solutions today or book a tour to discover why we’re the trusted partner of 42% of all Australian Federal Government agencies.
Frequently asked questions
What liquid is used for immersion cooling?
It differs. While in all cases, the immersion fluid is dielectric (meaning it is thermally conductive but not electrically conductive), the exact fluid used depends on a few factors.
Single-phase often uses hydrocarbon-based fluids like mineral oils. Two-phase uses fluorocarbons, though these are far more expensive. Some data centres use silicon-based fluids, which are often even more costly.
Does immersion cooling need more maintenance?
Not usually. There are fewer moving parts, and because everything is sealed, dust has fewer opportunities to build up. This means that immersion cooling generally needs less maintenance. Still, this isn’t absolute. Fluid leaks are still a concern, and regular checkups to assess fluid quality are still important. The hardware’s lifespan is longer but not infinite.
Can immersion cooling systems be retrofitted?
This is a common concern. One of the biggest barriers to immersion cooling for data centres is the belief that entirely new facilities must be constructed to support it. Thankfully, there are ways to retrofit immersion cooling, even if challenging.
It can be difficult to install, and several considerations surround compatibility and space. Data centre operators must also be trained to manage both systems while the installation takes place. Still, with the help of a qualified immersion cooling specialist, it is entirely possible.
Is immersion cooling a new development?
No. Immersion cooling has existed in some form since the 1960s. But it’s recently experienced a resurgence. This is largely because it is becoming more efficient.
This has coincided with the growing demand for high-performance computing and a number of environmental concerns surrounding data centres. In that sense, the increased interest in immersion cooling is really a case of ‘right time, right place’.
