Our digital world is expanding faster than ever, and data centers work hard to process all our Netflix binges and online transactions. Behind the scenes, there’s a crucial player that helps these data centers run smoothly: cooling systems. In general, data center operators have two options for keeping their equipment cool. Air cooling has long been the go-to method, but liquid cooling is gaining ground.

When it comes to cooling capacity, liquid cooling far outperforms air cooling and is also highly efficient and cost-effective. In this article, we’ll look at two types of liquid cooling: immersion and conductive (direct-to-chip).

Liquid Cooling: The Basics

Data centers are packed with racks on racks of servers, each one computing data at lightning speed. All that processing creates lots of heat. Liquid cooling exposes some or all of the servers to a coolant running through a network of pipes. When the liquid reaches the hot spots, it soaks up that excess heat.

The now-warm liquid then makes its way back to a heat exchanger, which transfers heat away from the liquid (typically to a water loop connected to a heat rejection system). With the heat dissipated, the liquid flows back to the equipment to repeat the cycle.

Liquid conducts heat far better than air, which makes this a much more efficient way to absorb excess heat. It’s more energy-efficient and consumes 10–50% less energy than conventional air cooling. But as we’ve noted, there are a number of different forms of liquid cooling solutions. Direct-to-chip (DTC) cooling—one of the most common methods—circulates liquid coolant through channels or cold plates that come into direct contact with the hot components, such as computer and graphics processing units (CPUs and GPUs).

On the other hand, immersion cooling fully immerses the IT equipment in a non-conductive liquid coolant. The coolant absorbs heat from all parts, not just specific components, and carries it away to a heat exchanger located outside the immersion tank. The thermal energy is transferred to a separate cooling system, such as a chilled water loop. After this, the coolant returns to the immersion tank, and the cycle repeats. However, it’s important to note that while this technique cools the hottest immersed components quite effectively, other components may still need supplemental air cooling.

Both methods are an efficient way to keep servers cool right at the source. But let’s explore the differences between these two methods to better determine which might be best for your needs.

Comparative Analysis: Immersion vs. Direct-to-Chip Cooling

Both liquid immersion and direct-to-chip cooling aim to tackle the heat generated by high-performance processors. But they have distinct features that make them unique. Here’s what you should consider when choosing the right solution for your data center:

Thermal Resistance

While both methods of cooling a dramatic imprvments over conventional chilled air, direct-to-chip cooling dissipates slightly more heat than liquid immersion cooling. This is because the fluid circulated through the direct to chip system is typically cooler than the fluid temperature in immersion cooling systems, providing lower resistance and transferring more heat away from components at the chip level. However, direct-to-chip cooling doesn’t cool other components such as the hard disk. You would need an additional means of cooling for other equipment—typically chilled air—negating many potential gains in operational efficiency.

Cost

If your organization currently has existing chilled air infrastructure in your data centers the initial setup costs for immersion cooling need to be considered, as you will need to invest in immersion tanks and coolant circulation systems. But in the long term, it will be worth the investment because it reduces energy consumption significantly, and if you are designing and building a data center from the ground up immersion’s simpler infrastructure requirements (no air handlers, chillers, in-row cooling, and so on) can dramatically reduce the capital expenses. Eliminating fans and insulating your IT equipment from dust, static electricity, condensation, and vibration can also extend the lifespan of your equipment.

Direct-to-chip cooling requires significant upfront investment in specialized equipment and infrastructure as well—coolant distribution units, cooling infrastructure to circulate fluid to each individual CPU or GPU, and so on. But, like immersion, it utilizes energy efficiently and reduces electricity consumption, delivering significant cost savings in the long term. However, because direct-to-chip only cools individual components, it doesn’t provide as dramatic a reduction in energy consumption as you still need supplemental cooling for the rest of your equipment.

Scalability

Both solutions are highly scalable and capable of cooling high-density deployments. Immersion tanks can easily incorporate additional servers or racks, allowing you to add capacity as needed without greatly increasing your compute footprint. Direct-to-chip cooling can also be integrated with additional individual processors, allowing you to leverage existing air cooled data center space and increase compute capacity without the need for significant space.

Maintenance

Immersion cooling systems require regular maintenance to check the integrity of the coolant and prevent contamination. You may also need to adjust the fluid levels periodically, such as when servers are added to or removed from an existing deployment.

Direct-to-chip cooling doesn’t need external coolant tanks. But this method also requires regular inspections and maintenance to ensure optimal performance.

Safety

The liquids used in immersion and direct-to-chip cooling are non-reactive and non-toxic. However, both methods do carry the risk of leakage—though the risk is greater with the direct-to-chip method. It has several touchpoints with your hardware, and a leak at any point can cause catastrophic damage and a complete shutdown.

Future Outlook and Trends in Liquid Cooling

The demand for high-performance computing is rising, which calls for better solutions to keep data centers cool. With the integration of AI, next-generation AI data centers are poised for even bigger growth. There’s an insatiable appetite for AI given that it can handle intensive computer applications. With that, the processing load will increase even more.

GPUs are also becoming exponentially powerful with escalating energy demands to match. For example, Nvidia’s latest GPU enables “supercharged generative AI workloads.” It’s projected to consume a maximum power of 10.2 kW, 160% more than the previous version.

Fortunately, liquid cooling systems can meet this challenge. While still in its early-adopter phase, experts project liquid cooling will grow into a $1.6 billion industry by 2027. This means data centers are eager to make the switch to a greener solution to support their growth.

Several key players are already noting this disruptive technology. To return to Nvidia’s news above, they made the pivot early, realizing the gains—they launched their first liquid-cooled AI system in 2022, and have since said they would be using liquid to cool GPUs going forward. Now it’s time for other data centers to follow suit. In the coming years, growing technologies such as cloud computing, machine learning, and media streaming will also turn to liquid cooling to keep their applications running.

Stay Cool With Liquid Immersion Cooling

Data centers need to scale their technologies rapidly. And while this growth increases the heat load tremendously, Green Revolution Cooling (GRC) can meet the demand. We’re the pioneers of single-phase immersion cooling technologies. Our solution is vetted by IT giants such as Dell and Intel, and used by an international network of major clients.

Contact GRC today to learn how your data center can slash costs and grow its computing power easily and efficiently.