Explanation of direct chip cooling, in single phase and dual phase.
Getting into trouble.
In the realm of data centers, heat management presents itself as one of the biggest challenges and at the same time, one of the most significant opportunities. When server components operate, they convert all the electrical energy they consume into heat, which needs to be dissipated. Historically, organizations have relied on air cooling, using fans to manage this heat. According to various estimates, about 80% of data centers are still primarily based on this method.
However, air cooling is considered inefficient and consumes a large amount of energy. A recent study by McKinsey indicates that this system can account for up to 40% of the total energy consumption of a data center. This has led 40% of operators of these facilities to explore alternative cooling methods that are eco-friendly and cost-effective, thus seeking sustainable operations.
From a fundamental perspective, this transition makes sense, as fluids are much easier to control than air, which can be difficult to direct. Liquid cooling techniques are evolving, especially with the emergence of server loads that tend to be more energy-intensive and therefore generate more heat.
One of the most common forms of liquid cooling is direct-to-chip cooling (DTC). This method has been used by both data centers and gamers for decades. It involves a metal plate placed over the processors, with a conductive material in between. This plate uses liquid through pipes to draw heat away from the chip and disperse it via a dissipation mechanism. This mechanism can be as simple as a single fan or, in industrial settings, dry coolers equipped with evaporative cooling mechanisms that work in tandem.
In warm climates or specific locations, evaporative cooling is complemented with dry coolers, where warm air passes through wet pads, evaporating water and cooling it. DTC techniques are more specific than air cooling, as the liquid in the pipes can be directed more precisely to the components that need cooling. Additionally, liquids often have greater thermal conductivity than gases, making them more efficient.
Nevertheless, even with DTC, it is generally necessary to use some air to address the challenges of adapting multiple designs of cold plates for all heat-generating equipment. While GPUs and CPUs are the main heat generators, RAM and hard drives also contribute, making a mix of air and water cooling common at an approximate ratio of 30%/70%.
In a move toward more advanced methods, immersion cooling is introduced, where the entire server is submerged in a liquid. This method has several advantages: all components are cooled simultaneously, and it can handle higher thermal loads. Furthermore, immersion prevents dust from entering the system, which extends hardware lifespan. However, immersion cooling is more complex than DTC, and its maintenance requires more attention.
There are two types of immersion cooling: single-phase and two-phase. In single-phase immersion cooling, the liquid remains in a liquid state throughout the process, while in two-phase immersion, the liquid boils and turns into vapor, which is then cooled and condensed back into liquid.
Additionally, residual heat management must be considered. Although dry coolers and evaporative methods can be employed to dissipate this heat, effective reuse remains a significant challenge, especially for older data centers located in industrial parks, far from areas where this heat could be repurposed. Although some data centers, like those in Germany, have started to use residual heat to warm offices, it is vital for the industry to continue improving these innovations.
Over the past two decades, liquid cooling technology has advanced considerably, capable of managing increasingly higher thermal and energy loads. Immersion cooling is still in its early stages of evolution, but it has great potential to address the cooling of components operating at high temperatures. However, it is unlikely that a single model will be adopted, and we can expect to continue seeing a combination of DTC, immersion cooling, and some air cooling across data centers in the coming years.
