As processor performances increase, rising heat dissipation has become a growing challenge to the cooling system hitherto used in computers. In recent developments, the power consumption of a CPU has increased from 150W to more than 300W. For example, the 3rd generation of Intel’s Xeon processors, “Ice Lake”, have a maximum thermal design power of 270W. Similarly, in the case of GPUs, it has become 700W per chip.
Therefore, a major challenge for the data center industry is to strike a balance between Power Usage Effectiveness (PUE) and Total Cost of Ownership (TCO).
The return of liquid cooling
When air cooling, the traditional solution, can no longer keep up with rising heat dissipation needs in data centers, cloud service providers like Microsoft have been searching for alternatives.
Lately, Microsoft has become the first cloud provider using a two-phase immersion cooling solution, cooperating with the Taiwanese cloud infrastructure provider Wiwynn Corporation. Microsoft’s datacenter in Quincy, Washington State, has also become the first location to deploy two-phase immersion cooling.
According to Microsoft, two-phase immersion cooling can reduce power consumption for any given server by 5% to 15%.
Liquid cooling is already a proven technology that has seen widespread adoption elsewhere. It is used in the automotive industry to prevent engines from overheating. Immersion cooling itself, moreover, was already used in the early days of the computing industry when supercomputers were submerged in liquid coolants. It was later replaced by air cooling.
In essence, immersion cooling entails directly immersing datacenter electronics in a dielectric liquid that absorbs the heat generated. Compared to air cooling, the method reduces the need for additional components such as heat sinks, shrouds and fans.
Two types of immersion cooling currently exist: in single-phase immersion cooling, pumps are used to drive the heated fluid to a heat exchanger where it is cooled. In two-phase immersion cooling, the fluid is instead boiled by the heat from the electronics, and later condensed through a heat exchanger. The later process achieves higher heat transfer efficiency.
Before its application in data centers, immersion cooling has most recently been used in the cryptocurrency industry for their mining computers. In fact, Wiwynn Corporation has just invested in LiquidStack – the world’s largest liquid cooling company and also an immersion cooling solution provider for various Bitcoin mining sites operated by its majority shareholder, the Bitfury Group. Through the investment, Wiwynn and LiquidStack will cooperate on two-phase immersion cooling for datacenters, edge and high-performance computing.
Ecosystem – the key to widespread deployment
Despite the advantages of two-phase immersion cooling, Sun-Lai Chang, President of Wiwynn Corporation, believes there are still many obstacles to overcome, among them the higher cost of the dielectric liquid involved.
Furthermore, to optimize two-phase immersion cooling, server and even data center designs have to change as well. The density of motherboards within servers, for example, ultimately decides the amount of coolants needed and the associated costs. However, changes in density inevitably require different motherboard designs, adding to the complexity of deploying two-phase immersion cooling systems.
Eventually, to hasten the widespread adoption of two-phase immersion cooling, standardization is the key step upon which a corresponding supply chain and ecosystem can be developed. Without a supportive ecosystem, Chang admits that it is hard to estimate the market potential of two-phase immersion cooling.
For that reason, Wiwynn joined the Open Compute Project (OCP), a community founded by Facebook, Intel and others to facilitate redesigning hardware technology to improve compute infrastructure. Notably, China’s Alibaba, developing its own cloud service, also joined the OCP and came up with a single-phase immersion cooling system.
Meanwhile, a Chinese report estimates that immersion cooling will account for 40% of the liquid cooling market by 2025, with the remaining 60% represented by cold-plate (also known as Direct-To-Chip) cooling. Cold-plate cooling involves pumping fluid through cold plates attached to electronics, without making direct contact with them. It is currently mainstream.