Limitations of traditional air cooling technology

Research shows that in traditional data centers, the cooling system may consume as much as 40% of the total electricity, which not only increases the operating costs but also runs counter to the global carbon reduction goal.
There is a physical upper limit on the heat dissipation capacity of an air-cooled system, and it is difficult to meet the demand for high-performance computing equipment with power density exceeding 30kW/ rack. In addition, the noise pollution and dust accumulation caused by large fans can not be ignored. These factors jointly promote the rapid development and application of data center liquid cooling technology.
Core advantages of data center liquid cooling technology
Liquid cooling technology dissipates heat by direct or indirect contact of a liquid medium with a heat source, and its heat conduction efficiency can reach 1000-3000 times that of air. This revolutionary cooling method has brought many obvious advantages:
First, the energy efficiency ratio has been greatly improved. A liquid cooling system can save about 30% of the total energy consumption, and PUE (energy efficiency) can be reduced to below 1.1, which is far better than the typical 1.6-1.8 of an air cooling system. Secondly, the heat dissipation capacity is significantly enhanced, and the power density of a single rack can be increased to more than 100kW, which meets the requirements of high-performance computing, such as AI and HPC. At the same time, the equipment runs quieter and the noise level is reduced by 20-35 decibels.
In the long run, the space utilization rate of a liquid-cooled data center is higher, which can reduce the floor space by about 50%, and it is easier to realize modular deployment and flexible expansion. These advantages make data center liquid cooling technology an inevitable choice for the future development of data centers.
Comparison of mainstream data center liquid cooling technology schemes
At present, data center liquid cooling technology is mainly divided into three types, each with its own characteristics and application scenarios:
1. Cold plate liquid cooling: the metal cold plate contacts with the heating parts, and the liquid circulates in the closed pipeline. This indirect contact method is compatible with existing equipment, and the cost of transformation is low. It is suitable for medium and high density scenes of 30-50kW/ rack, and it is the most widely used scheme at present.
2. Immersed liquid cooling: The equipment is completely immersed in non-conductive cooling liquid, which can be divided into phase change and non-phase change. Phase change immersion uses the latent heat of liquid vaporization, which has high heat dissipation efficiency and is suitable for ultra-high density environments above 50kW. Although the initial investment is large, the long-term operational benefit is remarkable.
3. Spray-type data center liquid cooling: accurately spray the cooling liquid onto the heating parts, and collect it by gravity reflux. This direct contact method has high efficiency, but the system is complex, and its application is relatively rare at present.
From the difficulty of implementation, the cold plate is most suitable for the reconstruction of existing data centers, while the submerged type is more suitable for the newly built high-performance computing center. In terms of cost, the initial investment of immersion may be 30-50% higher than that of a cold plate, but the total cost will be more advantageous after 5-7 years of operation.
Present situation and prospects of industrial application
Data center liquid cooling technology has shown great value in many fields. In the field of artificial intelligence, liquid cooling is widely used in large AI training clusters. For example, the AI supercomputer of a well-known technology company achieved the ultimate PUE of 1.08 through immersion cooling. Cloud computing service providers have also begun to deploy liquid-cooled servers, and Microsoft's Natick submarine data center project is a typical representative.
In the financial industry, high-frequency transactions have extreme requirements for low latency, and the liquid cooling system can increase the operation frequency by about 15% by lowering the chip temperature. In the field of scientific computing, six of the top 10 supercomputing centers in the world have adopted liquid cooling technology, including "Fuyue" in Japan and "Frontier" in the United States.
According to the forecast of market research institutions, by 2027, the global data center liquid cooling market will exceed 10 billion US dollars, with a compound annual growth rate of over 30%. With the development of new technologies such as 5G, Metaverse, and autonomous driving, the demand for computing power will only continue to grow, and liquid cooling technology will surely become the infrastructure standard of data centers.
Implementation challenges and solutions
Despite the broad prospects, the promotion of data center liquid cooling technology still faces some challenges:
Technically, the choice of coolant is very important, and it is necessary to balance thermal conductivity, insulation, environmental protection, and cost. At present, mineral oil, fluorinated liquid, or deionized water are used in the mainstream, and new nanofluids are being developed. Sealing and corrosion protection are also difficulties in system engineering, especially for submerged systems.
In terms of cost, although the long-term energy-saving benefit is obvious, the initial investment is 50-100% higher than that of the air-cooled system. It is necessary to lower the adoption threshold through innovative financing models such as energy-saving benefit sharing. The standardization process is also accelerating, and organizations such as OCP and ODCC have issued relevant design specifications.
Operation and maintenance transformation is another big challenge, which requires retraining technicians and establishing new maintenance processes. Some suppliers began to provide life-cycle services, from design, deployment, to operation and maintenance.
With the development of technology and the appearance of the scale effect, these challenges are gradually being overcome. It can be predicted that data center liquid cooling technology will change from "high-end option" to "mainstream standard" in the next 5-10 years, and redefine the energy efficiency benchmark of data centers.