In the AI era, is liquid cooling the only way to dissipate heat
According to predictions, the demand for computing power in the future will grow rapidly, and global intelligent computing power is expected to reach 105ZFLOPS by 2030, an increase of 500 times compared to 2020. According to IDC data, the scale of China's intelligent computing power reached 268.0EFLOPS (1018 floating-point operations per second) in 2022, and it is expected that the scale of intelligent computing power will enter the ZFLOPS level by 2026, reaching 1271.4EFLOPS.
With the continuous improvement of computing power, it is necessary to greatly improve chip performance to support it, which brings another major challenge, namely the thermal design power consumption (TDP) of chips. At present, the power consumption of the CPU has reached 350-500W, while the power of high-end GPUs and switch ASIC chips has reached over 700W. When the chip power is further increased to over 700W in the future, the heat dissipation design of the chip will become a serious problem.
At present, the most widely used method for chip cooling is air cooling, which means that a heat sink with good thermal conductivity is attached to a chip with high heat generation, and a small fan is fixed above the heat sink. The airflow generated by the high-speed rotation of the fan takes away the heat on the heat sink. With the continuous improvement of chip power, after exceeding 300W, the effect of using traditional heat sinks for heat dissipation is no longer obvious. Liquid cooling cooling technology is considered an ideal cooling solution in the AI era.
Liquid cooling technology can be divided into three types based on its different cooling methods: cold plate liquid cooling, immersion liquid cooling, and spray liquid cooling. Cold plate liquid cooling is an indirect contact type of liquid cooling that fixes the cold plate on the heatsource, and the liquid flows inside the cold plate to transfer heat away from the equipment, achieving heat dissipation. Spray liquid cooling is a technology that dissipates heat by spraying coolant onto the surface of IT equipment, with relatively low heat dissipation efficiency.
Immersion liquid cooling is considered the most mainstream and capable liquid cooling technology for large-scale deployment in liquid cooled data centers. It has the following advantages: firstly, it has high thrmal efficiency, as immersion liquid cooling directly immerses IT equipment in the coolant, allowing for comprehensive contact with heat sources, greatly improving cooling efficiency; Secondly, the noise reduction effect is good, as IT equipment is completely immersed in the coolant, which can reduce the noise emitted by IT equipment; The third is energy conservation and environmental protection. Immersion liquid cooling does not require the use of a large number of fans, which can reduce power consumption and carbon dioxide emissions. According to relevant data estimates, compared to air cooling, liquid cooling can save 20% -30% of the electricity required for the entire server operation.
Immersion liquid cooling technology will inevitably become the mainstream cooling technology in the AI era in the future. However, the current liquid cooling technology and products are still in a relatively preliminary application stage, but with the development of AI, data centers, and other applications, the application and popularization of liquid cooling technology will be accelerated. With the further development of immersion liquid cooling technology, it may become the 'only' choice for data center cooling in the future.
