Comparison of five server thermal management technologies, single-phase DLC is more effective
Recently, at a technical lecture organized by DCD, Dell technical expert Dr. Tim Shedd revealed in a special report titled "Performance comparison of five server thermal management technologies in data centers" that leading data center cooling technologies include air cooling, single-phase immersion, two-phase immersion, two-phase direct liquid cooling Comparison of single-phase direct liquid cooling (DLC, cold plate) research and testing.
By 2025, the power of CPU or GPU chips will generally reach up to 500 W, and artificial intelligence and machine learning have pushed the GPU power up to 700 W, with an expected 1000 W in the near future. More importantly, while increasing power, lower chip packaging temperatures and smaller temperature differences are required to ensure the normal operation of the chip. The more advanced the semiconductor technology, the smaller the transistor size, and the larger the leakage current, which increases exponentially with temperature. Therefore, the challenge of thermal management systems intensifies.
A few years ago, when the processor TDP was around 250W, these five thermal management technologies were able to provide very efficient cooling for typical data center cabinets, such as deploying 32 dual 250W rack mounted servers in data center cabinets. For a 2U rack mounted server, the report observed a temperature difference of approximately 26 ℃ between the chip packaging and the air flowing through the server. Therefore, it is quite reasonable to maintain the chip temperature around 51 ℃ with only 25 ℃ cold air. At this point, the efficiency of air cooling for a single server is equivalent to single-phase immersion cooling.
Currently, the power of a single processor has increased to 350W to 400W, and the temperature difference required to remove the heat from the chip to the cooling water of the facility is constantly increasing. Similarly, a cabinet with 32 dual 350W rack mounted servers is deployed for cooling. The temperature difference between the air and chip packaging during air cooling (1U) exceeds 50 ℃, which means that when the server is cooled with 25 ℃ cold air, the temperature of the processor will reach 75 ℃, approaching the operating temperature limit of the processor. It can be seen that the current processor power has increased to 350W-400W, and air cooling is very close to the actual limit, which means that cooler air is usually required, thereby exacerbating cooling energy consumption.
In the next two to three years, the TDP of processors will generally increase to 500W, and air cooling faces considerable challenges, requiring innovative heat sink design methods or relying on larger sizes to allow more air to enter and cool the processors. At this point, the temperature difference between air cooling (1U), single-phase immersion cooling, and chip packaging exceeds 60 ° C; The two-phase immersion cooling is still effective, and the temperature difference will increase to about 34 ℃; The temperature difference range between two-phase DLC and single-phase DLC (1 lpm) is not significant, about 25 ℃; The temperature difference range of single-phase DLC (2 lpm) is smaller, about 17 ℃.
Compared with the other four data center cooling methods, single phase direct liquid cooling (DLC) has the highest thermal efficiency and can provide a potential way to achieve better sustainability and improve efficiency.
