Optimization of heat sinks for data center server CPUs cooled via single-phase immersion cooling

초록

As modern data centers increasingly support high-performance computing and AI-driven applications, advanced thermal management becomes ever more critical. Single-phase immersion cooling offers an effective and energyefficient solution for managing high thermal loads. This study focuses on optimizing heat sink geometries and coolant conditions for a dual-CPU server with 24 DIMMs immersed in FC-40 dielectric fluid. Using response surface methodology with a Box-Behnken design (RSM-BBD), this study systematically evaluates the effects of design and operational variables on CPUs temperature, temperature uniformity, and the power required for coolant pumping and thermal restoration. The sensitivity analysis reveals that the coolant flow rate exerts the greatest influence across all optimization criteria. Apart from CPUs temperature, geometric parameters have a more pronounced effect on the remaining optimization metrics than the coolant inlet temperature. The optimized model satisfies all metrics for effective immersion cooling. Specifically, under a thermal design power (TDP) of 400 W, the CPUs temperature is maintained below 60 degrees C, with the temperature difference between the CPUs kept below 5 degrees C. Furthermore, the model can handle higher TDPs of up to 600 W while keeping the CPUs temperature below 85 degrees C. It also performs effectively for servers equipped with advanced chips, such as GaN-based devices, at TDPs up to 800 W. Among all the coolants evaluated, NovecTM 7200 demonstrates the ability to maintain the CPUs temperature at 58.3 degrees C while requiring the lowest pumping power. Furthermore, the proposed correlations accurately predict the thermal behavior of heat-generating components, with an average absolute deviation of less than 10 %.

키워드