Computational study of single-phase immersion cooling for high-energy density server rack for data centers

초록

As the cooling difficulty and power consumption of data centers increase in line with the growing demand for high-performance data centers, the importance of thermally effective and energy-efficient cooling of data centers has become apparent. However, the research on managing high heat generating electronic components in a server rack remains insufficient, failing to address the recent developments in the high-energy density server racks. This study narrows that gap through the computational analysis of the cooling systems for the server rack. Considering the varying impacts of different parameters of the cooling methods, this study comprehensively analyzed immersion cooling on a high-energy density server rack with high-performance components by comparing multiple cooling methods. Under equal power consumption, immersion cooling can achieve the lowest component temperature with the highest stability at various central processing unit (CPU) and memory thermal design powers, power consumptions, inlet temperatures, and server heights. For 800 W CPUs and 20 W memories, immersion cooling is the only method capable of maintaining component temperatures below the target temperature in an extreme condition, maintaining a CPU and memory temperature of 62.0 degrees C and 44.2 degrees C. Additionally, the temperature variation of the CPU was the lowest at 9.1 K, while those for air and hybrid cooling were 41.6 K and 10.7 K, respectively. The impact of various dielectric fluids on the component temperature for immersion cooling was investigated with a focus on the thermal properties of the fluid along with the thermal and hydraulic characteristics of the heatsink simplified through a porous media model.

키워드