Natural and forced convection heat transfer characteristics of single-phase immersion cooling systems for data centers

Abstract

Improving power usage effectiveness (PUE) in data centers becomes a key challenge for carbon neutrality and carbon peaking strategies. In this context, this paper investigates the single-phase immersion cooling (SPIC) system that promises to improve PUE. Three-dimensional mathematical models of pump-driven and buoyancy-driven immersion cooling tanks are established and numerically solved. The thermal characteristics, flow resistance, and energy consumption levels of immersion cooling tanks in three operating modes are investigated. Moreover, the roles of volume flow rate, inlet temperature, and device power in immersion cooling characteristics are analyzed. The results indicate that the heat dissipation of buoyancy-driven SPIC systems is less effective than pump-driven ones. Compared to buoyancy-driven SPIC systems, the average temperature of immersion coolants and PUE in pump-driven SPIC systems reduces by 55.5 and 11.6%, respectively. Interestingly, although natural convection in mixed convection operation enhances convective disturbance, it induces temperature stratification that weakens temperature homogeneity. Increases in inlet temperature and device power both worsen the immersion cooling performance. However, increasing inlet temperature improves temperature uniformity and reduces flow resistance losses, while increasing device power has the opposite effect. Besides, the increased flow rate enhances the convective heat transport and temperature uniformity but increases the flow resistance loss.