Thermal Characterization of Single-Phase Immersion Cooling Capability

As component and IT equipment level power and power densities continue to increase, traditional air-based cooling technologies are reaching their limits of thermal management in high performance data center servers and Edge applications. To meet these increasing performance demands and reduce total cost of ownership, more cost effective, capable, and efficient liquid cooling technologies such as cold plates and immersion are being actively investigated and implemented in many deployments. Single-phase immersion cooling is one such technology that involves immersing the entire IT equipment in a tank filled with a dielectric liquid. The dielectric liquid flows through the IT equipment, providing improved cooling compared to air via natural or forced convection. This technology has been seeing increasing interest in the industry in recent years due to its capability to cool high power servers at low cost and high efficiency as well as provide a holistic cooling solution for the entire IT equipment/server. Most installations of single-phase immersion tanks in the industry use pumps that provide circulation of the dielectric liquid through the heat exchanger to maintain tank liquid inlet temperature. However, these tanks have unique manifolds distributing the flow into the tank, which generates a unique flow inlet boundary condition to the servers for each tank design. On the market today, there are tanks that span from using only natural convection for cooling the servers, while most of the tanks use a mix between natural and forced convection. Furthermore, there is no means to measure the flowrates or understand the flow boundary conditions for each server or the flow going through the CPU heatsink. Hence, the design of efficient thermal solutions or performance prediction capabilities for such systems becomes challenging. The current paper describes thermal characterization of immersion optimized heatsinks under known and controlled boundary conditions. The results described here show the performance of the optimized heatsink in a liquid tunnel under natural convection and under forced convection regimes at a wide range of flowrates for an Intel CPU. Under natural convection, a large dependency of the heatsink cooling capability is observed on the operating power level of the CPU. As the flow rate increases, this dependence is reduced while the cooling capability improves significantly. The results show comparison of the thermal performance for air and immersion optimized heatsinks used in immersion in different flow regimes for known and controlled thermal and flow boundary conditions.