Analysis of Thermal Performance Metrics—Application to CPU Cooling in HPC Servers

Abstract

Optimization of heatsinks has a critical importance for the energy-efficient cooling of high-performance computing (HPC) servers to meet reliability, performance, and cost targets. This article provides a detailed application of the effectiveness as a performance metric to provide a qualitative comparison of the CPU heatsink solutions. First, the use of effectiveness as a metric to provide an absolute performance of heatsinks on a simple scale of 0–1 is highlighted. Then, the effectiveness and thermal resistances are compared to explore the limits of air cooling and the relative merits of liquid cooling. Second, a response surface method (RSM) for the heatsink optimization is presented. Furthermore, a response surface optimization (RSO) based on the design of experiments (DOE) is performed on the numerical models of the heatsink and cold plates using FloTHERM computational fluid dynamics (CFD) software. The simulation results obtained at 45 °C inlet are used to provide an objective comparison among the optimum heatsink and cold plates. The simulations show that heatsinks have the effectiveness of 0.65–0.75 and a cooling capability of 150–250 W. The cold plates have an effectiveness of 0.25–0.30 and a cooling capability of 325–425 W. Furthermore, the simulation results are compared with benchmark validation studies. Counterintuitively, the higher effectiveness obtained for the air-cooled heatsinks indicates limited scope for further improvement, whereas the cold plates (with lower effectiveness) provide great scope for further improvements to meet future cooling challenges.