Experimental Evaluation of Cooling Loop Configurations for Server-Level Direct-to-Chip Liquid Cooling Systems

Abstract

Effective heat removal has become a critical challenge in modern data centers as increasing computational demands from artificial intelligence and machine learning push GPUs, CPUs, and network switches toward unprecedented power densities. These heat loads often exceed the capabilities of traditional air-cooling solutions, accelerating the adoption of advanced thermal management strategies. Among them, direct-to-chip liquid cooling has emerged as a promising approach, using cold plates mounted directly on high-power components to enable efficient heat dissipation while reducing energy consumption and operational costs. This study experimentally evaluates the thermohydraulic performance of liquid-to-liquid (L2L) cooling systems employing a Coolant Distribution Unit (CDU). The CDU supplies a secondary coolant (25% propylene glycol) to server-level cooling loops, enabling heat exchange between facility water and cold-plate coolant. To emulate high-power server conditions, thermal test vehicles (TTVs) were developed with heaters of two footprints, generating distinct heat flux levels representative of modern computing hardware. The TTVs were characterized under natural convection to establish a baseline and under forced convection to assess high-power operation. Three cooling loop configurations featuring parallel and series flow paths were tested under identical thermal and hydraulic boundary conditions. A systematic methodology was applied to quantify temperature uniformity while isolating discrepancies arising from loop topology versus device-level assembly effects. The results provide a comparative assessment of cooling loop performance and offer practical, data-driven guidance for optimizing liquid-to-liquid cooling architectures in next-generation high-power data centers.