| Title |
A Numerical Analysis for Determining Stable Operating Temperature Limits by Coolant Type in Single-Phase Immersion-Cooled Data Centers |
| DOI |
https://doi.org/10.5659/JAIK.2026.42.10.347 |
| Keywords |
Immersion Cooling; Data Center; Dielectric Fluid; Junction Temperature; Computational Fluid Dynamics; Thermal Safety |
| Abstract |
High-density data centers driven by AI and HPC workloads are reaching rack power densities that exceed the limits of conventional air
cooling, making single-phase immersion cooling a promising alternative. However, the stable operating range of such systems depends strongly
on the dielectric fluid and the coolant supply temperature, which has not been systematically quantified. This study uses conjugate heat
transfer CFD analysis to determine the safe operating temperature limits of a 50 kW immersion-cooled GPU server (80 H100-class GPUs at
625 W each; junction-temperature limit of 87°C) for three single-phase fluids, FC-40, mineral oil, and synthetic ester, at coolant inlet
temperatures of 30, 40, and 50°C under a fixed 5°C inlet-to-outlet temperature difference. At inlet temperatures of 30°C and 40°C, all three
fluids keep the maximum GPU junction temperature below the 87°C limit, whereas at 50°C only FC-40 remains within the limit (85.8°C);
mineral oil (87.9°C) and synthetic ester (89.1°C) exceed it. The upper-to-lower junction-temperature gradient increases in the order FC-40 <
mineral oil < synthetic ester, governed primarily by volumetric heat capacity, viscosity, and thermal diffusivity rather than by flow rate. The
results indicate that the safe supply-temperature envelope is fluid-dependent: FC-40 offers the widest thermal margin but is subject to PFAS
phase-out, while non-PFAS fluids remain safe only up to a 40°C supply temperature. These findings provide quantitative guidance for the
cooling-system design and operation of high-density immersion-cooled data centers. |