Immersion Cooled ARM-Based Computer Clusters towards Low-Cost High-Performance Computing

This study aimed to investigate performance of ARM-based computer clusters using two-phase immersion cooling (IC) approach, and demonstrate its potential benefits over the air-based natural convection (NC) and forced convection (FC) approaches. ARM-based clusters were created using Raspberry Pi (Rpi) model 2 and 3, a commodity-level, single-board computer. The IC mode utilized two types of dielectric liquids, HFE-7100 and HFE-7000. Experiments involved running benchmarking programs Sysbench, and the combination of Sysbench and High-Performance Linpack (HPL), in order to quantify the key parameters of device junction temperature, operating frequency, execution time, computing performance, and energy consumption. Results indicated that the device junction temperature has direct effects on the computing performance and energy consumption. In the reference NC cooling mode, as the temperature approached the preset limit, the Rpi-3 cluster either decreased its operating frequency to save the internal cores from damage, leading to decrease in computing performance and increase in execution time and energy consumption (in the Sysbench test), or shut down all the nodes (in the HPL test). When the Rpi-3 cluster was exposed to a heavy load (combined Sysbench and HPL test) the IC modes compared to the FC mode indicated similar execution time, up to 3% higher computing performance, and up to 10% lower energy consumption. Although the results for the considered Rpi clusters do not represent a significant improvement, this study demonstrates that the two-phase IC method with its near-isothermal, high heat transfer capability would enable fast, energy efficient and reliable operation, particularly benefiting high performance and large scale computing applications where conventional air-based cooling methods would fail.