Optimizing HPL Performance on big.LITTLE Architectures: A Case Study with Odroid-MC1 Using ATLAS and OpenBLAS
Zdravko Krpić, Luka Loina, Filip Sušac · 2024
The rise of Edge Computing has brought Single Board Computing Clusters (SBCCs) back into the spotlight. With their efficient performance-to-power ratio, SBCs are becoming key players in Edge Computing infrastructure. Previous studies have explored the performance and power efficiency of SBCCs, yet often lack a systematic approach to benchmarking, especially for big.LITTLE ARM-based SBCs like the Odroid-MC1, Odroid$- \mathrm{N}2 +$, and ASUS’ Tinker Board 2 series. This study addresses the need for thorough benchmarking of such systems by employing the High Performance Linpack Benchmark (HPL). With precise tuning and improved cooling, we achieved up to 24% better performance on the same platforms using ATLAS and OpenBLAS, revealing more potential than initially anticipated and further improving their already notable performance-per-watt ratio.