Performance Evaluation of HPC Benchmarks on a RISC-V Based Cluster
Rushikesh Jadhav, Surendra Billa, Yogeshwar Sonawane, Sanjay Wandhekar · 2025
With the growing momentum of open hardware, the exploration of its role in high-performance computing (HPC) has become both timely and essential. This paper presents the design, deployment, and performance evaluation of a lightweight HPC cluster built entirely on RISC-V, a fully open-source processor architecture. The cluster comprises of four StarFive VisionFive boards interconnected via Gigabit Ethernet and runs a custom-built Fedora Linux distribution tailored for the RISC-V platform [1] [2]. A central focus of this work is the native compilation and integration of MPI libraries, along with the execution of High Performance Linpack (HPL) and High Performance Conjugate Gradient (HPCG) benchmarks, to assess parallel performance across varying node configurations [3]. Experimental results shed light on the portability of HPC software stacks to emerging architectures, demonstrating the feasibility of executing parallel workloads on modest, energy-efficient RISC-V hardware. This study highlights both the opportunities and the practical challenges of developing scalable, open-source HPC systems utilizing open silicon and open software.