The Open Edge and HPC Initiative (OEHI) has joined forces with the Gesellschaft für wissenschaftliche Datenverarbeitung mbH Göttingen (GWDG) to develop an easy-to-use software ecosystem focused on Scalable Matrix Extensions (SME) from Arm. SME is a recent instruction set extension comprising instructions that optimise matrix and tensor computations. These instructions will significantly increase the efficiency of processor architectures for machine-learning workloads. Significant research efforts are needed to facilitate the exploitation of these new instructions for scientific computing applications.

This collaboration aims to empower researchers with tools to design, implement, and test SME-enabled applications, fostering a growing knowledge base on SME-supported Arm architectures. The project also emphasises community engagement through workshops, webinars, and hackathons, helping researchers and companies to optimise for SME in their applications. A key objective is the development and benchmarking of SME-optimized numerical libraries, like BLAS and FFT, to demonstrate tangible performance gains in scientific HPC and AI applications.

The partnership will deliver an integrated development environment for exploring SME instructions comprising compilers, emulators, and simulators and establishing community channels. Furthermore, the project will deliver technical white papers, scientific publications and presentations at prominent conferences. Initial efforts focus on setting up a comprehensive SME development environment on OEHI’s HAICGU cluster (https://haicgu.github.io/), complete with suitable compilers and emulators, while also creating recipes for development on other HPC clusters. Future phases will explore performance optimizations, Proof of Concepts (PoCs), and benchmarking, culminating in shared results and insights through community meetings, promotional materials, and conference publications. This initiative marks a significant step towards building a vibrant SME-enabled HPC ecosystem in Europe.