First Experiences in Performance Benchmarking with the New SPEChpc 2021 Suites - Measurement Data

Modern High Performance Computing (HPC) systemsare built with innovative system architectures and novelprogramming models to further push the speed limit of computing.The increased complexity poses challenges for performanceportability and performance evaluation. The Standard PerformanceEvaluation Corporation (SPEC) has a long history ofproducing industry-standard benchmarks for modern computersystems. SPEC’s newly released SPEChpc 2021 benchmark suites,developed by the High Performance Group, are a bold attempt toprovide a fair and objective benchmarking tool designed for state-HPC, SPEC, HPG, SPEChpc 2021, benchmarks,performance benchmarking and analysis, heterogeneity, offloading,MPI, MPI+X, OpenMP, OpenACCof-the-art HPC systems. With the support of multiple host andaccelerator programming models, the suites are portable acrossboth homogeneous and heterogeneous architectures. Differentworkloads are developed to fit system sizes ranging from a fewcompute nodes to a few hundred compute nodes. In this work wepresent our first experiences in performance benchmarking thenew SPEChpc2021 suites and evaluate their portability and basicperformance characteristics on various popular and emergingHPC architectures, including x86 CPU, NVIDIA GPU, and AMDGPU. This study provides a first-hand experience of executingthe SPEChpc 2021 suites at scale on production HPC systems,discusses real-world use cases, and serves as an initial guidelinefor using the benchmark suites This deposit contains all measurement data for the results shown in the paper as well as all platform setups (compilers and flags) and also the errors encountered on the exascale test system "Spock"