SPEChpc 2021 Benchmark: Performance Measurement Data from HPC Systems

Holger Brunst from Technische Universität Dresden provides measurement data from benchmarking the new SPEChpc 2021 suites. The data captures performance and portability characteristics across diverse HPC architectures, including x86 CPUs, NVIDIA GPUs, and AMD GPUs. This deposit includes platform setups, compiler flags, and errors encountered during testing on the exascale system 'Spock'.

Use Cases

• Compare performance portability across homogeneous and heterogeneous architectures based on the described benchmark suites.
• Analyze basic performance characteristics of x86 CPU, NVIDIA GPU, and AMD GPU systems using the provided measurement data.
• Evaluate the effectiveness of different programming models (MPI, OpenMP, OpenACC) for performance benchmarking as mentioned in the description.
• Study scaling behavior from a few to hundreds of compute nodes using the different workload sizes described.

Strengths

• Data originates from first-hand execution of the industry-standard SPEChpc 2021 benchmark suites.
• Includes platform setup details such as compilers and flags, which are critical for reproducibility.
• Covers testing on the named exascale test system 'Spock', providing real-world scale data.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and dataset size are unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

Technische Universität Dresden

Collection Method

Benchmark execution and measurement on production HPC systems.