AMG Exascale Benchmark: 100,000-Rank Performance Data from Aurora Supercomputer

100,000 MPI ranks executed the AMG benchmark on the Aurora supercomputer at Argonne Leadership Computing Facility. Dragana Grbic collected this 3.0 GB database using the HPCToolkit measurement suite in April 2026. It contains fine-grained calling context trees, performance profiles, and detailed execution traces for each rank.

Use Cases

• Analyze scaling efficiency based on performance profiles across 100,000 MPI ranks
• Identify performance bottlenecks using fine-grained calling context trees (CCTs)
• Study communication patterns and load balancing from detailed execution traces
• Validate performance models for exascale systems using the AMG benchmark results

Strengths

• Massive-scale execution data from 100,000 MPI ranks on a leading exascale supercomputer
• Data collected using the established HPCToolkit measurement suite
• Database is structured for high-throughput ingestion and optimized for the hpcanalysis framework
• Provides multiple performance data types: CCTs, profiles, and traces for each rank

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment

Provenance

Source

Argonne Leadership Computing Facility (ALCF)

Collection Method

Collected using the HPCToolkit measurement suite during benchmark execution

Time Range

null

Freshness

Last updated 2026-04-26 06:27:45; freshness should be verified

Geography

null