MICM Performance Benchmarks

Every push to main runs the Chapman mechanism benchmark and adds a point to these charts.

Charts

• Instruction counts CIRRUS a10 GPU runner
• Wall-clock timing CIRRUS a10 GPU runner
• Instruction counts ubuntu-latest
• Wall-clock timing ubuntu-latest

How to read them

Instruction counts come from callgrind. They do not change with machine load, thermals, or turbo, so they show a hot-path change that the wall-clock time hides in noise. Use the timing charts for absolute cost and the instruction charts for regressions.

Each machine keeps its own series. Do not compare a value from one machine against a value from another.

Documentation

The MICM documentation moved to micm.readthedocs.io. Old documentation links on this site redirect there.