Statistical analysis across multiple runs.
Support for arbitrary shell commands.
Constant feedback about the benchmark progress and current estimates.
Warmup runs can be executed before the actual benchmark.
Cache-clearing commands can be set up before each timing run.
Statistical outlier detection to detect interference from other programs and caching effects.
Export results to various formats: CSV, JSON, Markdown, AsciiDoc.
Parameterized benchmarks (e.g. vary the number of threads).
Cross-platform