Benchmarks

The benchmark pages report the live per-tick Python-facing path:

• advance(...): update state only.
• update(...): update state and return a Python value/result.

These runs were collected on 2026-07-19 over SSH. Hostnames are intentionally omitted from the documentation; each full result page is identified by CPU type.

Each CPU page carries its own system metadata and full 247-algorithm latency table.

Run command:

python benchmarks/benchmark_readme.py --samples 50000 --repeat 5 --warmup 1 --output <benchmark-output.md>

Results By CPU

• Apple M4 Max: macOS-26.5.1-arm64-arm-64bit-Mach-O, arm64, Python 3.14.5, NumPy 2.5.1, RTTA 0.2.3; median advance(...) 29.9 ns/update, median update(...) 41.1 ns/update.
• Intel Xeon 6975P-C: Linux-7.0.0-1004-aws-x86_64-with-glibc2.43, x86_64, Python 3.14.4, NumPy 2.5.1, RTTA 0.2.3; median advance(...) 39.2 ns/update, median update(...) 52.2 ns/update.
• Loongson-3A6000: Linux-5.4.18-167-generic-loongarch64-with-glibc2.28, loongarch64, Python 3.14.4, NumPy 2.5.1, RTTA 0.2.3; median advance(...) 104 ns/update, median update(...) 139 ns/update.