Measuring the performance of CPython
summary
CPython performance is measured with the pyperformance benchmark suite (≈100 benchmarks), Linux perf sampling to attribute time to interpreter/memory/GC/library, and a GitHub Actions–backed continuous benchmarking on GitHub Actions fleet of bare‑metal runners. Recent work claims CPython 3.11 yields 20–60% speedups for some workloads, but typical noise is about ±1%, so CPU pinning, thermal control, warmup, hierarchical statistical tests, and donating representative benchmarks matter for trustworthy results.
AI-written summary · about
chapters (9)
- 0:00CPython performance measurement
- 1:50Benchmarking essentials
- 3:35Benchmark types in Python
- 13:34Benchmarks reveal changes
- 19:38Reliable benchmarking stopwatch
- 29:46Continuous benchmarking
- 33:17Optimizing benchmarking process
- 42:26Q&A noise reduction
- 47:07Parallelism and free threading
related talks
29mHow we are making CPython faster. Past, present and future.
57mPython and PyPy performance (not) for dummies
01h 03mA tour of CPython's runtime
22mPyPy in production
26mPython Performance at Scale - Making Python Faster at Instagram
33mWhy is Python slow?
39mCPython, Grumpy, PyPy - When, How, Why
48mPython Performance: Past, Present and Future