Performance benchmarks¶

Cross-runtime FPS comparison for the perf_bench workload: a small async Pygame game used to measure sustained frame rate on local desktop, pygodide, and pygbag.

Latest results¶

Generated: 2026-07-05T22:48:07Z

Higher is better. Each web runtime uses a headed browser window (not headless automation) so numbers are closer to manual testing.

Environment¶

Browser settings¶

Both pygodide and pygbag benchmarks used:

• Engine: Chromium (headed)
• Frame-rate uncap: disabled (normal browser pacing)
• Launch args: --disable-background-timer-throttling, --disable-backgrounding-occluded-windows, --disable-renderer-backgrounding

Raw JSON: benchmarks/results/latest.json

Reproduce locally¶

From the repository root:

uv sync --dev
uv run playwright install chromium
uv run python benchmarks/run.py

This opens a real browser window for each web runtime, runs the benchmark game for 5 seconds of warmup plus 20 seconds of measurement, and writes:

• benchmarks/results/latest.json
• benchmarks/output/benchmark-chart.html
• docs/assets/benchmark-chart.html (copied for the docs site)

Regenerate only the chart from saved JSON:

uv run python benchmarks/run.py --chart-only

Useful options¶

# Skip runtimes you cannot run on this machine
uv run python benchmarks/run.py --skip-local
uv run python benchmarks/run.py --skip-pygbag

# Measure pygbag in Firefox (after: uv run playwright install firefox)
uv run python benchmarks/run.py --pygbag-browser firefox

# Headless automation (faster, but often unrealistic FPS)
uv run python benchmarks/run.py --headless

You need a graphical session for the default headed runs (same as local pygame).

Methodology¶

The benchmark app logs:

[benchmark] ready
[benchmark] done fps_mean=432.65 samples=8584