Measuring the result

npx gameface-rasterize measure dist

gameface-rasterize measure - 3 alternating pairs

  live CSS  GPU p50 3.29 ms +/- 0.01    baked  GPU p50 2.19 ms +/- 0.00   -33.4% (n=4)
            Paint p50 0.165 ms             Paint p50 0.148 ms

Why frame time cannot tell you

The Player paces frames at a fixed cap of roughly 60 ms. An empty page reads 60 ms, and so does a page with hundreds of blurred, shadowed, gradient-filled boxes. Anything that measures only frame arrival cannot separate them.

What changes is the engine’s own Coherent_GPU counter from a CDP trace, which is per-frame GPU time. That is what this command reports.

What it does for you

• Alternates which variant runs first, so a first-run advantage cannot be mistaken for the effect.
• Repeats the run and reports the spread. If the gap is smaller than the run-to-run variation, it says so instead of letting you quote a meaningless number.
• Invalidates the page every frame, because cohtml repaints dirty regions only and a settled static page measures almost nothing.

npx gameface-rasterize measure dist --repetitions 5 --trace-seconds 6

Reading it

The delta is the number that matters, not the absolute level. Levels depend on the machine, driver, and what else is running.

A decoration that nothing disturbs is close to free on this engine, because it is not being redrawn. If your page is static, measure will correctly report that baking it changed very little.