CanvasLab

Canvas2D vs WebGPU

Same rectangle document, same camera path, same benchmark duration. Measurements represent synchronous main-thread render-call cost and observed requestAnimationFrame cadence — not raw GPU execution time.

initializing

Current workload: 4,000 plain root rectangles.

Canvas2Dproduction backend
WebGPUexperiment backend
How to read thisLower render-call time means less synchronous main-thread work. P95 is more useful than a single fastest frame because it exposes slower frames.Observed FPS is capped by display refresh rate, so two renderers may both show about 60 FPS even when their CPU cost differs substantially.WebGPU render-call time includes document-to-instance preparation, buffer upload, command encoding, and queue submission. It does not wait for GPU completion.This compares raw renderer backends. CanvasLab's production Canvas2D path also benefits from viewport culling and layered invalidation.