Measure the matrix.
A fixed scene makes backend and quality changes comparable. Run the same load ladder, capture timing and renderer evidence, then repeat on the companion backend URL before drawing conclusions.
Preparing the fixed actor pool and quality matrix...
- scenario
- -
- visible
- -
- quality
- -
- sensor
- -
- live p95
- -
- draws
- -
The runner warms each row before collecting a fixed sample. GPU timing is used when the backend exposes timestamp-query; CPU submission is the labeled fallback.
Rows stay empty until the fixed matrix is run.
| scenario | backend | sensor | p95 | draws | RTs |
|---|
The fixed pool remains scene-owned; graph resources are rebuilt only when a row changes quality.
This is a same-browser backend matrix, not a cross-device benchmark. Run the default WebGPU route and the ?forceWebGL=1 companion, then compare rows only after recording device and browser context.