Changelog
A record of every change to VolumeShader BM. Adjusting the test parameters changes whether scores stay comparable, so preset changes are listed separately.
1.1 · First public release: static pages and the full test flow
Released- The homepage, the pre-test environment check, the full-screen run page and the in-page result overlay are all live. A single test no longer moves you across multiple pages.
- Four tiers — Simple, Standard, Advanced and Extreme — change only the fractal iteration count and the ray-marching step cap; every other parameter is identical on every device. Standard keeps the classic original parameters (5 iterations × 1002 steps) and is the only tier that can be ranked.
- Timing uses a three-stage window: the first 5 seconds of warm-up are discarded, the middle 60 seconds count, and the last 5 seconds of wind-down are discarded. That removes first-compile cost and end-of-run frame drops from the result.
- The result reports average FPS, max FPS, 1% low, min FPS and a per-second frame-rate curve, and flags three conditions: refresh-rate capping, excessive drift, and too few valid frames.
- The whole run happens locally in your browser. No scores and no device information are uploaded.
preset v1.0 · Test parameters frozen
Released- The render canvas is fixed at 1920×1080. Device pixel ratio only affects the size you see on the page, not the number of pixels actually rendered, so a high-DPI screen is not penalised.
- The scene is a Mandelbulb ray-marching shader — the same rendering kernel as the original poison mushroom test. Each frame waits for the GPU to finish before timing, so the measurement is real draw cost rather than command submission.
- The leaderboard ranks by average FPS and keeps only the best result per GPU model, so repeated runs cannot inflate a position.
1.2 · Leaderboard and score collection
Planned- Open the unified Top 200 leaderboard and add a minimal write endpoint: validate field formats, rate-limit submissions, write to JSON.
- Tag devices on top of the collected results to reduce the comparability problem of mixing desktops, laptops and phones in one table.
- Fill in leaderboard-related Q&A and structured data so the rankings page can be indexed normally.