About
A browser GPU benchmark that tries to be simple and comparable.
Current version: 1.1 (site display and core benchmark flow)What we are solving
Too many GPU tests need downloads, accounts, or separate tables for phones, laptops and desktops. This tool runs in a tab, uses one shared preset for comparable frame rates, and shows device info alongside so you can judge comparability yourself.
Technical background
The renderer comes from the open-source Volumeshader_bm project: a WebGL fragment shader draws a Mandelbulb-style fractal in real time with per-pixel ray marching, producing a heavy, repeatable graphics load. Chinese communities often call it the 毒蘑菇测试 (literally "poison mushroom GPU test"). We added environment checks, per-frame timing, stable-window scoring and a result view. The rendering kernel itself is unchanged.
Relationship to the original project
This is an independent community project. It is not the official Volumeshader_bm site and is not affiliated with the original author or any GPU manufacturer. The original demo page is at cznull.github.io/vsbm. The source repository and licence details are still being verified, and attribution will be added once confirmed.
How the score is calculated
The standard run lasts 70 seconds: 5 seconds of warm-up (shader compilation and GPU clock ramp-up), 60 seconds of measurement, and 5 seconds of wind-down. Every frame interval and GPU duration is recorded, then average FPS, 1% Low, minimum FPS and the FPS curve are computed once at the end, with checks for stability and refresh-rate caps. We publish the rules here so readers can verify the method and judge whether a result is comparable.
Roadmap
1.1: site, environment checks, full benchmark flow and local scoring; leaderboard not populated yet.
1.2: result submission, leaderboard updates, preset refinement and data-quality rules.
Later: device labels, filtering and comparison based on collected data.
Contact
Suggestions, corrections or collaboration: [email protected]. If a phrase or a metric definition looks wrong, please tell us.