Does SIMD help your browser?
Compare complete physics steps, not graphics FPS. Same scenes, same precision, paired runs—with controls to show measurement noise.
Your setup
Checking…Reading available browser and hardware hints…
Collected information & privacy
Browser/OS versions, exposed architecture/device hints, logical processor and RAM hints, GPU renderer, screen, timer precision and battery status where available. A GPU name is not a CPU name. Missing/reduced data stays unknown.
No automatic result upload, analytics, geolocation or external hardware lookup. The latest run is saved only in this browser. Download the JSON and share it yourself. Most iOS browser apps share WebKit, so app names do not necessarily mean different engines.
Loading…
Run the test
Keep this tab visible, avoid other heavy work, and keep power settings consistent. Switching tabs or locking the screen stops the run and retains partial results. No physics is rendered while timing.
Each mode includes correctness checks, warmup, duration calibration and alternating paired measurements. Allow several minutes, especially on phones. Full uses more repeats and controls.
Results
Results will appear here. Download each run before starting another, then send the JSON file back for analysis.
Negative Δ time is faster. “Inconclusive” means the bootstrap interval does not clear the A/A noise margin (at least 1%). Small-sample intervals are exploratory—not a universal confidence guarantee. ORIGINAL and SIMD use different contact schedules; trajectories need not be identical.
The primary comparison uses ORIGINAL and SIMD in the same SIMD-enabled module. A/A repeats ORIGINAL to estimate noise. Module Δ compares ORIGINAL across scalar-only and SIMD-enabled binaries; it is not a SIMD solver speedup.
Protocol, correctness & raw diagnostics
No run yet.