SimulationWater ripples, wakes & caustics
The wave equation as compute on a ring of cells that follows the camera, a smooth Catmull-Rom read, an absorbing edge and a coarse second field for long wakes…
Roads as plain data (nodes, edges drawn through bends and rounded to arcs, a type on each) laid out into junctions with kerb returns and setbacks, then…
Measured 13 November 2026 on RTX 4060 Laptop, 13th Gen Intel(R) Core(TM) i9-13900H, Chrome 157, 1920×1080 at 1×. 600 frames × 3 rounds, median kept (rounds ranged CPU 0.29–0.34 ms). What ran: Measured running at 20 new cars a minute on each road in (about 100 cars at once): every car decides and moves every 1/20 s of simulated time, and the frame includes the Canvas 2D drawing of the cars, lights and overlays over a cached road layer. Phones: tiers are being measured next.
| Machine | GPU ms | CPU ms | CPU p95 |
|---|---|---|---|
| RTX 4060 Laptop · 13th Gen Intel(R) Core(TM) i9-13900Hreference · measuring tool | not timed | 0.31 | 0.48 |
| apple · Apple A19 Proin a browser tab · 16 November 2026 | not timed | 0.68 | 1.36 |
| Intel Arc 140V (Xe2-LPG) · Intel Core Ultra 7 258Vin a browser tab · 15 November 2026 | not timed | 0.50 | 0.89 |
Bars against one 16.7 ms frame. A browser-tab run is timed in a visible tab without the tool’s scripted input, so read it as close, not exact.
TRAFFIC_STEP) and keeps the rest of dt for the next call, so the same seed gives the same traffic at any frame rate; pose(car) looks ahead by that rest, so cars glide.From HaTac. Fully custom code.
Treiber, Hennecke and Helbing, the Intelligent Driver Model (2000): Car following: the desired gap s0 + vT + v·Δv / 2√(ab) and the acceleration a[1 − (v/v0)⁴ − (s*/s)²]. The published model, not code.
Dijkstra’s shortest paths (1959): Routing over lane paths, with arriving at the goal as a node of its own.
George Marsaglia, xorshift generators (2003): The seeded generator behind spawning and choices.
Fowler–Noll–Vo hash (FNV-1a): Each junction’s signal offset from its id.