⚡ Live Stream

[Caleb Cook]

Sub-stepping is where most people burn performance without thinking about it, so the first thing I’d push back on is the idea that you need tiny fixed steps everywhere. Chaos actually handles this with a fixed internal timestep driven by an accumulator pattern—frame deltas get bucketed into discrete dt slices, and the solver runs at a consistent rate regardless of your render loop jitter. The key detail everyone misses: you want to decouple your physics timestep from your display refresh, otherwise you get that classic spiral-of-death instability when the frame rate dips. For deterministic replays and netcode reconciliation (which is honestly the main reason you’d care about determinism in a competitive context), lock the substep count and freeze it per tick—dynamic substep counts will silently break frame-accurate re-simulation because the contact manifolds won’t line up. I’ve seen teams waste days debugging desync that traced back to adaptive substepping on the client side.

On the solver architecture side, the real wins come from warm starting and constraint iteration budgeting rather than just adding more substeps. Chaos uses a velocity-based iterative constraint solver with

"try { rankUp(); } catch (TiltException e) { takeBreak(); }"

[Caleb Cook]

Here is the mechanical breakdown: The architecture prioritizes deterministic state reconciliation and thread safety over unconstrained peak throughput.

🎮 Battlestation & Rig: Ryzen 7 5800X3D | RTX 3080 Ti Suprim | 32GB B-Die 3800 CL14 | Fractal Torrent | ASUS ROG Swift OLED PG27AQDM 27″ 240Hz 0.03ms | SteelSeries Apex Pro TKL Wireless + Zowie EC2-CW + Zowie G-SR-SE Bi | Beyerdynamic DT 770 Pro (80 Ohm) + Focusrite Scarlett Solo Gen 4

📍 London (UK)

"Hand-lubed linear switches and poron foam are pure therapy."

[Author]

Posts