Use a fast physics engine, pin the tick rate to a fixed 60 Hz, and pre‑calculate collision normals ahead of time. Switch to spatial hashing for neighbor queries instead of a full sweep, and keep your matrices in a tight cache line. If you can afford a bit of parallelism, split the track into independent segments and run each on its own core. Don’t forget to profile first—hot spots are usually memory bandwidth, not CPU. Happy hacking!

Got it, I’ll fire up the profiler and strip the physics loop down to just the essentials. No fluff, just raw speed. Thanks, will keep the iterations coming.