Continuous Collision Detection (CCD) vs Discrete Collision at Hyper Velocity: Advanced Implementation and Real-World Architecture

[Leo Carter]

The core problem I’m hitting is that discrete SAT checks are letting fast projectiles tunnel through thin colliders at high velocities, and I’m deciding between switching to swept-volume CCD or just shrinking my timestep. For anyone who’s actually implemented continuous capsule or swept-sphere tests in a real engine: what’s your threshold for switching from discrete to CCD based on projectile velocity relative to collider thickness? I’ve read the critical ratio is roughly when an object moves more than its own radius per frame, but I want real-world numbers, not textbook theory.

On the architecture front, I’m weighing a hybrid approach with swept broadphase against full GJK-based continuous tests only on narrowphase candidates.

"Wooting 60HE (Tachyon mode enabled, 0.1mm rapid trigger) + Lamzu Maya 4K"

[Noah Young]

The tunneling problem at hyper-velocity is fundamentally a sampling issue, so your first decision is whether you even need full CCD or just smarter substepping. Discrete detection only samples an object’s position at frame boundaries, so a bullet moving 200 units per frame through a 5-unit wall simply never registers the overlap—it’s gone. For competitive titles where frame-time consistency matters, I’d avoid blanket CCD because swept-volume/TOI (time-of-impact) tests blow up your narrow-phase cost, especially with SAT or GJK/EPA on convex shapes. The pragmatic architecture is a hybrid: keep discrete detection for anything under

"Crosshair placement and sound cues win 80% of gunfights. Keep your head cool."

[Author]

Posts