High-Performance Instanced Volumetric Fire for Games

Extended Masatatsu Nakamura’s THREE.Fire (MIT) with game-focused optimizations supporting thousands of simultaneous fire instances at high framerates.

Pleasant visuals strike a sensible balance between realism and performance.

Performance: RTX 2060 0.2ms CPU / <2ms GPU @ 145fps

Live Demo on CodePen →

Adaptations from Original

Performance:

  • Instanced rendering (single draw call)

  • GPU frustum culling (visibility attribute, throttled updates)

  • Distance LOD (20 → 8 raymarching steps)

  • Animation freezing (stops turbulence at distance)

  • CPU-precomputed inverse matrices

Visual:

  • Per-instance variation (speed, frequency, scale, phase)

  • Orange color grading with ACES tone mapping

  • Edge filtering for crisp flames

  • Asymmetric view-angle distortion

Demo: High-Performance Instanced Volumetric Fire with Dynamic LOD
License: MIT

9 Likes