r3f-performance
R3F performance optimization—LOD (Level of Detail), frustum culling, instancing strategies, draw call reduction, frame budgets, lazy loading, and profiling tools. Use when optimizing render performance, handling large scenes, or debugging frame rate issues.
Works with
---
name: r3f-performance
description: R3F performance optimization—LOD (Level of Detail), frustum culling, instancing strategies, draw call reduction, frame budgets, lazy loading, and profiling tools. Use when optimizing render performance, handling large scenes, or debugging frame rate issues.
license: MIT
---
# R3F Performance
Optimize render performance through draw call reduction, geometry optimization, smart loading, and profiling.
## Quick Start
```tsx
// Performance-optimized Canvas
<Canvas
dpr={[1, 2]} // Limit pixel ratio
performance={{ min: 0.5 }} // Adaptive performance
frameloop="demand" // Only render on change
gl={{
powerPreference: 'high-performance',
antialias: false // Disable for mobile
}}
>
<Suspense fallback={null}>
<Scene />
</Suspense>
</Canvas>
```
## Frame Budget
Target 60fps = 16.67ms per frame. Budget breakdown:
| Phase | Target | Notes |
|-------|--------|-------|
| JavaScript | < 4ms | useFrame logic, state updates |
| GPU Render | < 10ms | Draw calls, shaders |
| Compositing | < 2ms | Post-processing, overlays |
| Buffer | ~1ms | Safety margin |
## Draw Call Optimization
### The Golden Rule
**Fewer draw calls > fewer triangles**
A scene with 100 meshes of 1000 triangles each is slower than 1 mesh of 100,000 triangles.
### Reduction Techniques
| Technique | Draw Calls | When to Use |
|-----------|------------|-------------|
| Instancing | 1 per unique mesh | 100+ identical objects |
| Merged geometry | 1 per merged batch | Static scene parts |
| Texture atlases | Fewer materials | Many similar textures |
| LOD | Reduces complexity | Large/distant objects |
### Instancing (Best for Identical Meshes)
```tsx
// 10,000 cubes = 1 draw call
<instancedMesh args={[undefined, undefined, 10000]}>
<boxGeometry />
<meshStandardMaterial />
</instancedMesh>
```
### Geometry Merging (Static Scenes)
```tsx
import { useMemo } from 'react';
import { mergeGeometries } from 'three/examples/jsm/utils/BufferGeometryUtils';
import * as THREE from 'three';
function MergedScene() {
const mergedGeometry = useMemo(() => {
const geometries: THREE.BufferGeometry[] = [];
// Create many positioned geometries
for (let i = 0; i < 100; i++) {
const geo = new THREE.BoxGeometry(1, 1, 1);
geo.translate(
(Math.random() - 0.5) * 20,
(Math.random() - 0.5) * 20,
(Math.random() - 0.5) * 20
);
geometries.push(geo);
}
return mergeGeometries(geometries);
}, []);
return (
<mesh geometry={mergedGeometry}>
<meshStandardMaterial />
</mesh>
);
}
```
## Level of Detail (LOD)
Swap geometry based on camera distance:
```tsx
import { useMemo } from 'react';
import * as THREE from 'three';
function LODMesh() {
const lod = useMemo(() => {
const lodObject = new THREE.LOD();
// High detail (close)
const highGeo = new THREE.SphereGeometry(1, 64, 64);
const highMesh = new THREE.Mesh(highGeo, new THREE.MeshStandardMaterial({ color: 'red' }));
lodObject.addLevel(highMesh, 0);
// Medium detail
const medGeo = new THREE.SphereGeometry(1, 32, 32);
const medMesh = new THREE.Mesh(medGeo, new THREE.MeshStandardMaterial({ color: 'orange' }));
lodObject.addLevel(medMesh, 10);
// Low detail (far)
const lowGeo = new THREE.SphereGeometry(1, 8, 8);
const lowMesh = new THREE.Mesh(lowGeo, new THREE.MeshStandardMaterial({ color: 'green' }));
lodObject.addLevel(lowMesh, 30);
return lodObject;
}, []);
return <primitive object={lod} />;
}
```
### Drei LOD Helper
```tsx
import { Detailed } from '@react-three/drei';
function AdaptiveSphere() {
return (
<Detailed distances={[0, 10, 30]}>
{/* Close: high detail */}
<mesh>
<sphereGeometry args={[1, 64, 64]} />
<meshStandardMaterial />
</mesh>
{/* Medium distance */}
<mesh>
<sphereGeometry args={[1, 32, 32]} />
<meshStandardMaterial />
</mesh>
{/* Far: low detail */}
<mesh>
<sphereGeometry args={[1, 8, 8]} />
<meshStandardMaterial />
</mesh>
</Detailed>
);
}
```
## Frustum Culling
Objects outside camera view are not rendered. Enabled by default, but:
```tsx
// Disable for objects that animate into view unpredictably
<mesh frustumCulled={false}>
<boxGeometry />
<meshStandardMaterial />
</mesh>
// Force bounding sphere update for dynamic geometry
useEffect(() => {
geometry.computeBoundingSphere();
}, [geometry]);
```
## Adaptive Performance
R3F's adaptive performance system automatically adjusts DPR:
```tsx
<Canvas
performance={{
min: 0.5, // Minimum DPR under stress
max: 1, // Maximum DPR
debounce: 200 // Debounce time for changes (ms)
}}
/>
```
### Manual Performance Control
```tsx
import { useThree } from '@react-three/fiber';
function PerformanceMonitor() {
const { performance } = useThree();
useFrame(() => {
// Check current performance
if (performance.current < 1) {
// System is under stress, reduce complexity
}
});
// Trigger performance drop
const triggerRegress = () => {
performance.regress(); // Temporarily lower DPR
};
}
```
## Lazy Loading
### Code Splitting with Suspense
```tsx
import { Suspense, lazy } from 'react';
const HeavyModel = lazy(() => import('./HeavyModel'));
function Scene() {
return (
<Suspense fallback={<SimpleLoader />}>
<HeavyModel />
</Suspense>
);
}
```
### Progressive Loading
```tsx
import { useGLTF } from '@react-three/drei';
function Model() {
// Preload in background
useGLTF.preload('/model.glb');
const { scene } = useGLTF('/model.glb');
return <primitive object={scene} />;
}
// Preload before component mounts
useEffect(() => {
useGLTF.preload('/next-model.glb');
}, []);
```
### View-Based Loading
```tsx
import { useInView } from 'react-intersection-observer';
function LazySection() {
const { ref, inView } = useInView({
triggerOnce: true,
rootMargin: '200px' // Start loading 200px before visible
});
return (
<group ref={ref}>
{inView && <HeavyContent />}
</group>
);
}
```
## Memory Management
### Dispose Unused Resources
```tsx
// Manual disposal
useEffect(() => {
return () => {
geometry.dispose();
material.dispose();
texture.dispose();
};
}, []);
// Drei helper for GLTF
import { useGLTF } from '@react-three/drei';
useEffect(() => {
return () => {
useGLTF.clear('/model.glb');
};
}, []);
```
### Texture Optimization
```tsx
import { useTexture } from '@react-three/drei';
import * as THREE from 'three';
// Compress and optimize
const texture = useTexture('/texture.jpg', (tex) => {
tex.minFilter = THREE.LinearMipmapLinearFilter;
tex.generateMipmaps = true;
tex.anisotropy = 4; // Lower = faster, higher = sharper
});
// Use compressed formats (KTX2)
import { useKTX2 } from '@react-three/drei';
const texture = useKTX2('/texture.ktx2');
```
## Profiling
### Stats Panel
```tsx
import { Stats } from '@react-three/drei';
<Canvas>
<Stats /> {/* FPS, MS, MB counters */}
<Scene />
</Canvas>
```
### Performance Panel
```tsx
import { Perf } from 'r3f-perf';
<Canvas>
<Perf
position="top-left"
showGraph // Show FPS graph
minimal={false} // Full or minimal view
/>
<Scene />
</Canvas>
```
### Manual Profiling
```tsx
import { useThree } from '@react-three/fiber';
function ProfileInfo() {
const { gl } = useThree();
useEffect(() => {
const info = gl.info;
console.log({
drawCalls: info.render.calls,
triangles: info.render.triangles,
points: info.render.points,
lines: info.render.lines,
textures: info.memory.textures,
geometries: info.memory.geometries
});
});
return null;
}
```
### Frame Time Measurement
```tsx
function FrameProfiler() {
const frameTimeRef = useRef<number[]>([]);
useFrame(() => {
const start = performance.now();
// ... your logic ...
const elapsed = performance.now() - start;
frameTimeRef.current.push(elapsed);
if (frameTimeRef.current.length > 60) {
const avg = frameTimeRef.current.reduce((a, b) => a + b) / 60;
console.log(`Avg frame time: ${avg.toFixed(2)}ms`);
frameTimeRef.current = [];
}
});
return null;
}
```
## Common Bottlenecks
| Symptom | Likely Cause | Fix |
|---------|--------------|-----|
| Low FPS, high draw calls | Too many meshes | Instance, merge, or LOD |
| Low FPS, few draw calls | Heavy shaders/materials | Simplify shaders, use cheaper materials |
| Stuttering on load | Large assets | Lazy load, compress, use LOD |
| Memory growth | No disposal | Dispose on unmount |
| Mobile issues | High DPR, AA | Limit DPR, disable antialias |
## Optimization Checklist
```markdown
[ ] Draw calls < 100 for complex scenes
[ ] Instancing for repeated objects
[ ] LOD for large/distant objects
[ ] Geometry merged where possible
[ ] Textures compressed (KTX2/Basis)
[ ] DPR capped at 2
[ ] Lazy loading for heavy assets
[ ] Proper disposal on unmount
[ ] Frustum culling enabled
[ ] Shadows optimized or disabled
```
## File Structure
```
r3f-performance/
├── SKILL.md
├── references/
│ ├── profiling-guide.md # Deep profiling techniques
│ ├── mobile-optimization.md # Mobile-specific tips
│ └── large-scenes.md # Handling massive scenes
└── scripts/
├── utils/
│ ├── lod-helper.ts # LOD setup utilities
│ ├── merge-helper.ts # Geometry merging
│ └── perf-monitor.ts # Performance monitoring
└── presets/
├── mobile.ts # Mobile-optimized Canvas config
└── desktop.ts # Desktop-optimized Canvas config
```
## Reference
- `references/profiling-guide.md` — Deep profiling with browser DevTools
- `references/mobile-optimization.md` — Mobile-specific optimization
- `references/large-scenes.md` — Handling 100k+ object scenesMore Debugging skills
diagnosing-bugs
mattpocock/skills
Diagnosis loop for hard bugs and performance regressions. Use when the user says "diagnose"/"debug this", or reports something broken/throwing/failing/slow.
explore-code
lllllllama/rigorpilot-skills
Rigor Improve implementation leaf skill for auditable candidate implementation in deep learning research repositories. Use when the researcher explicitly authorizes exploratory work on an isolated branch or worktree to transplant modules, adapt a backbone, add LoRA or adapter layers, replace a head, or stitch together meaningful low-risk migration ideas with rollback-aware records in `explore_outputs/`. Do not use for end-to-end exploration orchestration on top of `current_research`, trusted baseline reproduction, conservative debugging, environment setup, verified contribution claims, or default repository analysis.
safe-debug
lllllllama/rigorpilot-skills
Rigor Debug / Rigor Audit skill for deep learning research work. Use when the user pastes a traceback, terminal error, CUDA OOM, checkpoint load failure, shape mismatch, NaN loss symptom, or training failure and wants conservative diagnosis before any patching, with debug fixes clearly separated from research contributions. Do not use for broad refactoring, speculative adaptation, automatic exploratory patching, or general repository familiarization.

