2nd/10_Wiki/Topics/Frontend/Instancing.md

---
id: wiki-2026-0508-instancing
title: Instancing
category: 10_Wiki/Topics
status: verified
canonical_id: self
aliases: [GPU instancing, instanced rendering, drawInstanced]
duplicate_of: none
source_trust_level: A
confidence_score: 0.95
verification_status: applied
tags: [graphics, gpu, webgl, webgpu, threejs, performance]
raw_sources: []
last_reinforced: 2026-05-10
github_commit: pending
tech_stack:
  language: GLSL/WGSL
  framework: Three.js/WebGPU
---

# Instancing

## 매 한 줄
> **"매 동일 mesh 를 N 번 그릴 때 single draw call 로 묶는 GPU 기법 — 매 per-instance attribute (transform, color) 만 다르게"**. 매 thousands → millions of objects 가 60fps 로 가능. 매 grass, particles, crowd, foliage 의 핵심. WebGL 2 / WebGPU / Vulkan / Metal 모두 native 지원.

## 매 핵심

### 매 왜 빠른가
- **Draw call overhead 제거**: 매 CPU→GPU command 1번 만.
- **Vertex buffer reuse**: 매 base mesh 1개, instance attr 만 streaming.
- **GPU parallelism**: 매 instance 가 각 SM/CU 에 분산.

### 매 per-instance attribute 종류
- **Transform matrix** (mat4) — 매 가장 흔함.
- **Color / tint** (vec4)
- **Texture index** / atlas UV offset
- **Animation frame** (skinned crowd)

### 매 응용
1. Three.js `InstancedMesh` — 매 50k tree, 100k particle.
2. Unreal HISM / Niagara, Unity GPU Instancer.
3. WebGPU compute-driven instancing — 매 frustum culling on GPU.
4. Game KvK map — 매 thousands of city/troop sprite.

## 💻 패턴

### Three.js InstancedMesh
```javascript
import * as THREE from "three";

const geometry = new THREE.BoxGeometry(1, 1, 1);
const material = new THREE.MeshStandardMaterial({ color: 0x44aa88 });
const count = 10000;
const mesh = new THREE.InstancedMesh(geometry, material, count);

const m = new THREE.Matrix4();
for (let i = 0; i < count; i++) {
  m.setPosition(
    (Math.random() - 0.5) * 100,
    0,
    (Math.random() - 0.5) * 100,
  );
  mesh.setMatrixAt(i, m);
}
mesh.instanceMatrix.needsUpdate = true;
scene.add(mesh);
```

### Per-instance color
```javascript
const colors = new Float32Array(count * 3);
for (let i = 0; i < count; i++) {
  colors[i * 3]     = Math.random();
  colors[i * 3 + 1] = Math.random();
  colors[i * 3 + 2] = Math.random();
}
geometry.setAttribute(
  "instanceColor",
  new THREE.InstancedBufferAttribute(colors, 3),
);

material.onBeforeCompile = (shader) => {
  shader.vertexShader = shader.vertexShader
    .replace("#include <common>", "#include <common>\nattribute vec3 instanceColor; varying vec3 vColor;")
    .replace("#include <begin_vertex>", "#include <begin_vertex>\nvColor = instanceColor;");
  shader.fragmentShader = shader.fragmentShader
    .replace("#include <common>", "#include <common>\nvarying vec3 vColor;")
    .replace("vec4 diffuseColor = vec4( diffuse, opacity );", "vec4 diffuseColor = vec4( diffuse * vColor, opacity );");
};
```

### Update single instance (game troop move)
```javascript
function updateTroop(idx: number, x: number, z: number) {
  m.setPosition(x, 0, z);
  mesh.setMatrixAt(idx, m);
  mesh.instanceMatrix.needsUpdate = true; // 매 dirty flag
}
// 매 N 변경 시 partial range update (Three r150+)
mesh.instanceMatrix.addUpdateRange(start * 16, count * 16);
```

### WebGPU instanced draw
```javascript
// WGSL vertex shader
const wgsl = `
struct Instance { @location(3) m0: vec4f, @location(4) m1: vec4f, @location(5) m2: vec4f, @location(6) m3: vec4f };
@vertex fn vs(@location(0) pos: vec3f, instance: Instance) -> @builtin(position) vec4f {
  let model = mat4x4f(instance.m0, instance.m1, instance.m2, instance.m3);
  return uniforms.viewProj * model * vec4f(pos, 1.0);
}`;

// JS — drawIndexedIndirect 또는 draw with instanceCount
pass.draw(vertexCount, instanceCount, 0, 0);
```

### GPU frustum culling (compute → instanced draw)
```wgsl
@compute @workgroup_size(64)
fn cull(@builtin(global_invocation_id) gid: vec3u) {
  let idx = gid.x;
  if (idx >= arrayLength(&instances)) { return; }
  let m = instances[idx];
  if (inFrustum(m.bbox)) {
    let out = atomicAdd(&visibleCount, 1u);
    visibleInstances[out] = m;
  }
}
// 매 visibleInstances + visibleCount → drawIndirect
```

### Foliage with wind (vertex shader animation)
```glsl
attribute mat4 instanceMatrix;
uniform float uTime;
void main() {
  vec3 p = position;
  float wind = sin(uTime + instanceMatrix[3].x * 0.1) * 0.1 * p.y;
  p.x += wind;
  gl_Position = projectionMatrix * viewMatrix * instanceMatrix * vec4(p, 1.0);
}
```

## 매 결정 기준
| 상황 | Approach |
|---|---|
| Same mesh, ≥100 copies | InstancedMesh |
| Different meshes, similar | BatchedMesh (Three r155+) |
| Dynamic count + culling | GPU compute culling + drawIndirect |
| Skinned crowd | Texture-baked anim + instancing |
| Just 10 copies | 매 그냥 N draw — 매 instancing overhead 작음 |

**기본값**: ≥100 copies of same mesh → InstancedMesh.

## 🔗 Graph
- 부모: [[Draw Call Optimization]]
- 변형: [[BatchedMesh]] · [[Indirect Drawing]]
- 응용: [[Three.js]] · [[WebGPU]]
- Adjacent: [[Frustum Culling]] · [[LOD]]

## 🤖 LLM 활용
**언제**: large-scale rendering 설계, particle / foliage / crowd 구현.
**언제 X**: 매 single object, 매 매우 다른 mesh — 매 instancing overhead 정당화 X.

## ❌ 안티패턴
- **needsUpdate = true on every frame**: 매 모든 instance 변경 안 했어도 전체 upload — 매 partial update range 사용.
- **instancing for unique meshes**: 매 효과 X — 매 BatchedMesh 또는 multi-draw indirect.
- **CPU-side culling per instance**: 매 GPU compute 가 더 빠름 (10k+).

## 🧪 검증 / 중복
- Verified (Three.js docs, WebGPU spec, Real-Time Rendering 4th ed.).
- 신뢰도 A.

## 🕓 Changelog
| 날짜 | 변경 |
|---|---|
| 2026-05-08 | Phase 1 |
| 2026-05-10 | Manual cleanup — Three.js + WebGPU instancing 패턴 |