Lecture 04：游戏引擎中的渲染实践

• Objects with one type of effect
• Focus on representation and math correctness
• No strict performance requirements
• Realtime(30 FPS) / interactive(10 FPS) / offline
• Out-of-core rendering

GPU 架构

GPU v.s. CPU -> SIMT v.s. SIMD

GPU Architecture:

• GPC(Graphics Processing Cluster)
• SM(Streaming Multiprocessor)
• Texture Units
• CUDA Core
• Warp

Data Flow from CPU to GPU

CPU and Main Memory

• Data Load / Unload
• Data Preparation

CPU to GPU

• High Latency
• Limited Bandwidth

GPU and Video Memory

• High Performance Parallel Rendering

Tips

Always minimize data transfer between CPU and GPU when possible.

尽可能数据单向传输，同时尽可能不要从显卡中读数据

GPU Bounds and Performance

• Memory Bounds
• ALU Bounds
• TMU(Texture Memory Unit) Bounds
• BW(Bandwidth) Bounds

可渲染物体

• Material
• Texture
• Shader
• Mesh/Submesh
• Resource Pool

可见性裁剪（Visibility Culling）

只渲染可见的物体，在摄像机锥形视野内的物体

贴图压缩

Block Compression

Cluster-based Mesh Pipeline

一个未来渲染的发展趋势

---

最后更新: 2022年4月27日 15:29:45
创建日期: 2022年4月22日 15:02:30

评论