WebGPU Compute Shaders vs. WebGL Fragment Shaders: Architecture & Performance
Evaluating GPGPU compute limitations in WebGL against raw WGSL parallel execution models for heavy web-based computations.
Before WebGPU, general-purpose GPU computing (GPGPU) on the web required tricking WebGL by encoding numerical data into floating-point textures and evaluating algorithms inside fragment shaders. WebGPU compute shaders bypass these rendering overheads completely.
1. The Hacky Era of GPGPU in WebGL
In WebGL, performing parallel array operations forced developers to attach 2D texture framebuffers and execute full-screen quad render passes. This approach incurred fixed rasterization pipeline overhead and rigid texture output dimensions.
2. Explicit Memory Access with Storage Buffers
WebGPU compute shaders operate directly on arbitrary Storage Buffers without dummy render targets. Threads in WGSL compute workgroups can read and write to shared memory blocks concurrently with atomic memory operations.
3. Benchmarking Compute Workloads
For workloads like particle physics, matrix multiplications, and spatial sorting algorithms, WebGPU compute dispatches reduce driver overhead significantly compared to WebGL texture-based framebuffers, delivering up to 4x higher throughput.