GPU.JS - GPU Accelerated JavaScript
summary
Library claims to run JavaScript on the GPU in the browser by compiling a subset of JS into WebGL fragment shaders and packing arrays into textures. Built in a 24‑hour hackathon, it ends up using compiles JS to WebGL shaders, uses float textures and square textures and an SPMD kernel model. Benchmarks show 18× on GTX 1080 and 87× on iPhone 7 for large matrix and ray‑tracing workloads. Limitations include reduced floating‑point accuracy, branch divergence, a JS subset, watchdog timeouts, and hardware quirks.
AI-written summary · about
chapters (22)
- 0:00Intro
- 0:33Fazli Sapuan
- 1:07Eugene Cheah Pico Creator
- 1:47The Story Behind GPU.JS
- 2:20And it was a Nintendo fun start
- 3:22Literally under 24 hours
- 4:10Lesson learnt?
- 7:05All in one convenient API package
- 8:07Technology Overview
- 10:22Compiling to GLSL
- 11:46Turning our inputs into textures
- 14:19Writing our output
- 15:43Running the Program
- 16:20The Magic in 3 Sentences
- 18:30Caveats (cont'd...)
- 18:51SPMD Computation Model
- 19:25Example: Matrix Multiplication
- 22:55Demo: Ray Tracer
- 23:28Real-Time Ray Tracing
- 24:56Summary
- 25:54Interested to find out more?
- 26:44Extra: Join Hack&Roll 2017 (students only)
related talks
38mWebGPU: Next-generation 3D Graphics on the Web
59mYou can't do that with JavaScript!
01h 08mKEYNOTE: Use.GPU - Declarative/Reactive 3D Graphics
24mArbitrary Computation on the GPU Using WebGL
59mYou can't do that with JavaScript!
26mBeyond HTML and CSS: Fusing Javascript and shaders
33mHigh-Performance WebGL Apps with LLJS and asm.js
38mMy journey with WebGL