WebGPU FFT

English

简体中文

English

简体中文

Appearance

WebGPU FFTBrowser-native FFT core

GPU-accelerated 1D/2D FFT and real-input RFFT APIs for JavaScript and TypeScript. High-throughput transforms run on WebGPU; spectrum analysis and image filtering remain CPU-based utilities.

Quick Start
Evaluate the API
View on GitHub
WebGPU FFT
⚡

GPU FFT core

WebGPU compute shaders accelerate the FFT engine itself. The repository includes a benchmark script that reports measured CPU and WebGPU results for the current environment.

📊

Honest capability boundaries

Complex FFT and real-input RFFT APIs are available on both CPU and GPU surfaces. Spectrum analysis and image filtering are built-in helpers, but they currently use CPU FFT internally.

🎵

TypeScript-first package

ESM + CJS exports, no runtime dependencies, and a clean typed API surface for browser or Node-based workflows.

🔍

Fast evaluation path

Start with Quick Start, verify supported APIs in the reference, then inspect the architecture pages if you need implementation details.

What this site is for ​

This site helps you answer four questions quickly:

  1. Is the FFT core actually GPU-accelerated? Yes — the FFTEngine and 2D FFT pipeline are the GPU-centered parts.
  2. What is CPU-only today? createSpectrumAnalyzer() and createImageFilter() are convenience utilities built on the CPU FFT path.
  3. Can I benchmark it honestly? Yes — npm run benchmark reports measured CPU results and measured WebGPU results only when WebGPU is actually available.
  4. Where should I start? Use Quick Start for usage, API Reference for contracts, and Architecture for internals.

Good fit / Not a fit ​

Good fit ​

  • Browser-native FFT workloads where WebGPU is available
  • Large 1D or 2D transforms that justify GPU setup overhead
  • Real-valued signal or image workloads that benefit from compressed half-spectrum APIs
  • Projects that want a typed FFT core without runtime dependency sprawl

Not a fit ​

  • Workloads that require GPU-native spectrum analysis or GPU-native image filtering today
  • Arbitrary non-power-of-two FFT sizes without preprocessing
  • A broad DSP framework with many unrelated signal-processing features

Evaluate quickly ​

Released under the MIT License.

Copyright © 2024-2026 WebGPU FFT Contributors