VideoDB: A GPU-Accelerated Data Storage Library
Hi everyone,
I’d like to share a personal project called VideoDB. It’s a small TypeScript/JavaScript library that uses WebGPU to store data directly in GPU memory (VRAM) instead of the usual system memory. Think of it like a lightweight IndexedDB for the browser, but GPU-accelerated.
Key Points
- High Throughput: With 1KB rows, add operations can reach ~500K rows/sec (~500 MB/s) depending on the data type. Batched reads can exceed ~200K gets/sec.
- Supports JSON & Typed Arrays: You can store arbitrary JSON or binary data, including typed arrays.
- Deferred Writes: Writes are batched to minimize overhead.
I’ve also put together a demo page that shows how to:
- Create stores
- Add data
- Fetch data
- Run stress tests
The page includes full class documentation, usage examples, and features like importing JSON datasets from data.gov, loading them into your GPU, and viewing them in a grid with paging.
Thanks for taking a look! This is a free tool with no advertising or "premium" membership levels.
Author Of article : dgriebel2014 Read full article
