Local AI Advisor

Browser speed test

How fast a local model writes is limited mostly by memory bandwidth: how quickly your GPU can read the model's weights. This test measures it directly with WebGPU, right here in your browser.

1. Your computer (optional)

Tell us what it is to compare the result with its published memory bandwidth.

2. Measure memory bandwidth

Takes about 5 seconds and uses up to 512 MB of GPU memory. Nothing is downloaded or uploaded.

The test streams a large random buffer through a WebGPU compute shader and times it. Browsers add overhead, so results are usually below what native runtimes reach on the same machine. Prompt processing speed (how fast long inputs are read) depends on compute instead and isn't measured here.