Select a workload

Run benchmark variants in your browser and compare custom WebGPU kernels against ONNX Runtime, TensorFlow.js, and jax-js.

01

Matmul

Square matrix multiplication across WebGPU kernels and ML frameworks.

02

Matvec

Matrix-vector multiplication repeated to amortize dispatch overhead.

03

Conv2d

2D convolution in NCHW layout with framework comparisons.