All GPUs. All NPUs. All CPUs. Working together.
Detecting compute devices...
50,000 particles split across detected devices. Each device computes its slice independently.
Run model on NPU + GPU simultaneously. WebNN auto-partitions: NPU takes small fast ops, GPU takes large parallel ops.
Live chart of how each device is being used. Updates as workloads run.
WebNN is the W3C standard for multi-chip ML (CPU + GPU + NPU). One API, all devices.