Sub-60ms stereo depth on Orange Pi Zero 3w via custom NPU kernels

_William_F_ · x · 2026-08-28

Achieved stereo depth inference with sub-60ms latency on an Orange Pi Zero 3w (14g, 30mm x 65mm PCB). This was accomplished by hand-writing custom NPU kernels for the hardware, eliminating the need for expensive GPUs like the 4090 and demonstrating extreme on-device optimization.

Related event: 14g Board Runs Stereo Depth Inference Under 60ms With Hand-Written NPU Kernels(3 posts)→

Original post →

More from Embodied

Embodied channel →