Hailo-10H

Hailo-10H is an edge inference accelerator intended to run computer-vision and generative neural networks inside a host device. Its dataflow architecture performs model operations locally, reducing dependence on a remote service for each inference. The part can therefore serve embedded cameras, robots and other equipment, while the host processor remains responsible for application logic, sensor handling and physical control.

Hailo rates the device at 40 TOPS for INT4 and 20 TOPS for INT8 workloads with typical consumption of 2.5 watts. Offerings include chip-on-board and M.2 2242 or 2280 modules, with LPDDR4 or LPDDR4X support and industrial or automotive temperature grades. Those peak arithmetic figures are precision-specific; realised latency and throughput depend on model compilation, memory traffic, host interfaces, thermal design and the complete software pipeline.