Retrofittable and scalable
AI‑plug-in cards

Two versions with two variants each are available for the expansion of standard computer systems. A version with four as well as a version with eight Hailo-8™ AI processors provide sufficient inference performance for AI applications. Variant one with four Hailo-8™ processors achieves a performance of 104 TOPs, 5,300 FPS inference performance when using ResNet, and 900 FPS with YOLO v5m. Variant two with eight Hailo-8™ processors even doubles performance.

Machine learning frameworks such as ONNX, Tensorflow and pyTorcg are supported by the MiAi-H8. Furthermore, the expansion card offers GStreamer, Native C/C++ and ONNX Runtime support. Beside the Dataflow Compiler, which enables the model conversion and -compilation, further software packages are available. HailoRT as runtime environment and driver package. Model Zoo provides pre-trained models. TAPPAS contains the development framework, templates and multi-network pipelines.

Subsequent system expansion
is possible

Retrofitting of systems is facilitated by compatibility with Linux Ubuntu 22.04, as well as Microsoft Windows 10 Professional or IoT 2021 and Windows 11 Professional. Multiple cards can be run in a single system for scaling. An actively cooled variant is available for more demanding operating temperatures. Here, the quad Hailo™ processor version supports -40 °C to 85 °C and the eight Hailo™ version -40 °C to +75 °C.
Power consumption is typically between 20 and 30 watts, reaching a maximum of 74 watts. The MiAi-H8 underlines its industrial suitability with its design and extended temperature range. ▪ ds