The Hailo-10H M.2 AI Acceleration Module, part number HM22HB1C2FAE, adds dedicated AI inference capability to a compatible PC or embedded edge computer. This configuration combines the Hailo-10H accelerator with 8GB of on-module LPDDR4/4X memory in an M.2 2280 Key M form factor.
The module is designed for supported generative AI and computer vision models. Development teams can assess applications such as image understanding, object detection, video analytics, and compatible language or vision-language inference. Processing on a local edge system can help reduce the need to transfer application data to a remote service. The result depends on the model, host system, and overall application design.
Hailo specifies up to 40 TOPS at INT4 precision or 20 TOPS at INT8 precision for the Hailo-10H. These ratings describe different numerical precision modes and should be considered alongside memory requirements and measured application performance. The 8GB memory configuration gives engineers more on-module memory for supported models than the 4GB configuration. Model compatibility still depends on the chosen model and software release.
HM22HB1C2FAE connects through PCIe Gen 3 x4. Its 2280 format requires an M.2 Key M socket with a 22 × 80 mm mounting position, suitable PCIe connectivity, and enough space for the module and its thermal solution. Check electrical compatibility, software support, and cooling before selecting a host platform.
The module is specified for an extended operating temperature range of −40°C to 85°C. Hailo lists 2.5 W typical power consumption, while actual power and heat output vary with the workload. Thermal performance should be assessed in the finished enclosure, particularly when running sustained AI inference.
Hailo’s software suite supports model preparation, compilation, and deployment. It includes the Hailo Dataflow Compiler and HailoRT runtime, with support for models developed using TensorFlow, TensorFlow Lite, Keras, PyTorch, and ONNX. Hailo lists x86 and Arm hosts and Linux, Windows, and Android operating systems for the module family.
