Part of the SkyAI™ family of intelligent computing subsystems, the NANOhpc-AI is a high-performance and power-efficient artificial intelligence subsystem composed of a base NANOhpc-obc unit connected via a high-speed PCIe interface to an AI extension built around dual Hailo-8 accelerators.

Each Hailo-8 delivers up to 26 TOPS of inference performance, providing a combined peak throughput of 52 TOPS. A PCIe switch manages the high-bandwidth interconnect between the NANOhpc-obc and the AI accelerators, ensuring sufficient data throughput. 

The NANOhpc-obc provides supervision, housekeeping, and general control capabilities, while the AI extension is dedicated to real-time inference for demanding onboard applications such as video analytics and sensor fusion.

Base NANOhpc-obc

• Main CPU cluster4x RISC-V 64-bit processors @ max 600 MHz

• Supervisor RISC-V monitoring core@ max 600 MHz

• Main memory 4 GB DDR4 

• Memory Management MMU with Physical memory protection

• Mass storage 92 GB NAND (EDAC protected)

• TM storage 256 kB MRAM with SECDED

• Interfaces:

  • Redundant CAN,

  • 4x LVDS/RS422/485/UART/16xGPIO,

  • 2x Gigabit ETH

• Software Linux (Yocto)

AI extension board

• AI Technology Dual Hailo 8 accelerator

• Performance Up to 52 TOPS

• Typical power Combined 5W consumption

• Supported Frameworks:

  • TensorFlow,

  • TensorFlow Lite,

  • Keras,

  • PyTorch & ONNX