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

Datasheet for NANOhpc-AI - High performance AI accelerator by SkyLabs

Datasheet