REVIEWS AI kit
Raspberry Pi 5 AI Kit
Trying to find some intelligence on Earth, Les Pounder fires up the official AI kit for the Raspberry Pi.
SPECS
Support: Pi 5 Form: Pi M.2 HAT+ NPU: Hailo-8L Entry-Level AI Accelerator, 2242 package TOPS: 13 Support: TensorFlow, TensorFlow Lite, Keras, PyTorch and ONNX Host: x86, Arm
B
ased around the Hailo-8L entry-level NPU (neural processing unit) AI accelerator and the Raspberry Pi M.2 HAT+, this kit provides all the hardware that a Raspberry Pi 5 would need to beef up its AI powers.
We got our hands on an early unit, upon which this review is based. But before we begin, let’s caveat the review. At the time of writing, the software was behind the rather excellent hardware, and while we were able to run the preview demos, we were unable to write any code of our own. As such, we’ll be revisiting this kit once the software is updated.
Installation and setup
Using the Raspberry Pi M.2 HAT+ board is a smart move. It already exists, and it has the connection and bandwidth for the Hailo-8L board. It also means that it comes with plastic spacers and a GPIO passthrough, which were a bone of contention for us in our previous review. Installation of the Hailo-8L is as easy as installing an M.2 NVMe SSD, but it does introduce one issue: unless you have a board with dual M.2 connectors, you can only run the AI or an NVMe SSD. Not both. Fear not, we shall be testing a dual M.2 board later, and (spoiler) it works if you put the effort in.