Raspberry Pi announces 'Raspberry Pi AI HAT+', an add-on module for AI processing, with a processing performance of 26 TOPS and the ability to run multiple AI simultaneously



The Raspberry Pi AI HAT+ is an add-on module that allows you to add an NPU for AI processing to the Raspberry Pi 5. The Raspberry Pi AI HAT+ comes in two models: 13 TOPS and 26 TOPS, and is capable of running multiple AIs simultaneously.

Introducing the Raspberry Pi AI HAT+ with up to 26 TOPS - Raspberry Pi

https://www.raspberrypi.com/news/raspberry-pi-ai-hat/

Raspberry Pi AI HAT+ – Raspberry Pi
https://www.raspberrypi.com/products/ai-hat/

The Raspberry Pi AI HAT+ looks like this. It is equipped with the Hailo -8L or Hailo-8 NPU, with the Hailo-8L model providing 13 TOPS of AI processing power, and the Hailo-8 model providing 26 TOPS of AI processing power.



Raspberry Pi also sells the 'Raspberry Pi AI Kit' equipped with the Hailo-8L, but the Raspberry Pi AI Kit has problems such as 'the system becomes complicated' and 'heat dissipation is reduced' because it uses an M.2 connector. The newly released Raspberry Pi AI HAT + can be connected directly to the main board without the M.2 connector, simplifying the system and improving heat dissipation. In addition, explicit settings are required to use the PCIe 3.0 connection with the Raspberry Pi 5, but the Raspberry Pi AI HAT + can automatically connect to PCIe 3.0 without any settings.

Review of 'Raspberry Pi AI Kit' that can add an AI dedicated processor exceeding Core Ultra to Raspberry Pi 5 and turn it into an AI camera - GIGAZINE



The Raspberry Pi AI HAT+ is intended to be used in combination with a camera module and other components as follows:



The Hailo-8 equipped model of the Raspberry Pi AI HAT+ can run multiple AIs simultaneously. In the example below, three types of image recognition AIs are running simultaneously.

Raspberry Pi AI HAT+ - YouTube


The Raspberry Pi AI HAT+ is priced at $70 (about 11,000 yen) for the Hailo-8L model and $110 (about 17,000 yen) for the Hailo-8 model.

in Hardware, Posted by log1o_hf