Raspberry Pi SC2166 Raspberry Pi AI HAT+ 2

The Raspberry Pi SC2166 AI HAT+ 2 brings generative AI capability to Raspberry Pi 5. It is the latest entry in our range of AI/ML accelerator products. Building on the success of the original AI HAT+.

With 8GB of dedicated on board RAM, the Raspberry Pi AI HAT+ 2 enables running of large language models (LLMs) and vision-language models (VLMs) locally and securely. This leaves the host Raspberry Pi 5 free for handling other tasks and ensuring your AI models run smoothly.

Designed for both privacy and security, the AI HAT+ 2 delivers reliable, low-latency AI processing at the edge and without a network connection, simplifies data security, minimises infrastructure requirements, and reduces cloud API costs.

AI HAT+ 2 is built around the Hailo-10H neural network inference accelerator, and delivers 40 TOPS of INT4 inference performance. This supports all AI HAT+ capabilities (including object detection, camera post-processing, robotics, moderate neural workloads) and adds local LLM and VLM capabilities, enabling applications such as local chat over documents.

A set of sample models is provided by Hailo, allowing users to also train custom vision models or fine-tune generative AI models using Low-Rank Adaptation (LoRA) to suit their application, such as speech to text, translation, or visual scene analysis.
The AI HAT+ 2 communicates using Raspberry Pi 5's PCI Express interface. When the host Raspberry Pi 5 is running an up-to-date Raspberry Pi OS image, it automatically detects the on board Hailo accelerator and makes it available for AI computing tasks. The built-in rpicam apps camera applications in Raspberry Pi OS natively support the AI module, automatically using the accelerator to run compatible post-processing.

• Compatible with Raspberry Pi 5
• Hailo 10H neural network accelerator, delivering 40 TOPS (INT4) inferencing performance
• Onboard 8GB LPDDR4X-4267 SDRAM
• Similar performance for computer vision models to the Raspberry Pi AI HAT+ (26 TOPS)
• Full integration into Raspberry Pi's camera software stack
• Conforms to Raspberry Pi HAT+ specification
• Accelerates selected LLMs and VLMs
• Expected LLM/VLM support at launch o Llama-3.2-3B-Instruct o QWEN2.5-VL-3B o ASR (whisper-base)