Raspberry Pi has announced the release of the AI HAT+ 2, an add-on board designed to enable the Raspberry Pi 5 to run generative AI models locally. Priced at $130, the upgraded AI HAT+ 2 offers 8GB of RAM and a Hailo 10H chip with 40 TOPS of AI performance, expanding the capabilities of the previous model. This new board allows the Raspberry Pi 5 to offload AI workloads to the AI HAT+ 2, freeing up the main board’s Arm CPU for other tasks.
Unlike its predecessor focused on image-based AI processing, the AI HAT+ 2 comes equipped with onboard RAM and can handle small generative AI models like Llama 3.2, DeepSeek-R1-Distill, and Qwen models. Users can also train and fine-tune AI models using this device.
A demonstration showcased by Raspberry Pi illustrates how the AI HAT+ 2 can power AI models for text-based description generation from a camera stream and language translation tasks, such as translating French to English with Qwen2.
However, tech enthusiast Jeff Geerling discovered that a standalone Raspberry Pi 5 with 8GB of RAM generally outperformed the AI HAT+ 2 across supported models due to power draw limitations. The Pi 5 can operate at up to 10 watts, whereas the AI HAT+ 2 is capped at 3W.
While the AI HAT+ 2 may excel in specific scenarios, Geerling suggests that the additional 8GB of RAM may not provide a significant advantage over opting for the more versatile 16GB Pi with enhanced performance capabilities. The previous AI HAT+ model, priced at $70, remains a cost-effective solution for AI image processing.
Source: The Verge