Now loading...
Raspberry Pi has unveiled a fresh accessory designed to bring generative AI capabilities directly to the Raspberry Pi 5 without relying on cloud services. The new $130 AI HAT+ 2, revealed this week, builds on the previous model by incorporating 8GB of dedicated memory and a powerful Hailo 10H processor that delivers 40 tera operations per second for AI tasks.
This add-on snaps onto the Raspberry Pi 5, allowing it to offload intensive machine learning operations to the HAT while the device’s core Arm processor handles everyday computing duties. Departing from its predecessor’s emphasis on visual AI processing, the updated version supports compact generative models such as Llama 3.2, DeepSeek-R1-Distill, and various Qwen variants. Users can even customize these models through training and fine-tuning right on the setup.
In demonstrations from Raspberry Pi, the board powers an AI system that analyzes live video feeds to produce textual summaries and responds to queries about elements like human presence in the footage. Another example highlights real-time language translation, converting French text to English using the Qwen2 model.
Independent testing by technology enthusiast Jeff Geerling revealed that a standard Raspberry Pi 5 equipped with 8GB of RAM often surpasses the AI HAT+ 2 in model performance. He attributes this to differences in power allocation, with the Pi 5 capable of drawing up to 10 watts compared to the HAT’s strict 3-watt limit. While the extra onboard memory offers some benefits for targeted applications, Geerling points out that it falls short of justifying the cost over simply opting for a higher-spec Pi 5 with 16GB of RAM, which provides greater versatility and speed.
For those seeking more affordable AI image handling, the original AI HAT remains available starting at $70. Raspberry Pi anticipates releasing additional, more substantial AI models compatible with the HAT+ 2 in the coming weeks, expanding its potential for hobbyists and developers alike. For details, check the official announcement or the product page.
