Enhanced Raspberry Pi Board with Expanded RAM for AI Tasks

The Raspberry Pi lineup is expanding yet again with a novel add-on board designed to enhance the AI capabilities of the Raspberry Pi 5. This new accessory, priced at $130 and known as the AI HAT+ 2, represents a significant upgrade from its predecessor. It boasts 8GB of RAM and integrates the Hailo 10H chip, capable of delivering 40 TOPS in AI performance metrics.

Optimizing AI Workloads

Once integrated, the AI HAT+ 2 enables the Raspberry Pi 5 to efficiently process AI-centric tasks. This configuration allows the mainboard's Arm CPU to remain available for other computing activities. Unlike the former AI HAT+, which specialized in AI image processing, this updated module comes equipped with onboard memory enabling it to run compact generative AI models such as Llama 3.2 and DeepSeek-R1-Distill. Moreover, it also facilitates training and refining AI models.

Practical AI Implementations

Demonstrations illustrate how this board can support an AI model that generates descriptive text from a video feed or determine the presence of individuals in a scene. Another example shows streamlined translation capabilities, like converting text from French to English using the Qwen2 model.

Nonetheless, the AI HAT+ 2’s performance, in certain scenarios, doesn't surpass the Raspberry Pi 5's standalone processing. According to tech influencer Geerling, the Pi 5 with 8GB RAM can exceed the capabilities of this add-on, primarily due to the higher power engagement permitted by the Pi 5 at up to 10 watts, while the accessory is confined to 3 watts.

Comparative Analysis

Despite its niche advantages, Geerling highlights that the supplementary 8GB memory doesn’t necessarily give the HAT the edge over opting for a more comprehensive 16GB Pi variant, which offers more adaptability and quicker model processing. The original AI HAT, available since last year, facilitates AI-driven image processing at a lower $70 beginning price.

Future Prospects

Raspberry Pi has announced plans for more robust AI models, indicating that they are currently being prepped for deployment. These models should be ready for installation shortly after the launch, further extending the capabilities of the Raspberry Pi systems.