How good is Raspberry Pi’s AI Kit

🌈 Abstract

The article discusses the release of Raspberry Pi's single board computer AI Kit, which is powered by the Hailo-8L accelerator and Raspberry Pi's own M.2 HAT+ board. The kit offers 13TOPS of AI performance and is aimed at the embedded edge computing market, providing a significant power advantage for computer vision applications.

🙋 Q&A

[01] Raspberry Pi AI Kit

1. What are the key features of the Raspberry Pi AI Kit?

• The AI kit is powered by the Hailo-8L accelerator and Raspberry Pi's own M.2 HAT+ board
• It offers 13TOPS of AI performance, which is a significant improvement over the 2TOPS offered by the RISC-V Linux laptop
• The Hailo-8L accelerator consumes about 1-2 watts of power for typical embedded workloads like processing 60 FPS video in real-time
• The kit supports a wide range of AI frameworks including TensorFlow, TensorFlow Lite, Keras, PyTorch, and ONNX

2. Why did Raspberry Pi choose to use a separate accelerator board rather than an integrated NPU?

• Raspberry Pi already has a segregated architecture, using 40nm for I/O functions and 16nm for the CPU and GPU
• Adding an NPU to the 16nm core would increase the die area and cost, which Raspberry Pi wanted to avoid
• By using a separate accelerator board, users who don't need the AI capabilities can avoid paying for the additional NPU die area

3. What are the current limitations of the Raspberry Pi AI Kit?

• The software support is not yet fully developed, and users may need to put in some "elbow grease" to get the AI features working
• Raspberry Pi OS updates are expected to enable more of the AI features in the near future
• The demo reviewed by Tom's Hardware team did not include speed measurements, but they felt that some of the AI tasks were faster

4. What are the potential use cases for the Raspberry Pi AI Kit?

• The low power advantage of the Hailo-8L accelerator makes it well-suited for computer vision and image recognition applications at the edge
• Raspberry Pi also mentions models suited for adding AI features to games, though the details are not yet clear
• The maker community is expected to come up with unique use cases to solve their own problems, especially in the field of robotics

[02] Future Outlook

1. How does Raspberry Pi see the future of the partnership with Hailo?

• Raspberry Pi CEO mentioned that the association with Hailo doesn't end with this AI kit, suggesting further collaboration in the future
• The article notes that it will be interesting to see what the future holds for edge AI, as the software support and capabilities continue to evolve