By Bobby Jefferson on Thursday, 24 October 2024
Category: Tech News

New Raspberry Pi HAT+ board doubles AI processing performance - where to find it

Raspberry Pi AI HAT+ with Hailo-8 AI accelerator.

Raspberry Pi

Artificial intelligence (AI) is everywhere -- and the Raspberry Pi foundation is taking notice. Hot on the heels of the release of an  M.2 2242-format board with the powerful Hailo 8L AI chip that was designed to fit into the M.2 HAT+ board are boards with Hailo accelerator chips directly integrated onto the main PCB.

Also: Raspberry Pi just released its own high-performance SSDs and SSD kits

So why integrate the chip into the PCB? Two reasons. First, it simplifies the setup as you just need a single board rather than the AI board and the M.2 board (although Raspberry Pi did offer both as a kit, but it's still two boards). But the biggest advantage is that by integrating the two, the engineers have achieved better thermal dissipation.

And this better cooling has allowed Raspberry Pi to offer the board in two speeds. First is the standard 13 TOPS (tera-operations per second) Hailo-8L AI accelerator chip found in the earlier M.2 board, but also add to the lineup a board powered by the Hailo-8 AI accelerator which can deliver 26 TOPS. 

Also: How to add AI superpowers to your Raspberry Pi

More power allows you to do more AI with the board in real time.

Both versions of the AI HAT+ are fully backward compatible with the AI Kit, and the Hailo accelerator integration in the camera software stack works in exactly the same way with the AI HAT+, so nothing needs to change.

One thing to be aware of, though, is that while the neural network model compiled for the Hailo-8L will run fine on the more powerful Hailo-8 chip, models built to take advantage of the extra power offered by the Hailo-8 may not work on the Hailo-8L.

The 13 TOPS HAT+ running the same Hailo-8L accelerator as the AI Kit is priced at $70, while the more powerful 26 TOPS HAT+ equipped with the Hailo-8 accelerator is $110. Right now, stocks are limited, but I've found that CanaKit has both in stock. 

Original link
Leave Comments