Artificial intelligence has already taken over, but now it faces new challenges. One of them is efficiency: sustaining so many online models demands a lot of energy, and some companies bet on "performance per watt" to promote their products. That brings us to the Raspberry Pi AI Kit, a hardware package compatible with the Raspberry Pi 5 designed to accelerate "machine learning" tasks such as object recognition, pose detection, image segmentation, and other applications that rely on a camera.
Want a local artificial intelligence? These days it's enough to install Ollama and download some compatible models. You'll need some firepower (including disk space) and a good dose of patience when they start to hallucinate, but the process is not a headache, and all the software is free.
However, there are other applications that go beyond chat. The Raspberry Pi Foundation points in that direction with its new product, the Raspberry Pi AI Kit. This package includes two components: an M.2 HAT+ adapter and the accelerator module itself, based on the Hailo-8L chip.
What can you do with the Raspberry Pi AI Kit?
Jeff Geerling explains in his latest video that the strength of the Raspberry Pi AI Kit lies in visual tasks. Most notably, the Hailo-8L chip delivers between 3 and 4 TOPS per watt, beating its direct competitors in efficiency and price. Now, there's no point in training a model on this hardware (you'll have to wait several days), but the performance leap in certain environments is remarkable. For example, YOLOv5 runs at two or three frames per second on the Raspberry Pi's CPU, but the offloading is complete with the AI module, freeing the processor for other applications.
Obviously, this doesn't mean the Raspberry Pi AI Kit is suitable for every workload. Image generation and language models (read 'ChatGPT') are things that require much more hardware… but it's not impossible either. If the module is accessible and well-documented, the software will appear, and with a $70 price tag, it will probably make more noise than we imagine.
Official site: Click here
