Tel Aviv-based AI chip maker Hailo has launched its latest processor family, the Hailo-15 — which includes three variants, the Hailo-15H, Hailo-15M, and Hailo-15L.
These new processors are designed to bring AI models to the edge, particularly in intelligent cameras. The Hailo-15 family runs from 7 tera operations per second (TOPS) to 20 TOPS, which is five times faster than similarly priced solutions. This release brings vision-based transformers for real-time object detection to cameras.
The new Hailo family can run real-time detection models like YOLOv5 at a resolution of 1280 x 1280 and a benchmark ResNet-50 classification model at 700 frames per second. For video input, it can handle up to 4k60 if real-time processing is not needed or if the number of frames analyzed per second can be small.
Hailo-15-powered cameras offer several advantages over traditional cameras. By offloading the cloud analytics load, they save video bandwidth and processing, reduce the total cost of ownership for large-scale camera deployments, and ensure greater privacy by anonymizing data at the edge. As a result, ultra-high-quality AI-powered video analytics solutions provide safety, ensure privacy, reduce network infrastructure costs and complexity, and enable organizations to operate more efficiently.
Hailo has expanded on the revolutionary architecture it developed for the original chip, allowing it to map the physical resources according to the neural networks running on it. This keeps the computing, memory, and control units closer together for optimal performance and turns off parts of the processor that are not needed.
Overall, the launch of the Hailo-15 family marks an important step in making Edge AI more scalable and affordable. With its breakthrough performance and comprehensive software suite, it has the potential to revolutionize the way we approach intelligent cameras and video analytics solutions.