Hailo-10H AI HAT+2 40 TOPS for Pi 5

Description

Hailo10

The Hailo-10H 40 TOPS AI HAT+ for Raspberry Pi 5 is a dedicated AI accelerator board that unlocks powerful generative AI and vision capabilities at the edge. Equipped with a Hailo-10H NPU delivering 40 TOPS (INT4) and 8GB of onboard memory, it enables local execution of large language models (LLMs) and vision-language models (VLMs) with up to ~6 billion parameters. Fully compliant with the Raspberry Pi HAT+ standard and plug-and-play ready, it seamlessly integrates with the Raspberry Pi camera stack and Hailo’s comprehensive toolchain, making it ideal for robotics, real-time vision analytics, and secure offline AI deployment.

40 TOPS AI Performance with Dedicated 8GB Memory

Features a Hailo-10H NPU delivering 40 TOPS (INT4) for efficient AI inference, paired with 8GB of onboard memory to run LLMs and VLMs locally without consuming Raspberry Pi 5’s main RAM—freeing the host for other tasks and enabling complex multitasking.

Plug-and-Play Compatibility with Full Camera & OS Support

Fully complies with Raspberry Pi HAT+ specification and connects via PCIe Gen3. Automatically recognized by the latest Raspberry Pi OS, it natively integrates with libcamera, rpicam-apps, and Picamera2 for accelerated AI vision and generative tasks right out of the box.

Comprehensive AI Toolchain & Ready-to-Run Model Library

Supported by Hailo’s complete software suite including model conversion, optimization, and deployment tools. Comes with pre-trained models for object detection, segmentation, pose estimation, and generative AI, plus tutorials for custom model development.

Proven Edge AI Acceleration for Real-Time Applications

Demonstrated to accelerate vision-language model inference from minutes to seconds (e.g., Qwen2-VL-2B) and boost frame rates in real-time computer vision tasks—ideal for robotics, industrial inspection, offline analytics, and low-latency edge AI solutions.

Optimized Thermal Design for Sustained Performance

Includes an attached heatsink to maintain stable performance under load, with support for active cooling solutions to prevent thermal throttling during continuous high-intensity AI workloads.