Official Raspberry Pi AI HAT+ 13 TOPS

Specifications:

• Operating Voltage: 5V DC (powered via Raspberry Pi GPIO)
• Neural Processing Unit (NPU): Hailo-8L
• Inferencing Performance: 13 Tera-Operations Per Second (TOPS)
• Communication Interface: Single-lane PCIe 2.0 / 3.0 (via 16-way FPC flex cable)
• Form Factor: Standard Raspberry Pi HAT format
• Software Integration: Fully integrated with Raspberry Pi OS camera software stack
• Supported Frameworks: TensorFlow, TensorFlow Lite, PyTorch, ONNX
• Operating Temperature: 0°C to 50°C
• Compatibility: Raspberry Pi 5 Only

Unlock New Capabilities

• If you've already experimented with running basic object detection on a standard Raspberry Pi CPU only to hit 2–3 frames per second, or if you've maxed out the capabilities of older USB-based neural sticks, you already understand the bottleneck of edge AI. You are ready to move from choppy, delayed inferencing to real-time hardware acceleration.
• This official AI HAT+ unlocks high-framerate, complex vision tasks locally on your Pi. You can now build real-time multi-object tracking systems for security, instant facial recognition gates, high-speed license plate readers, and fluid gesture-controlled robotics—all running locally without relying on expensive, latency-heavy cloud APIs.
• Effort Level: Intermediate

Minimum hardware to get started:

• Raspberry Pi 5: The HAT+ exclusively uses the Pi 5's dedicated single-lane PCIe interface; it will not work with older Pi models.
• Raspberry Pi 5 Active Cooler:  Heavy AI inferencing generates significant heat; without the official Active Cooler, the Pi 5 will thermally throttle and degrade your AI framerates.
• 27W USB-C Power Supply: The Hailo-8L module draws brief, heavy power spikes during complex matrix calculations; standard 15W supplies will cause brownouts.
• Raspberry Pi Camera (e.g., Camera Module 3): To provide the live visual data stream for the Hailo-8L to process.

Package Includes:

• 1 X Official Raspberry Pi AI HAT+ 13 TOPS

Frequently Asked Questions (FAQ)

• What NPU chip is on the Official Raspberry Pi AI HAT+ 13 TOPS?
  It uses the Hailo-8L M.2 AI acceleration module running on a custom HAT+ carrier board.
• Is the Raspberry Pi AI HAT+ 13 TOPS compatible with Raspberry Pi 4?
  No, it is exclusively compatible with the Raspberry Pi 5 because it relies on the PCIe Gen 2/3 ribbon interface introduced on the Pi 5.
• What is the difference between the 13 TOPS and 26 TOPS AI HAT+ models?
  The 13 TOPS version utilizes the Hailo-8L module (ideal for entry to mid-level vision tasks), while the 26 TOPS version utilizes the full Hailo-8 module, offering double the raw inference compute.
• Does the 13 TOPS AI HAT+ require an external power supply?
  No separate power input is needed; it draws power through the Raspberry Pi 5 PCIe interface and GPIO header.
• Why do I need a 27W power supply when using the AI HAT+?
  The AI HAT+ draws additional current during high-speed inferencing. A standard 15W supply may cause under-voltage throttling on the Pi 5.
• How does the AI HAT+ connect to the Raspberry Pi 5?
  It connects via a 16-pin PCIe FPC ribbon cable and the standard 40-pin GPIO header.
• Can I use the GPIO pins while the AI HAT+ is installed?
  Yes, it includes a pass-through 40-pin stacking header allowing full access to unused GPIO pins.
• Which neural network frameworks are supported by Hailo-8L?
  It supports TensorFlow, TensorFlow Lite, PyTorch, ONNX, and Keras through compilation via the Hailo Dataflow Compiler.
• Can I run YOLOv8 natively on the 13 TOPS AI HAT+?
  Yes, YOLOv8 models can be compiled to Hailo .hef binary format and run efficiently at high frame rates.
• Does the Raspberry Pi Camera software natively support this AI HAT+?
  Yes, Raspberry Pi OS libcamera and Picamera2 frameworks feature native integration for post-processing AI pipelines.
• What is the frame rate for object detection on 1080p video?
  Depending on model complexity (e.g., YOLOv8s vs YOLOv8n), frame rates typically range between 30 FPS to 60+ FPS.
• Does the board heat up significantly during continuous inferencing?
  The Hailo-8L module is low-power (~2.5W to 5W peak), but using the Raspberry Pi Official Active Cooler is recommended to maintain system-wide thermal balance.