Run Vision AI on NVIDIA Jetson at the Edge

NVIDIA Jetson AI Vision Solution

Jetson puts an NVIDIA GPU in a small, low-power module, which makes it the go-to box for running vision models on a production line, inside a robot, or on a pole. Roboflow Inference runs on JetPack 5 and 6 (opens in a new tab): one command picks the right container for your device, and TensorRT acceleration is available when you need faster models.

The Jetson Orin Nano Super Developer Kit (opens in a new tab) (67 TOPS, 7 to 25 W) is the place to start. For production, we recommend Jetson Orin NX (opens in a new tab) with 16 GB or more, while AGX Orin (up to 275 TOPS) handles many camera streams at once. Jetson AGX Thor (opens in a new tab) is built for the heaviest workloads, such as running vision-language models at the edge.

• Featured: Orin Nano Super Developer Kit, Orin NX 16 GB, AGX Orin, AGX Thor
• Attributes: NVIDIA GPU in a compact module, low power draw, TensorRT acceleration, wide range of carrier boards and rugged enclosures
• Integration: Run Roboflow Inference on JetPack, connect USB, CSI, RTSP, or GigE cameras, and send results to PLCs, MES, or alerting systems

Label images from your cameras, train a model in Roboflow, build a Workflow that counts, tracks, or flags what matters, then deploy it to the Jetson with Roboflow Inference. Your team owns the model and the data, and can retrain as conditions change.

‍

Want AI without the integration work? Roboflow's AI1 deployment device (opens in a new tab) packages a Jetson Orin NX, integrated lighting, and an industrial camera into a plug-and-play system for real-time computer vision, with no separate camera, lighting, or edge computer to spec.