NVIDIA® Jetson™ Thor Edge AI for Physical AI

EyePop.ai on NVIDIA Jetson Thor brings real-time computer vision to the most powerful Jetson platform yet, delivering server-class detection, tracking, and generative-AI inference at the edge, no cloud round-trip required.

Qualcomm Dragonwing IQ-30175 EVK small electronic development board with various ports and slots.
Circular graphic with dots

Trusted by industry leaders

How it works

EyePop.ai's inference engine is built to take full advantage of Jetson Thor's Blackwell GPU, routing detection, tracking, and generative-AI workloads across the module's GPU, CPU, and always-on DSP. Point a camera feed at a Thor-based system, and within seconds you're receiving live, structured predictions — no cloud dependency, no added latency.

The result? Server-class vision, running at the edge.

Start building

1,000+ TOPS

Blackwell GPU AI performance

Up to 128GB

LPDDR5X memory

75–130W total power

Configurable power modes

1
Camera feeds
Up to 6 cameras, CSI-2/Holoscan
Two white security cameras mounted on adjustable wall brackets.
2
Jetson Thor
NVIDIA Jetson Thor SoM
Diagram showing video decode/encode, detection and tracking, and GPU pre/post-processing functions.
3
EyePop runtime
Docker container, on-device
Logo of a light blue whale with a stack of blue squares on its back.
4
Structured output
Boxes, tracks, JSON
Icon of a document with code brackets and two horizontal lines representing text.

NVIDIA® Jetson™ Thor

Checkmark icon
Blackwell GPU architecture — 2,560 CUDA cores, 96 5th-gen Tensor Cores, 1,000+ TOPS AI performance
Checkmark icon
14-core Arm Neoverse V3AE CPU (up to 2.6GHz) for real-time robotics and control workloads
Checkmark icon
128GB LPDDR5X memory (273GB/s bandwidth) for large multimodal and generative-AI models at the edge
Checkmark icon
4× 25GbE networking (100Gbps total) plus PCIe Gen5 for high-speed sensor throughput
Checkmark icon
16× CSI-2 camera lanes (up to 6 cameras); Holoscan Sensor Bridge support for multi-camera vision
Checkmark icon
Configurable 75–120W power envelope (max 130W)
Compact black and gray electronic device with USB ports, power button, headset jack, and cooling fins on top.

Designed uniquely for…

Computer monitor interface icon.

Developers

Build smarter by embedding real-time object detection, tracking, and classification onto Snapdragon Dragonwing hardware.

Read the dev docs

An external link icon with an arrow pointing up and to the right inside a box.
Lightbulb puzzle pieces icon.

Business leaders

Stay ahead of competitors by deploying Dragonwing-accelerated vision AI on-site, cutting bandwidth costs and eliminating cloud latency.

Icon of a target with a bullseye

Industry innovators

Explore new ways to bring Snapdragon Dragonwing's compute to factories, fleets, and cities where connectivity and power are constrained.

Build Edge AI

It's your turn to explore how Snapdragon Dragonwing-accelerated computer vision can help your business detect, analyze, and act on visual data at the edge, in real time and at scale.