NVIDIA Jetson

Definition

NVIDIA Jetson is a line of small, low-power embedded computing boards that combine an ARM-based CPU with an NVIDIA GPU, designed to run AI models directly on robots, drones, cameras, and other edge devices rather than in the cloud. Jetson modules run the same CUDA software stack used on NVIDIA's larger GPUs, letting developers train a model on a workstation or in the cloud and then deploy it to run locally on a Jetson board with limited changes. The line spans entry-level modules for hobbyists up to more powerful boards for industrial robotics.

Overview

Jetson addresses a problem distinct from NVIDIA's data center and consumer GPU lines: running AI inference directly on a physical device in the field, where sending data to the cloud for processing would be too slow, too costly in bandwidth, or impossible due to unreliable connectivity. Robots, drones, security cameras, and industrial sensors often need to interpret camera or sensor data in real time on the device itself, and Jetson boards are built specifically to make that practical within tight power and size budgets. Mechanically, a Jetson module packages an ARM-based multi-core CPU alongside an NVIDIA GPU with CUDA cores and, on higher-end variants, dedicated tensor cores for AI acceleration, all on a compact circuit board designed to be embedded into another product rather than used as a standalone computer. Power consumption is measured in watts rather than the hundreds of watts a data center GPU draws, since Jetson devices often run on batteries or limited power budgets in robots and drones. NVIDIA provides a software layer, JetPack, that includes drivers, libraries, and the CUDA toolkit adapted to run on the embedded ARM processor. Within NVIDIA's broader product lineup, Jetson is positioned distinctly from Jetson's data center cousins like the H100 or gaming-focused GeForce cards: it prioritizes power efficiency and physical compactness over raw throughput, since its target workloads run one model on one device rather than training massive models across a cluster. Compared with other embedded AI hardware from companies like Qualcomm or Google's edge TPU offerings, Jetson's main differentiator is compatibility with the same CUDA ecosystem developers already use for training on larger NVIDIA GPUs, easing the path from development to deployment. In practice, Jetson boards appear in autonomous mobile robots, drones performing real-time object detection, smart cameras doing on-device video analytics, and industrial equipment running quality-inspection AI models, as well as in robotics research and hobbyist projects due to lower-cost entry-level modules. Manufacturers often embed Jetson modules directly into their own products rather than exposing them to end users. The main trade-off is that Jetson's power and size constraints cap how large or complex a model it can run compared with cloud or data center inference, so models often need to be compressed or optimized specifically for edge deployment before running well on a Jetson board. Developers also need to manage the ARM-based embedded Linux environment, which differs in some respects from a standard desktop Linux setup. Projects that can tolerate cloud latency and connectivity requirements, or that need to run very large models, typically rely on cloud-based inference instead of Jetson.

Key Features

• Combines an ARM CPU with an NVIDIA GPU on a compact embedded board
• Runs CUDA software stack shared with NVIDIA's larger GPU products
• Optimized for low power consumption suited to battery-powered devices
• JetPack software layer bundles drivers, libraries, and CUDA tooling
• Spans entry-level hobbyist modules to industrial-grade robotics boards
• Enables on-device AI inference without cloud connectivity
• Used to embed AI directly into robots, drones, and cameras
• Distinct from NVIDIA's data center and gaming GPU product lines

Use Cases

Alternatives

Frequently Asked Questions

Related in Computer Science