NVIDIA Jetson AGX Orin 64GB

Data-center AI performance at the edge

The NVIDIA Jetson AGX Orin 64GB is a compact compute module that delivers up to 275 TOPS of AI inference performance — approximately 8× the throughput of its predecessor — in a form factor small enough to embed in robots, drones, and industrial vision systems. It belongs to the top tier of NVIDIA's Jetson edge AI platform.

How it is built

The module integrates the following hardware into a single unit:

• GPU: 2048-core NVIDIA Ampere architecture with 64 Tensor Cores, the same GPU generation used in NVIDIA's data-center products
• CPU: 12-core Arm processor
• AI accelerators: Dual deep-learning accelerators (DLAs) for offloading dedicated inference workloads
• Memory: 64 GB LPDDR5, shared across CPU, GPU, and accelerators in a unified memory architecture
• Power envelope: Configurable between 15 W and 60 W, allowing the module to be tuned to the thermal and power constraints of a given deployment

Note on the 275 TOPS figure: This headline number reflects sparse INT8 computation. Real-world dense inference throughput is roughly half that value, which is the more relevant figure for most production workloads.

Software compatibility

The AGX Orin runs NVIDIA's full JetPack SDK, which includes CUDA, TensorRT, DeepStream, and the Isaac robotics framework — the same toolchain used on NVIDIA's data-center GPUs. This means AI models and processing pipelines developed in a data-center environment can generally be deployed to the AGX Orin without a fundamental rewrite, which is a significant practical advantage for teams moving from cloud-based development to edge deployment.

What it is used for

The AGX Orin 64GB is suited to applications that require processing multiple high-resolution sensor or camera streams locally, in real time, without relying on a network connection to a remote server:

• Autonomous mobile robots (AMRs): onboard perception, path planning, and sensor fusion
• Multi-camera perception systems: simultaneous ingestion and analysis of several video streams
• Medical imaging devices: on-device inference for imaging workloads in clinical or field settings
• Industrial inspection and vision: real-time defect detection and quality control on the factory floor
• Edge servers: local AI inference nodes in environments where latency or connectivity constraints rule out cloud processing

Who uses it

The module is primarily used by robotics engineers, computer vision researchers, and systems integrators building production hardware. The Jetson AGX Orin Developer Kit is the common starting point for development and prototyping; production deployments use the standalone module, typically integrated onto a custom carrier board.