Jetson Orin

Practical

Jetson Orin is NVIDIA’s edge AI computing platform launched in 2022, designed for robotics, autonomous machines, and AI applications requiring real-time performance at the edge. All Orin modules remain in active production with support extended through January 2032.

Status: Active Production

All Orin modules are in active production with LTS extended through January 2032. JetPack 6.2 introduced “Super Mode” enabling higher performance on NX and Nano modules.

The Orin Family

AGX Orin

Jetson AGX Orin — The flagship module

Spec	64GB	32GB
AI Performance	275 TOPS (INT8)	200 TOPS (INT8)
GPU	2048 CUDA cores, 64 Tensor cores	1792 CUDA cores, 56 Tensor cores
CPU	12-core Arm Cortex-A78AE	8-core Arm Cortex-A78AE
Memory	64GB LPDDR5	32GB LPDDR5
Power	15W - 60W	15W - 40W
Status	Long-term support	Long-term support

Best for: Production deployments, multi-sensor fusion, existing Orin-based designs

Orin NX

Jetson Orin NX — High performance, compact form factor

Spec	16GB	8GB
AI Performance	100 TOPS (157 w/ Super Mode)	70 TOPS (117 w/ Super Mode)
GPU	1024 CUDA cores	1024 CUDA cores
CPU	8-core Arm Cortex-A78AE	6-core
Memory	16GB LPDDR5	8GB LPDDR5
Power	10W - 25W (40W Super Mode)	10W - 20W
Status	Active production

Best for: Delivery robots, drones, industrial AMRs, cost-sensitive applications

Orin Nano

Jetson Orin Nano — Entry-level AI at the edge

Spec	8GB	4GB
AI Performance	40 TOPS (67 w/ Super Mode)	20 TOPS (32 w/ Super Mode)
GPU	1024 CUDA cores	512 CUDA cores
CPU	6-core Arm Cortex-A78AE	6-core
Memory	8GB LPDDR5	4GB LPDDR5
Power	7W - 15W (25W Super Mode)	5W - 10W
Status	Active production

Best for: Entry-level robots, education, prototyping, high-volume deployments

Orin vs Thor: Which to Choose?

Consideration	Choose Orin	Choose Thor
AI Performance needed	<300 TOPS	2,070 TFLOPS (FP4)
Budget	Cost-sensitive	Performance-critical ($3,499 dev kit)
Existing design	Migrating from Xavier	New humanoid/advanced robot
Transformer workloads	Limited	Native Transformer Engine
Memory	Up to 64GB	128GB LPDDR5X

Architecture Overview

┌─────────────────────────────────────────────────────────────┐
│                    Jetson Orin SoC                          │
├─────────────────────────────────────────────────────────────┤
│  ┌─────────────────┐  ┌─────────────────────────────────┐  │
│  │   Arm CPU       │  │        NVIDIA Ampere GPU        │  │
│  │  Cortex-A78AE   │  │  ┌─────────┐  ┌─────────────┐  │  │
│  │  Up to 12 cores │  │  │  CUDA   │  │   Tensor    │  │  │
│  │                 │  │  │  Cores  │  │   Cores     │  │  │
│  └─────────────────┘  │  └─────────┘  └─────────────┘  │  │
│                       └─────────────────────────────────┘  │
│  ┌─────────────────┐  ┌─────────────────────────────────┐  │
│  │  Deep Learning  │  │         Video Engines           │  │
│  │  Accelerator    │  │  NVENC │ NVDEC │ JPEG │ OFA    │  │
│  │  (DLA x2)       │  │                                 │  │
│  └─────────────────┘  └─────────────────────────────────┘  │
│  ┌─────────────────────────────────────────────────────┐   │
│  │              Memory: LPDDR5 (256-bit)               │   │
│  └─────────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────────┘

Key Components

• Ampere GPU: CUDA and Tensor cores for parallel compute and AI inference
• DLA (Deep Learning Accelerator): Dedicated AI inference engines (2x on AGX Orin)
• PVA (Programmable Vision Accelerator): Computer vision preprocessing
• Video Engines: Hardware encode/decode for camera streams

Software Stack

JetPack 6.x (LTS)

# JetPack 6.2.1 - Long Term Support (Recommended for Orin)
# Ubuntu 22.04, CUDA 12.6, TensorRT 10.3, cuDNN 9.3
sudo apt update
sudo apt install nvidia-jetpack

Recommended for production Orin deployments requiring stability.

JetPack 7.x

# JetPack 7.x - Latest features (Primary target: Thor)
# Ubuntu 24.04, CUDA 13.0, Linux Kernel 6.8
# Orin support expected in JetPack 7.2 (Q1 2026)

Primary target is Jetson Thor; Orin support expected in JetPack 7.2.

┌─────────────────────────────────────────┐
│           Your Application              │
├─────────────────────────────────────────┤
│  Isaac ROS │ ROS 2 │ DeepStream │ TAO  │
├─────────────────────────────────────────┤
│  TensorRT │ cuDNN │ CUDA │ OpenCV      │
├─────────────────────────────────────────┤
│           JetPack SDK (L4T)             │
├─────────────────────────────────────────┤
│        Linux Kernel + Drivers           │
└─────────────────────────────────────────┘

Getting Started

1. Flash the Device

# Using NVIDIA SDK Manager (recommended)
# Or command line:
sudo ./flash.sh jetson-agx-orin-devkit internal

2. Install JetPack Components

sudo apt update
sudo apt install nvidia-jetpack

3. Verify Installation

# Check CUDA
nvcc --version

# Check TensorRT
dpkg -l | grep tensorrt

# Monitor system
tegrastats

Power Management

Orin supports multiple power modes via nvpmodel:

# List available modes
sudo nvpmodel -q --verbose

# Set to max performance (AGX Orin)
sudo nvpmodel -m 0  # MAXN: 60W

# Set to power-efficient mode
sudo nvpmodel -m 3  # 30W

# Maximize clocks (for benchmarking)
sudo jetson_clocks

Mode	AGX Orin Power	Use Case
MAXN	60W	Maximum performance
50W	50W	High performance
30W	30W	Balanced
15W	15W	Power-constrained

Related Terms

Jetson Thor Next-gen platform for humanoid robots (2,070 TFLOPS)

Isaac ROS GPU-accelerated ROS 2 packages for Jetson

DGX Spark Desktop AI supercomputer for development

Learn More

• Jetson Orin Official Page
• JetPack 6.x LTS Documentation
• Orin → Thor Migration Guide
• Jetson Developer Forums

Sources

• NVIDIA Jetson Orin — Official specifications
• NVIDIA AGX Orin Technical Brief — AGX Orin 64GB/32GB variant specifications
• NVIDIA JetPack SDK — Current JetPack 6.2.1 versions
• JetPack 6.2 Super Mode Blog — Super Mode feature details
• Jetson Product Lifecycle — LTS extended through January 2032
• NVIDIA Newsroom - Jetson Thor — Thor availability and specs