NVIDIA Jetson Nano 4G: Edge Computing and Embedded AI

1. Jetson Nano Hardware Introduction

NVIDIA Jetson Nano is a compact artificial intelligence computer designed specifically for edge computing and embedded AI applications. It has gained significant attention in the AI field due to its powerful performance and low power consumption.

Hardware Specifications

• Size and Power Consumption: The Jetson Nano development board measures only 70mm x 45mm, smaller than a credit card, with a power requirement between 5 to 10 watts.

• GPU and CPU: This development board is equipped with a 128-core NVIDIA Maxwell architecture GPU and a quad-core ARM Cortex-A57 CPU, with the GPU’s maximum frequency at 921MHz and the CPU’s maximum frequency at 1.43GHz.

• Memory and Storage: It is equipped with 4GB of 64-bit LPDDR4 memory and 16GB of eMMC storage.

• Multimedia Capabilities: Supports one 4K@30p video encoding, or four 1080p@30, or two 1080p@60, as well as one 4K@60p, or two 4K@30, or eight 1080p@30 video decoding.

• Interfaces and Expansion: Provides Gigabit Ethernet, HDMI 2.0, DP1.2 display interfaces, PCIe, USB 3.0 high-speed interfaces, as well as UART, SPI, I2C, I2S, GPIOs for expansion.

Software Support

• Operating System and Framework: Jetson Nano supports the Linux for Tegra operating system and provides a complete Linux BSP and L4T package for developers to quickly get started. It also supports various neural network frameworks such as TensorFlow, PyTorch, Caffe/Caffe2, Keras, and MXNet.

• Pre-trained Models: Developers can quickly implement image classification, object detection, language translation, and other functions through pre-trained models in the NGC™ catalog.

Application Areas

• Embedded AI Applications: Jetson Nano is suitable for entry-level Network Video Recorders (NVR), home robots, and smart gateways with comprehensive analysis capabilities in embedded IoT applications.

• Autonomous Robots and Complex AI Systems: With its powerful computing performance and low power consumption, Jetson Nano can be used to build autonomous robots and complex artificial intelligence systems.

1.1 Application Scenarios

• Autonomous Driving Cars

• JD.com Autonomous Delivery Vehicle

1.2 Interface Introduction

• Core Board

  

• Base Board

• Base Board + Core Board Assembly

1.3 TF Card

• It is recommended to be at least 32GB. Depending on usage, it can be 64GB.

• The quality of the card must be good, with fast read and write speeds.

• Previously, a 128GB card of the same brand was much slower than a 64GB one. So, appropriate size is sufficient.

1.4 Camera

• USB Camera

• Raspberry Pi V2 CSI Camera (IMX219)

• CSI does not support hot swapping, install before booting

1.5 Wireless Network Card

• Wired 1000M network port connection

• USB Wireless Network Card

• Wireless Network Card AC8265 (Smart Car)

1.6 Development Package

2. Jetson Nano Flashing

2.1 SD Card Formatter Installation

Download link:

https://www.sdcard.org/downloads/formatter/

We provide the downloaded software

2.2 BalenaEtcher Installation

• Download link: https://www.balena.io/etcher/

• We provide the downloaded software

2.3 NVIDIA JetPack Introduction

• Includes OPENCV version 4.1

• Supports DeepStream version 6.0

• Download link:

https://developer.nvidia.com/embedded/jetpack-sdk-461

2.4 Practical Flashing Process

• Insert TF card reader into the computer

• Warning:

• The card reader must also support USB 3.0 or higher

• The computer’s USB port must also be USB 3.0 or higher

3. Jetson Nano System Installation Process

3.1 Agree to the Agreement Interface

3.2 Select Language

3.3 Select Keyboard Layout

3.4 Select Time Zone, Click the Red Circle, Select Shanghai

3.5 Fill in Username and Password (Choose Whether to Auto-Login Below)

3.6 Wait for System Configuration

3.7 Restart After Configuration

3.8 Enter the System

4. Experience the GPU Computing Power of Nano

4.1 CUDA Example

• Default installation directory for CUDA: /usr/local/cuda/

• CUDA can usually be added to the system path

• export CUDA_HOME=/usr/local/cuda

• export PATH=/usr/local/cuda/bin:$PATH

•export LD_LIBRARY_PATH=/usr/local/cuda/lib64/:$LD_LIBRARY_PATH

4.2 View CUDA Demo

• /usr/local/cuda/samples/

• /usr/local/cuda/samples/5_Simulations/fluidsGL

•/usr/local/cuda/samples/5_Simulations/smokeParticles

4.3 View VisionWorks Example

• Install and compile examples:

• cd /usr/share/visionworks/sources/

• sudo ./install-samples.sh ~/

• cd ~/VisionWorks-1.6-Samples

• sudo make -j4

• View examples:

• ~/VisionWorks-1.6-Samples/bin/aarch64/linux/release

• ./nvx_demo_feature_tracker Feature Tracking

• ./nvx_demo_hough_transform Hough Transform

• ./nvx_demo_motion_estimation Motion Estimation

• ./nvx_demo_video_stabilizer Video Stabilizer

Summary:

Overall, the NVIDIA Jetson Nano development board, with its compact size, powerful computing performance, rich interfaces, and wide application areas, has become an ideal choice for embedded AI applications. Whether for beginners or professional developers, Jetson Nano allows for easy prototyping and productization of AI projects.

Finally, it is said that the newly released Jetson Orin Nano Super is much more powerful.