Jetson Nano开箱(图为&沥拓)

1、Jetson Nano 官方资料下载
Jetson Nano Developer Kit官方介绍
Get-Started-With-Jetson-Nano-Devkit
Jetson-Nano-Dev-Kit-Sd-Card-Image
NVIDIA Jetson-Nano-Resources
Jetson Nano Wiki
Jetson Nano Upstream

图为科技Jetson Nano 运行通用案例分享(持续更新）

Jetson Nano 对应的DeepStream 2019Q2 才会发布，目前采用Nano 做的项目还是比较受限，期待DS 有更高的性能表现！

image.png

2、Jetson Nano刷机

Nano MicroSD卡刷机工具：SDCardFormatter 和 balenaEtcher
Nano MicroSD卡系统镜像：jetson-nano-sd-r32.1-2019-03-18.zip (5.6GB)，但解压缩后大约12GB多，运行在Micro SD卡，因此SD卡务必16GB及以上且高速UHS（参考NV 说明)，否则刷TF卡镜像和系统软件运行时会慢如牛，你想砸了Nano或TF卡!
MicroSD卡选型可参考文章: 一文看懂各种储存卡:TF卡、SD/SDHC/SDXC卡、CF卡和Class等级

image.png

TF卡（Nano运行）测试方法 （无其他软件运行)：
SanDisk --- UHS-I接口兼容(Speed: Class10 + A1)

写入速度测试：
$ dd count=1k bs=1M if=/dev/zero of=/home/nvidia/tst.img
1024+0 records in
1024+0 records out
1073741824 bytes (1.1 GB, 1.0 GiB) copied, 47.5029 s, 22.6 MB/s

读取速度测试：
$ sudo apt-get install hdparm
$ sudo hdparm -t /dev/mmcblk0p1
/dev/mmcblk0p1:
 Timing buffered disk reads: 244 MB in  3.01 seconds =  81.19 MB/sec

SanDisk --- UHS-I接口兼容(Speed: U3 + V30 + A1)

读取速度测试：
$ sudo apt-get install hdparm
$ sudo hdparm -t /dev/mmcblk0p1
/dev/mmcblk0p1:
 Timing buffered disk reads: 258 MB in  3.01 seconds =  85.61 MB/sec

写入速度测试：
nvidia@tw-Nano:~$ dd count=1k bs=1M if=/dev/zero of=/home/nvidia/tst.img
1024+0 records in
1024+0 records out
1073741824 bytes (1.1 GB, 1.0 GiB) copied, 8.68056 s, 124 MB/s

刷机过程如图：

image.png

3、供电USB 5V 2A 供电自动开机（外加了一个风扇）

image.png

AC Adapter 必须5V 开机

image.png

4. Jetson Nano套件特性

硬件接口

• Ports & Interfaces
• 4x USB 3.0 A (Host)
• USB 2.0 Micro B (Device)
• MIPI CSI-2 x2 (15-position Camera Flex Connector)
• HDMI 2.0
• DisplayPort
• Gigabit Ethernet (RJ45)
• M.2 Key-E with PCIe x1
• MicroSD card slot
• (3x) I2C, (2x) SPI, UART, I2S, GPIOs

软件列表

• JetPack 4.2
• Linux4Tegra R32.1 (L4T)
• Linux kernel 4.9
• Ubuntu 18.04 LTS aarch64
• CUDA Toolkit 10.0
• cuDNN 7.3.1
• TensorRT 5.0.6
• TensorFlow 1.31.1
• VisionWorks 1.6
• OpenCV 3.3.1
• OpenGL 4.6
• OpenGL ES 3.2
• EGL 1.5
• Vulkan 1.1
• GStreamer 1.14.1
• V4L2 media controller support

5. 指导资料和教程文档

System Tools系统工具

• L4T Kernel Development Guide
• Power Supply Considerations
• Upstream Development Guide

Deep Learning深度学习资料

• Hello AI World (jetson-inference)
• TensorFlow 1.13.1 Installer (pip wheel)
• PyTorch 1.1 Installer (pip wheel)
  See the NVIDIA AI-IoT GitHub for other coding resources on deploying AI and deep learning.

Robotics 机器人应用

• NVIDIA JetBot (AI-powered robotics kit)
• jetbot_ros (ROS nodes for JetBot)
• ROS Melodic (ROS install guide)
• ros_deep_learning (jetson-inference nodes)

Camera 模组

• Leopard Imaging LI-IMX219-MIPI-FF-NANO
• Raspberry Pi Camera v2 (IMX219)
• Stereolabs ZED (stereo camera)

CUDA 10（路径：/usr/local/cuda/，但未加入环境变量）
运行如下命令添加环境变量：
export PATH=${PATH}:/usr/local/cuda/bin
export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/local/cuda/lib64

sudo apt-get update
sudo apt-get install samba
sudo apt-get install python3-pip

Tensorflow 1.13.1
pip3 install --extra-index-url https://developer.download.nvidia.com/compute/redist/jp/v42 tensorflow-gpu==1.13.1+nv19.3 --user
安装过程若报错hdf5，请安装：
sudo apt-get install libhdf5-serial-dev

PS: Nano features list:
Nano 不支持SATA 硬盘，但可以通过如下方式：
1.USB3 SATA dongle
2.M.2 Key-E to Mini-PCIe adapter (未验证）
"$sudo nvpmodel -m 1" ---> Nano run in 5W (Power：5V1A)
"$sudo nvpmodel -m 0" ---> Nano run in 10W (Power：5V2A)
Supported video codecs: H.265, H.264, VP8, VP9 (VP9 decode only)
HEVC encoder supports 10-bit color, but B-frames are not supported