Skip to content

Instantly share code, notes, and snippets.

Show Gist options
  • Save lgg/a81314128ee8044c82fe5c68ec3ca0f2 to your computer and use it in GitHub Desktop.
Save lgg/a81314128ee8044c82fe5c68ec3ca0f2 to your computer and use it in GitHub Desktop.
Tips and Tricks for Jetson AGX Xavier

Tips and Tricks for Jetson AGX Xavier

This document contains useful software and links to documentation that we have found to be useful when working with Jetson AGX Xavier.

Copied from: https://git.its.aau.dk/WW82ZE/docs_xavier/src/branch/master

Links

Nvidia Developer forum Jetson AGX Xavier topic

Jetson Community Projects

Embedded Linux wiki with many further links

Further links to resources

Jetson Nano tips:

Build Jetson AGX Kernel and Modules:

NVIDIA Container Runtime on Jetson:

Utils:

Software packages

JetPack

JetPack 4.6 Production Release with L4T 32.6.1

Nvidia Jetson-specific components are collected in JetPack. They can be installed as Debian packages

This enables, for example, installation of all JetPack components via the nvidia-jetpack metapackage. These commands on the developer kit will result in a full JetPack install:

sudo apt update
sudo apt install nvidia-jetpack

To view individual packages which are part of nvidia-jetpack metapackage, enter the command:

sudo apt-cache show nvidia-jetpack

Boot from NVME

git clone https://github.com/jetsonhacks/rootOnNVMe.git
cd rootOnNVMe
./copy-rootfs-ssd.sh
./setup-service.sh
sudo reboot
# If you want to boot from SD again, remove the file /etc/setssdroot.conf from the SD card.

Remove unused packages

sudo apt remove --purge libreoffice*
sudo apt remove --purge thunderbird*
sudo apt clean
sudo apt autoremove

Remove Docker

If you don't use a docker container, remove it. Docker daemons use some system resources.

apt-get remove docker docker-engine docker.io containerd runc

Docker Default Runtime

https://github.com/dusty-nv/jetson-containers/#docker-default-runtime

To enable access to the CUDA compiler (nvcc) during docker build operations, add "default-runtime": "nvidia" to your /etc/docker/daemon.json configuration file before attempting to build the containers:

{
    "runtimes": {
        "nvidia": {
            "path": "nvidia-container-runtime",
            "runtimeArgs": []
        }
    },

    "default-runtime": "nvidia"
}

You will then want to restart the Docker service or reboot your system before proceeding.

NVIDIA Container Runtime on Jetson

# Allow containers to communicate with Xorg
$ sudo xhost +si:localuser:root
$ sudo docker run --runtime nvidia --network host -it -e DISPLAY=$DISPLAY -v /tmp/.X11-unix/:/tmp/.X11-unix nvcr.io/nvidia/l4t-base:r32.3.1

root@nano:/# apt-get update && apt-get install -y --no-install-recommends make g++
root@nano:/# cp -r /usr/local/cuda/samples /tmp
root@nano:/# cd /tmp/samples/5_Simulations/nbody
root@nano:/# make
root@nano:/# ./nbody

Building CUDA in Containers on Jetson

Docker gives you the ability to build containers using the “docker build” command. Let's start with an example of how to do that on your Jetson device:

$ mkdir /tmp/docker-build && cd /tmp/docker-build
$ cp -r /usr/local/cuda/samples/ ./
$ tee ./Dockerfile <<EOF
FROM nvcr.io/nvidia/l4t-base:r32.3.1

RUN apt-get update && apt-get install -y --no-install-recommends make g++
COPY ./samples /tmp/samples

WORKDIR /tmp/samples/1_Utilities/deviceQuery
RUN make clean && make

CMD ["./deviceQuery"]
EOF

$ sudo docker build -t devicequery .
$ sudo docker run -it --runtime nvidia devicequery

CUDA Device Query (Runtime API) version (CUDART static linking)

Detected 1 CUDA Capable device(s)

Device 0: "Xavier"
  CUDA Driver Version / Runtime Version          10.0 / 10.0
  CUDA Capability Major/Minor version number:    7.2
  Total amount of global memory:                 15692 MBytes (16454430720 bytes)
  ( 8) Multiprocessors, ( 64) CUDA Cores/MP:     512 CUDA Cores
  GPU Max Clock rate:                            1500 MHz (1.50 GHz)
  Memory Clock rate:                             1377 Mhz
  Memory Bus Width:                              256-bit
  L2 Cache Size:                                 524288 bytes
...

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 10.0, CUDA Runtime Version = 10.0, NumDevs = 1
Result = PASS

Known limitation: The base l4t image doesn’t allow you to statically compile with all CUDA libraries. Only libcudadevrt.a and libcudart_static.a are included.

[Jetson Nano] Increasing swap memory

When you build a large software packages like openCV, you may experience an out of memory phenomenon. Increasing the swap file size can prevent this malfunction.

git clone https://github.com/JetsonHacksNano/installSwapfile
cd installSwapfile
./installSwapfile.sh

Above script file will increase 6GB swap files. You can change the swap file size by modifying the scripts. If you want to uninstall the swap setting, open the fstab file and delete the swap file line and reboot.

sudo vi /etc/fstab

There's a good explanation on this site. https://www.jetsonhacks.com/2019/04/14/jetson-nano-use-more-memory/

LXDE

Use more memory by changing Ubuntu desktop to LXDE Changing Ubuntu 18.04 desktop to LXDE desktop, you can make extra 1GB memory.

sudo apt remove --purge ubuntu-desktop
sudo apt install lxdm 
sudo apt remove --purge gdm3
sudo apt install lxde
sudo apt install --reinstall lxdm 

After reboot, your Jetson Nano's desktop would be changed.

GUI on boot

To disable GUI on boot, run:

    sudo systemctl set-default multi-user.target

To enable GUI again issue the command:

    sudo systemctl set-default graphical.target

to start Gui session on a system without a current GUI just execute:

    sudo systemctl start gdm3.service

VNC Setup (Method 1)

https://developer.nvidia.com/embedded/learn/tutorials/vnc-setup

Setup VNC server on the Jetson developer kit Enable the VNC server to start each time you log in If you have a Jetson Nano 2GB Developer Kit (running LXDE)

mkdir -p ~/.config/autostart
cp /usr/share/applications/vino-server.desktop ~/.config/autostart/.

For all other Jetson developer kits (running GNOME)

cd /usr/lib/systemd/user/graphical-session.target.wants
sudo ln -s ../vino-server.service ./.

Configure the VNC server

gsettings set org.gnome.Vino prompt-enabled false
gsettings set org.gnome.Vino require-encryption false

Set a password to access the VNC server

# Replace thepassword with your desired password
gsettings set org.gnome.Vino authentication-methods "['vnc']"
gsettings set org.gnome.Vino vnc-password $(echo -n 'thepassword'|base64)

Reboot the system so that the settings take effect

sudo reboot

The VNC server is only available after you have logged in to Jetson locally. If you wish VNC to be available automatically, use the system settings application on your developer kit to enable automatic login.

VNC Setup (Method 2)

By default there are somes problems to enable desktop sharing.

To activate it, it nesessary to do this

  1. Edit the Xml file like : sudo nano /usr/share/glib-2.0/schemas/org.gnome.Vino.gschema.xml
  2. Add this part on the XML file :
<key name='enabled' type='b'>
   <summary>Enable remote access to the desktop</summary>
   <description>
   If true, allows remote access to the desktop via the RFB
   protocol. Users on remote machines may then connect to the
   desktop using a VNC viewer.
   </description>
   <default>false</default>
</key>
  1. Compile the Gnome schema with : sudo glib-compile-schemas /usr/share/glib-2.0/schemas
  2. After this, you can enable desktop sharing.

Jetson enable desktop sharing

Source information : http://bit.ly/2onnLIc

Auto SSH login + no sudo passwd

# https://superuser.com/questions/8077/how-do-i-set-up-ssh-so-i-dont-have-to-type-my-password
# on host:
ssh-keygen -t rsa

# upload the public key to the remote server:
ssh-copy-id -i ~/.ssh/id_rsa.pub remote-user@remote-host

# https://askubuntu.com/questions/147241/execute-sudo-without-password
sudo visudo
username	ALL=(ALL) NOPASSWD:ALL

Docker

https://ngc.nvidia.com/catalog/containers/nvidia:l4t-base

In case of docker error:

# https://github.com/moby/moby/issues/13008
# [1.6.0][graphdriver] prior storage driver "devicemapper" failed: error intializing graphdriver
sudo rm -rf /var/lib/docker
sudo reboot

OpenCV

https://github.com/raspberry-pi-maker/NVIDIA-Jetson/tree/master/useful_scripts https://forums.developer.nvidia.com/t/installing-opencv4-on-xavier-solved/65436 https://github.com/AastaNV/JEP/blob/master/script/install_opencv4.5.0_Jetson.sh

Samba

sudo apt-get install samba
sudo nano /etc/samba/smb.conf

Configure like:

#======================= Share Definitions =======================

# Un-comment the following (and tweak the other settings below to suit)
# to enable the default home directory shares. This will share each
# user's home directory as \\server\username
[homes]
   comment = Home Directories
   browseable = yes

# By default, the home directories are exported read-only. Change the
# next parameter to 'no' if you want to be able to write to them.
   read only = no

# File creation mask is set to 0700 for security reasons. If you want to
# create files with group=rw permissions, set next parameter to 0775.
   create mask = 0755

# Directory creation mask is set to 0700 for security reasons. If you want to
# create dirs. with group=rw permissions, set next parameter to 0775.
   directory mask = 0755

# By default, \\server\username shares can be connected to by anyone
# with access to the samba server.
# Un-comment the following parameter to make sure that only "username"
# can connect to \\server\username
# This might need tweaking when using external authentication schemes
   valid users = %S

Then:

sudo service smbd restart
sudo smbpasswd -a username

Python and PyTorch

Install useful Python packages

sudo apt install python3-pip libfreetype6-dev libffi-dev -y
pip3 install cython
pip3 install numpy jupyter jupyterlab matplotlib pandas

ROS2

https://forums.developer.nvidia.com/t/ros2-on-agx-xavier/121836/9 https://github.com/dusty-nv/jetson-containers/blob/master/Dockerfile.ros.foxy https://github.com/AndreV84/jetson-containers/blob/master/ros2_realsense/Dockerfile

PyTorch

https://github.com/AastaNV/JEP/blob/master/script/install_pyTorch_Xavier.sh

Install PyTorch: Follow the instructions from Nvidia forums

wget https://nvidia.box.com/shared/static/9eptse6jyly1ggt9axbja2yrmj6pbarc.whl -O torch-1.6.0-cp36-cp36m-linux_aarch64.whl
sudo apt-get install python3-pip libopenblas-base libopenmpi-dev 
pip3 install Cython
pip3 install numpy torch-1.4.0-cp36-cp36m-linux_aarch64.whl

Install torchvision

sudo apt-get install libjpeg-dev zlib1g-dev
git clone --branch release/0.7 https://github.com/pytorch/vision torchvision
cd torchvision
sudo python3 setup.py install

Tools

jetson_stats

Command line resource monitor jetson_stats. Works well in SSH.

sudo apt install python-pip python3-pip
sudo -H pip install -U jetson-stats
jtop

Visual Studio Code

https://github.com/JetsonHacksNano/installVSCode

VERSION=latest
wget -N -O vscode-linux-deb.arm64.deb https://update.code.visualstudio.com/$VERSION/linux-deb-arm64/stable
sudo apt install ./vscode-linux-deb.arm64.deb

Arduino IDE

https://github.com/JetsonHacksNano/installArduinoIDE

INSTALL_DIR=${HOME}
# Direct Jetson support starts at 1.8.15
ARDUINO_VERSION=1.8.15

# Only download if newer version exists
wget -N https://downloads.arduino.cc/arduino-$ARDUINO_VERSION-linuxaarch64.tar.xz
tar -C $INSTALL_DIR/ -xvf arduino-${ARDUINO_VERSION}-linuxaarch64.tar.xz
cd $INSTALL_DIR/arduino-${ARDUINO_VERSION}
sudo ./install.sh
./arduino-linux-setup.sh "$USER"
echo "You can delete the tar file if desired: arduino-"${ARDUINO_VERSION}"-linuxaarch64.tar.xz"

Netdata

Web-based resource monitor Netdata

sudo apt install curl
bash <(curl -Ss https://my-netdata.io/kickstart.sh)

open your browser to http://192.168.55.1:19999/ to see the Netdata interface.

To disable Netdata, run

systemctl mask netdata
systemctl stop netdata

To enable it again, use

systemctl unmask netdata
systemctl start netdata

GPIO pins

The GPIO pins can be controlled in Python using jetson-gpio

Install Jetson.GPIO

git clone https://github.com/NVIDIA/jetson-gpio.git
cd jetson-gpio/
sudo python3 setup.py install

Setting User Permissions https://github.com/JetsonHacksNano/ServoKit/blob/master/scripts/setPermissions.sh

sudo groupadd -f -r gpio
sudo usermod -a -G gpio $USER
sudo usermod -aG i2c $USER
sudo cp lib/python/Jetson/GPIO/99-gpio.rules /etc/udev/rules.d/
sudo udevadm control --reload-rules && sudo udevadm trigger

Synergy

https://github.com/symless/synergy-core/wiki/Compiling#linux

sudo apt install xorg-dev libssl-dev qtbase5-dev libgdk-pixbuf2.0-dev libnotify-dev qttools5-dev-tools qttools5-dev
git clone https://github.com/symless/synergy-core.git
cd synergy-core
mkdir build
cd build
cmake ..
make
# https://github.com/symless/synergy-core/wiki/Command-Line
# ./bin/synergyc $SYNERGY_SERVER_IP

RealSense SDK

https://github.com/JetsonHacksNano/installLibrealsense

https://dev.intelrealsense.com/docs/nvidia-jetson-tx2-installation

sudo apt-key adv --keyserver keys.gnupg.net --recv-key F6E65AC044F831AC80A06380C8B3A55A6F3EFCDE || sudo apt-key adv --keyserver hkp://keyserver.ubuntu.com:80 --recv-key F6E65AC044F831AC80A06380C8B3A55A6F3EFCDE

sudo add-apt-repository "deb https://librealsense.intel.com/Debian/apt-repo bionic main" -u

sudo apt-get install librealsense2-utils librealsense2-dev

With librealsense2-dev package installed, you can compile an application with librealsense using g++ -std=c++11 filename.cpp -lrealsense2 or an IDE of your choice. To get started with RealSense using CMake check out librealsense/examples/cmake

Reconnect the RealSense device and run: realsense-viewer to verify the installation

Tech specs

Jetson AGX Xavier module

The module itself contains the SOC and acts as the brains of the operation. Its specifications are shown in the table.

Jetson AGX Xavier module
GPU 512-core Volta GPU with 64 Tensor Cores
CPU 8-core Nvidia Carmel CPU (ARM v8.2 64-bit CPU, 8MB L2 + 4MB L3) (up to 2,2656GHz)
Memory 16GB 256-Bit LPDDR4x | 137GB/s
Storage 32GB eMMC 5.1
DL Accelerator (2x) NVDLA Engines
Vision Accelerator 7-way VLIW Vision Processor
Encoder/Decoder (2x) 4Kp60 | HEVC/(2x) 4Kp60 | 12-Bit Support
Size 105 mm x 105 mm x 65 mm
Deployment Module (Jetson AGX Xavier)

Developer Kit interface board

In-depth specification of the port capabilities, as well as pinouts for the GPIO pins on the Developer kit board can be found in the Carrier Board Specification.

A short summary of the ports is shown in the table.

Jetson AGX Xavier Developer Kit Interface Board
PCIe x16 x8 PCIe Gen4/x8 SLVS-EC
RJ45 Gigabit Ethernet
USB-C 2x USB 3.1, DP (Optional), PD (Optional) Close-System Debug and Flashing Support on 1 Port
Camera Connector (16x) CSI-2 Lanes
M.2 Key M NVMe
M.2 Key E PCIe x1 + USB 2.0 + UART (for Wi-Fi/LTE) / I2S / PCM
40-Pin Header UART + SPI + CAN + I2C + I2S + DMIC + GPIOs
HD Audio Header High-Definition Audio
eSATAp + USB3.0 Type A SATA Through PCIe x1 Bridge (PD + Data for 2.5-inch SATA) + USB 3.0
HDMI Type A HDMI 2.0
uSD/UFS Card Socket SD/UFS

Disk Speed

The disk speed of the AGX Xavier is around 116MB/s write and 293MB/s read, if you get read speed much over that (eg. 360MB/s), then it might be the cache you are reading from.

You can perform a disk speed test using dd by first creating a test file and then copying it to /dev/null

dd if=/dev/zero of=testfile bs=2G count=1 oflag=direct
dd if=testfile of=/dev/null

Performance Comparison

In a small AI training test which was mildly CPU-intensive, the Jetson Xavier took 17 sec per epoch, while a Intel i7-4790K with a Nvidia GTX 960 took around 11 sec per epoch. This might give an idea about the performance of the Jetson for training, however your results may differ depending on how CPU- or GPU-intensive the workload is.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment