This project provides a docker image which supports ubuntu desktop (xfce4, lightweight, fast and low on system resources), so that you can run virtual ubuntu desktop in container, you can access it by using ssh or remote desktop just like a virtual machine.
Hardware GPU accelerated rendering for 3D GUI application is supported in container, it's based on EGL by using VirtualGL, and doesn't require /tmp/.X11-unix. if you needn't hardware GPU accelerated rendering, you can also run this container on headless host without GPU (for exmaple, Cloud Server), remote desktop and 3d GUI based on software rendering (high cpu usgae) is also supported.
Tip: Hardware GPU accelerated rendering is only verified on
ubuntu desktop systemhost withmonitor. i'm not sure if hardware GPU accelerated rendering can work for headlessubuntu server systemhost.
- Remote access by ssh and nomachine (remote desktop).
- OpenGL rendering based on software rasterizer (LLVMpipe) with high CPU usgae. (default)
- OpenGL rendering based on Nvidia GPU hardware-accelerated.
- Pre-installed Firefox web browser.
- pre-installed CUDA toolkit, which is useful for deep learning, such as pytorch, tensorflow.
Tip: it's useful to share public computer resources in labs, you can run a independent computer environment like a virtual machine, but more lightweight, and easier to deploy.
- fast to deploy multiple independent developing environment on a single computer.
- easy to share files with host or another container.
- easy to transfer environment to another new computer (save and load image).
xfce4 desktop:
Supported Tags (you can find here Github Tag)：
- Tags of base image：
18.04,20.04,22.04 - Tags of image with cuda (based on
nvidia/cuda)：18.04-cu11.0.3,20.04-cu11.0.3etc. - naming rules is
{UBUNTU VERSION}-cu{CUDA VERSION}, you can find supported{CUDA VERSION}in Docker Image <nvidia/cuda>
Supported {CUDA VERSION}:
- Ubuntu18.04：
11.0.3,11.1.1,11.2.2- Ubuntu20.04：
11.0.3,11.1.0,11.2.2,11.3.1,11.4.3,11.5.2,11.6.2,11.7.1- Ubuntu22.04：
11.7.1,11.8.0,12.0.1,12.1.1
- install
nvidia driver - install
dockerandnvidia-container-runtime.
Tip: the newer cuda version isn't supported if you use older nvidia driver.
pull docker image
docker pull gezp/ubuntu-desktop:20.04-cu11.0.3
# use aliyuncs mirror for chinese users (国内用户可使用阿里云仓库)
# docker pull registry.cn-hongkong.aliyuncs.com/gezp/ubuntu-desktop:20.04-cu11.0.3create conatiner
# create conatiner
docker run -d --restart=on-failure \
--name my_workspace \
--cap-add=SYS_PTRACE \
--gpus all \
--shm-size=1024m \
-p 10022:22 \
-p 14000:4000 \
gezp/ubuntu-desktop:20.04-cu11.0.3- the default username and password are both ubuntu.
access conatiner by ssh
ssh ubuntu@host-ip -p 10022- it's recommended to use vscode + remote ssh plugin
access conatiner by nomachine client (remote desktop)
- download and install nomachine software.
- the ip is host's ip, the port is 14000.
configure USER, PASSWORD, GID, UID when you create conatiner，for example：
docker run -d --restart=on-failure \
--name my_workspace \
--cap-add=SYS_PTRACE \
--gpus all \
-e USER=cat \
-e PASSWORD=cat123 \
-e GID=1001 \
-e UID=1001 \
--shm-size=1024m \
-p 10022:22 \
-p 14000:4000 \
gezp/ubuntu-desktop:20.04-cu11.0.3vglrun glxinfo | grep -i "opengl"- hardware-accelerated is enable successfully if it's output contain
NVIDIA Product Series.
you need add prefix vglrun for command when you run 3D software, for example vglrun gazebo.
# vulkan info
vulkaninfo | grep -i "GPU"
# vulkan demo
vkcube- it's output should contain
NVIDIA Product Seriesif vulkan works well.
add shell in .bashrc to update environment variable
export CUDA_HOME=/usr/local/cuda
export PATH=/usr/local/cuda/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH- for detailed usage, you can refer to
nvidia/cudaDocker Image.
for example
git clone https://github.com/gezp/docker-ubuntu-desktop.git
cd docker-ubuntu-desktop
# for 20.04
./docker_build.sh 20.04
# for 20.04-cu11.0 (based on nvidia/cuda)
./docker_build.sh 20.04-cu11.0thanks to the authors of following related projects:
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent