xorg-docker

Completely Dockerized instance of X server with support for HW acceleration from nVidia GPUs, OpenGL and Vulkan.

How to build

You need cpk to build this project. Install it from pip3 install cpk.

cpk build

How to run

Pick the GPU to use

You need to specify which GPU to use inside the container. You can do so by passing the variable BUS_ID. Use the following command to find a list of GPUs on your system and take note of the PCI BusID of the GPU you want to use.

nvidia-xconfig --query-gpu-info

A good output looks like the following,

GPU #1:
  Name      : GeForce GTX 1080 Ti
  UUID      : GPU-2e02c872-4cad-80d7-5128-780d01085bcb
  PCI BusID : PCI:129:0:0

In this case, the value for BUS_ID is PCI:129:0:0.

Mounting graphics drivers

This repository supports only nvidia GPUs at this time. You need the package nvidia-docker2 installed for this to work.

The nvidia runtime for Docker mounts the GPU drivers from the host so that you don't need them installed inside the container. Though this runtime is designed for CUDA applications, so it doesn't mount graphics libraries, we will need to do that manually.

Run with OpenGL

cpk run -L opengl -- -e BUS_ID=PCI:129:0:0 --runtime=nvidia -v /usr/lib/x86_64-linux-gnu/nvidia/xorg:/usr/lib/xorg/modules/drivers

Run with Vulkan

cpk run -L vulkan -- -e BUS_ID=PCI:129:0:0 --runtime=nvidia -v /usr/lib/x86_64-linux-gnu/nvidia/xorg:/usr/lib/xorg/modules/drivers -v /usr/share/vulkan:/etc/vulkan

Debug

If you want to find out whether a process successfully landed on the GPU and is not being run using CPU rendering instead, use the command,

cat /sys/kernel/debug/dri/BUS_DEVICE_ID/clients

where BUS_DEVICE_ID is the number that appears in the BUS_ID variable, in the case shown above BUS_DEVICE_ID=129.

Troubleshooting

(EE) no screens found

When the X server inside the container dies with this message, it is very likely that it wasn't the display that it wasn't found but your GPU drivers. In particular, make sure that you have this file on your host system /usr/lib/x86_64-linux-gnu/nvidia/xorg/nvidia_drv.so. If this file is not there, you either have installed the server version of the nvidia drivers (which do not carry graphics drivers), or it is somewhere else, if you find it, maybe with locate nvidia_drv.so, you can update the cpk command above accordingly.

afdaniele/xorg-docker