The various tags of the adamrehn/ue4-runtime image provide minimal, pre-configured environments for running packaged Unreal Engine projects with full GPU acceleration via the NVIDIA Container Toolkit. (For more details on the NVIDIA Container Toolkit, see the NVIDIA Container Toolkit primer on the Unreal Containers community hub.) Note that these images will work with packaged Linux builds from any source, not just builds packaged using the container images from the ue4-docker project.

Both OpenGL+Vulkan and OpenGL+Vulkan+CUDA variants are provided, along with preconfigured images for common GPU-accelerated frameworks such as TensorFlow. Each image variant is also available in a configuration with VirtualGL bundled for displaying the output of OpenGL applications using the host system's display. See the section Using the VirtualGL images for usage details.

For details on using these images to perform cloud rendering via NVIDIA Docker, see the Cloud rendering guide on the Unreal Containers community hub. There are also example Dockerfiles available that demonstrate various uses of Unreal Engine containers, including multi-stage build workflows that encapsulate packaged projects in variants of the ue4-runtime image.

The following tags are provided as convenient aliases for the fully-qualified tags of common image variants:

• adamrehn/ue4-runtime:latest is an alias for adamrehn/ue4-runtime:18.04-vulkan
• adamrehn/ue4-runtime:16.04-opengl is an alias for adamrehn/ue4-runtime:16.04-vulkan
• adamrehn/ue4-runtime:18.04-opengl is an alias for adamrehn/ue4-runtime:18.04-vulkan
• adamrehn/ue4-runtime:tensorflow is an alias for adamrehn/ue4-runtime:18.04-tensorflow-1.13.1
• adamrehn/ue4-runtime:virtualgl is an alias for adamrehn/ue4-runtime:18.04-vulkan-virtualgl
• adamrehn/ue4-runtime:tensorflow-virtualgl is an alias for adamrehn/ue4-runtime:18.04-tensorflow-1.13.1-virtualgl

• adamrehn/ue4-runtime:18.04-vulkan: Ubuntu 18.04 + OpenGL + Vulkan
• adamrehn/ue4-runtime:18.04-cudagl9.2: Ubuntu 18.04 + OpenGL + Vulkan + CUDA 9.2
• adamrehn/ue4-runtime:18.04-cudagl10.0: Ubuntu 18.04 + OpenGL + Vulkan + CUDA 10.0
• adamrehn/ue4-runtime:18.04-cudagl10.1: Ubuntu 18.04 + OpenGL + Vulkan + CUDA 10.1
• adamrehn/ue4-runtime:18.04-cudagl10.2: Ubuntu 18.04 + OpenGL + Vulkan + CUDA 10.2
• adamrehn/ue4-runtime:18.04-tensorflow-1.13.1: Ubuntu 18.04 + OpenGL + Vulkan + CUDA 10.0 + TensorFlow 1.13.1
• adamrehn/ue4-runtime:18.04-vulkan-virtualgl: Ubuntu 18.04 + OpenGL + Vulkan + VirtualGL
• adamrehn/ue4-runtime:18.04-cudagl9.2-virtualgl: Ubuntu 18.04 + OpenGL + Vulkan + CUDA 9.2 + VirtualGL
• adamrehn/ue4-runtime:18.04-cudagl10.0-virtualgl: Ubuntu 18.04 + OpenGL + Vulkan + CUDA 10.0 + VirtualGL
• adamrehn/ue4-runtime:18.04-cudagl10.1-virtualgl: Ubuntu 18.04 + OpenGL + Vulkan + CUDA 10.1 + VirtualGL
• adamrehn/ue4-runtime:18.04-cudagl10.2-virtualgl: Ubuntu 18.04 + OpenGL + Vulkan + CUDA 10.2 + VirtualGL
• adamrehn/ue4-runtime:18.04-tensorflow-1.13.1-virtualgl: Ubuntu 18.04 + OpenGL + Vulkan + CUDA 10.0 + TensorFlow 1.13.1 + VirtualGL

• adamrehn/ue4-runtime:16.04-vulkan: Ubuntu 16.04 + OpenGL + Vulkan
• adamrehn/ue4-runtime:16.04-cudagl9.0: Ubuntu 16.04 + OpenGL + Vulkan + CUDA 9.0
• adamrehn/ue4-runtime:16.04-cudagl9.1: Ubuntu 16.04 + OpenGL + Vulkan + CUDA 9.1
• adamrehn/ue4-runtime:16.04-cudagl9.2: Ubuntu 16.04 + OpenGL + Vulkan + CUDA 9.2
• adamrehn/ue4-runtime:16.04-cudagl10.0: Ubuntu 16.04 + OpenGL + Vulkan + CUDA 10.0
• adamrehn/ue4-runtime:16.04-cudagl10.1: Ubuntu 16.04 + OpenGL + Vulkan + CUDA 10.1
• adamrehn/ue4-runtime:16.04-cudagl10.2: Ubuntu 16.04 + OpenGL + Vulkan + CUDA 10.2
• adamrehn/ue4-runtime:16.04-tensorflow-1.13.1: Ubuntu 16.04 + OpenGL + Vulkan + CUDA 10.0 + TensorFlow 1.13.1
• adamrehn/ue4-runtime:16.04-vulkan-virtualgl: Ubuntu 16.04 + OpenGL + Vulkan + VirtualGL
• adamrehn/ue4-runtime:16.04-cudagl9.0-virtualgl: Ubuntu 16.04 + OpenGL + Vulkan + CUDA 9.0 + VirtualGL
• adamrehn/ue4-runtime:16.04-cudagl9.1-virtualgl: Ubuntu 16.04 + OpenGL + Vulkan + CUDA 9.1 + VirtualGL
• adamrehn/ue4-runtime:16.04-cudagl9.2-virtualgl: Ubuntu 16.04 + OpenGL + Vulkan + CUDA 9.2 + VirtualGL
• adamrehn/ue4-runtime:16.04-cudagl10.0-virtualgl: Ubuntu 16.04 + OpenGL + Vulkan + CUDA 10.0 + VirtualGL
• adamrehn/ue4-runtime:16.04-cudagl10.1-virtualgl: Ubuntu 16.04 + OpenGL + Vulkan + CUDA 10.1 + VirtualGL
• adamrehn/ue4-runtime:16.04-cudagl10.2-virtualgl: Ubuntu 16.04 + OpenGL + Vulkan + CUDA 10.2 + VirtualGL
• adamrehn/ue4-runtime:16.04-tensorflow-1.13.1-virtualgl: Ubuntu 16.04 + OpenGL + Vulkan + CUDA 10.0 + TensorFlow 1.13.1 + VirtualGL

Offscreen rendering with Vulkan requires projects built with Unreal Engine 4.25.0 or newer. To render offscreen, specify the -RenderOffscreen flag when running your packaged Unreal project.

Vulkan rendering under Unreal Engine 4.24 or older will require bind-mounting the X11 socket from the host system and propagating the DISPLAY environment variable so that output can be rendered to a window. See the section below for details on the required docker run flags.

The virtualgl configuration of each image variant adds the following components:

• The X11 libraries needed for running applications that create X11 windows
• VirtualGL itself, which provides the vglrun command for interposing OpenGL function calls

To run a container using a VirtualGL-enabled image, the Docker host system will need to be running an X11 server and you will need to bind-mount the host's X11 socket inside the container like so:

# Replace "adamrehn/ue4-runtime:virtualgl" with your chosen image tag
docker run --gpus=all -v/tmp/.X11-unix:/tmp/.X11-unix:rw -e DISPLAY adamrehn/ue4-runtime:virtualgl bash

The manner in which you need to invoke UE4 projects inside the container depends on your use case:

• If you are running the container locally on a machine with an OpenGL-enabled X11 configuration (e.g. a standard desktop installation of Ubuntu 18.04) then the GLVND dispatch library provided by the NVIDIA base images will handle the relevant OpenGL function calls without the need to use VirtualGL. Running UE4 projects via vglrun in this scenario will actually reduce performance due to the additional interposition overheads, so be sure to run projects directly. (e.g. ./MyProject.sh)

• If you are running the container on a remote host and are using X11 forwarding to display the window on your local machine then you will need to run UE4 projects via vglrun in order to ensure OpenGL functionality will work from within an SSH session. (e.g. vglrun ./MyProject.sh)

Building the container images from source requires Python 3.5 or newer and the dependency packages listed in requirements.txt.

To build the images, simply run build.py. This will automatically query Docker Hub to retrieve the list of available nvidia/cudagl base images based on Ubuntu LTS releases and build all variants of the adamrehn/ue4-runtime image accordingly.

Copyright © 2019 - 2020, Adam Rehn. Licensed under the MIT License, see the file LICENSE for details.

Initial development of the TensorFlow 1.13.1 image variant was funded by Deepdrive, Inc.