Packaged Project does not start properly when run inside official runtime container

Distribution
, ,
1

Hi everybody,I want to run a packaged project inside an official runtime container from Epic. Yet I run into the problem that it starts with the following output to the shell:

5.5.4-37670630+UE5 1013 0
Disabling core dumps.

Here are the steps I go through before running the project inside the container:

  1. Build the project inside the official unreal container
  2. Copied the build outside via a shared volume
  3. Removed the chmod +x *-Linux-Shipping line from the start up script
  4. Containerize the project based on the ghcr.io/epicgames/unreal-engine:runtime
  5. Start the project inside the container via ./Project.sh -RenderOffscreen
  6. Receive the above output

Note:

  • I removed the chmod as Docker already sets the permission that every user can execute the file and the chmod itself caused permission issues inside the container as it would require root access
  • The build itself with all its modifications was tested on a native Linux machine as well and worked without any problems
  • When I change the -RenderOffscreen to -nullrhi the project does not exit immediately but the rendering does obviously not work
2

Hey, I ran into the exact same issue and was able to fix it — this is NOT a chmod or startup script problem.

The root cause is that the container is not using the NVIDIA runtime, so Vulkan cannot access the GPU.

That’s why you see:
“Disabling core dumps.”
with no further logs — Unreal fails during RHI (rendering) initialization.

The key clue is that -nullrhi works (no rendering), but -RenderOffscreen fails.

Fix:

  1. Configure Docker to use NVIDIA runtime:

Edit /etc/docker/daemon.json:

{
“default-runtime”: “nvidia”,
“runtimes”: {
“nvidia”: {
“path”: “nvidia-container-runtime”,
“runtimeArgs”: 

}
}
}

Then restart Docker:

sudo systemctl restart docker

  1. Verify:

docker info | grep -i runtime

You should see:

Default Runtime: nvidia

  1. Run container again (with --gpus all)

After this, inside the container you should see:

/usr/share/vulkan/icd.d/nvidia_icd.json

Before fixing this, only Mesa / lavapipe (CPU Vulkan) was available, which causes Unreal to exit immediately.

Once NVIDIA ICD is available, Unreal starts correctly and Pixel Streaming works.

Important:
–gpus all alone is NOT enough — Docker must use the NVIDIA runtime to expose Vulkan properly.