Is there any way to use "gpus all" when launching container from Docker Desktop UI?

[progmars]

I’m using WSL2 and I followed the GPU support | Docker Docs article about enabling GPU (by the way, it had two mistakes in command examples, which I reported in GitHub). The test commands passed as expected, CUDA is working.

Then I downloaded unsloth/unsloth image using Docker Desktop and launched a container. I tried to do that multiple times, but the Python scripts inside the container kept reporting that CUDA is not available. Only when I run the container from a command line with --gpus all flag, then it works.
So, it is not clear what flags exactly are being passed when running a container from the UI and why there is no option to enable the GPU flag.

I understand that experts use command line anyway. However, this might cause confusion for novice users. I already found a topic here that was closed seemingly because the responder did not understand what’s there more to enable if --gpus all command line works. Yeah, but the UI does not. Or are there any additional reasons or settings why running a container from Desktop UI does not give it GPU access automatically?

[meyay (Metin Y.)]

The Docker Desktop ui does allow configuring exactly 4 configuration options for a container: the image, port mappings (if the image has at least on EXPOSE instruction), volume mappings, and environments.

The other 100+ configuration options, --gpu being one of them, are not supported from the UI.

You can try opening a feature request in the docker roadmap repo and ask for the feature.

@progmars

I already found a topic here that was closed seemingly because the responder did not understand what’s there more to enable if --gpus all command line works.

It is more likely the topic was auto closed after a month due to the absence of new posts in the topic.

Just to be sure: this is a community forum, where community users help other community users. Everyone is trying to answer questions to the best of their knowledge.