feat: add support for nvidia gpu access

[MondoGao]

Add a new environment to enable Nvidia gpu access when creating containers. It's similar to /dev/dri support. See #1525.

Note: I didn't fully inspect and test this change since I don't have php development environment set up in my machine. Feel free to inherit this pr to enhance its maintainability.

[szaimen]

Hi @MondoGao thanks for your PR! Can you please fix DCO? Wee need that in order to be able to merge it later on. See https://github.com/nextcloud/all-in-one/pull/5132/checks?check_run_id=28844781358

[MondoGao]

Hi @MondoGao thanks for your PR! Can you please fix DCO? Wee need that in order to be able to merge it later on. See https://github.com/nextcloud/all-in-one/pull/5132/checks?check_run_id=28844781358

Sign added :)

[szaimen]

Hi, I had a fast look at this and I think it would add the capability to all containers that are controlled by AIO. Better would probably be to add this as a capability to containers-schema.json and add this only to certain containers via containers.json

Also this is missing some places, e.g. adding documentation on it in the readme. See #1659 as inspiration.

Additionally, do you know if this is also going to work with AMD and Intel GPUs? In best case we create only one setting that works for all of them.

[MondoGao]

Hi, I had a fast look at this and I think it would add the capability to all containers that are controlled by AIO. Better would probably be to add this as a capability to containers-schema.json and add this only to certain containers via containers.json

Also this is missing some places, e.g. adding documentation on it in the readme. See #1659 as inspiration.

Additionally, do you know if this is also going to work with AMD and Intel GPUs? In best case we create only one setting that works for all of them.

I believe docker runtime only supports Nvidia GPU passthrough. Intel & AMD CPU/GPU's hardware acceleration is exposed through /dev/dri.

Looks like I have to install php dev env to polish this pr, please expect a late response.

[luzfcb]

@MondoGao Thank you for starting this effort.

@MondoGao @szaimen

Besides supporting Nvidia GPU passthrough, there are cases where the container may require the same specific version of the NVIDIA driver installed on the host system. It might be a good idea to create an environment variable for the Nvidia driver version and let the container handle the Nvidia driver setup and configuration logic inside the container.

That is, Nextcloud AIO could either provide some way for the user to manually specify the version or, if it is not provided by the user, then have an automatic way to obtain this information and fill an NVIDIA_DRIVER_VERSION environment variable that will be available to all containers managed by Nextcloud AIO.

I have exactly this with docker-steam-headless, and it works great.

This is the driver download and installation script: https://github.com/Steam-Headless/docker-steam-headless/blob/860451da74b397385f1b1658545d2bb891aa8e46/overlay/etc/cont-init.d/60-configure_gpu_driver.sh

This script creates the X Server configuration files and other related configurations. It probably does not apply to Nextcloud and plugins since they do not use X Server, but I am including it here just for reference.

https://github.com/Steam-Headless/docker-steam-headless/blob/860451da74b397385f1b1658545d2bb891aa8e46/overlay/etc/cont-init.d/70-configure_xorg.sh

[docjyJ]

It would be nice to have this feature!

We can have Nextcloud Assistant servers with high performance easily!!!!!!

I agree with @szaimen the configuration should be done in the containers.json schema.

Otherwise it looks good to me.

AMD GPU : https://rocm.docs.amd.com/projects/install-on-linux/en/latest/how-to/docker.html