Lambda Stack and Nvidia Data Science Workbench

Technical Help
1

Can Lambda Stack and Nvidia Data Science Workbench co-exist or are they going to trip all over each other?

– Rich

2

NVIDIA Data Science Workbench is writes all over the system replacing packages.
The way it is written it is not directly compatible with any normal install, and can break usage not just for the user installing but for other users on system, also it does not isolate to a virtual environment, so it may break other packages you are working on.

However, if you are going to move to use ‘NVIDIA Data Science Workbench’. I think I can make it work with Lambda Stack by adding other repositories. (nvidia tends to use their own version of the deprecated nvidia-docker2 versus docker.io). I will need to reboot and again re-run the installer script as it just finished the first stage of the install. It looks like it uses spyder which brings in other issues, but some like spyder.

I am installing right now. However, it is a dangerous install.

I was going to install this in the python venv environment with system-defaults.
(So it does not mess up other applications usage). But it changes system wide software.

  • This package installs deb packages - so it requires a user to have root
  • It installs packages from git repositories
  • And installs packages from snap repositories
  • It sets up system wide configurations

Also currently the install appears to be hung, and finally continued after about 10 minutes.
Then requires you to reboot, and re-run the installer tool.

I will come back after the reboot

1 Like
3

I would recommend setting up a separate root so you can back to a healthy image.

I ended up cleaning up a lot after installing to get it back to normal. And I will likely
just reinstall. (But note I have 3 boot devices and shared space).

I will setup a clean Ubuntu root and test install Nvidia’s workbench above
and works out of the box.

It may be easier to simply just install docker the normal way and use NGC directly.
NVIDIA NGC Tutorial: Run a PyTorch Docker Container using nvidia-container-toolkit on Ubuntu

Also it is a much cleaner and standard solution from desktop to HPC machines.

1 Like