SDKManager "Not supported on Linux"

@sgroesz you can’t even use a VM! It must be a ‘real’ computer. I had to dig out an old ThinkPad to install Ubuntu 18.04 onto it just so I could flash the Jetson.

@moshammer.a At least with the GUI you don’t have to know what options can be given for cli options like:

  • --target
  • --version
  • --select

The example given use--target P3668-0000, but I have no idea what that model number means, so I end up googling it. And then I find an unrelated NVIDIA page that happens to list the model numbers. This is the type of problem I’m talking about - nothing is made easy. I have to flick to a browser, open Google and search for “jetson P3448 xavier agx” hoping that it gives me something. Or maybe NVIDIA expects their customers to read every page of every doc and website before they begin using their products?

I agree.

The sdkmanager is already painful, and we previously had to jump through hoops to get it to work by faking lsb_release etc. Now that doesn’t work. Just the sheer amount of time wasted with all these strange tools.

Ugh… two hours wasted today already, now going down the docker route. This sucks tbh

2 Likes

I built a docker container to run SDK Manager GUI on newer Linux distributions. GitHub - raffraffraff/nvidia-sdkmanager: Run the NVIDIA SDK Manager gui from Docker

It volume mounts several directories in your $HOME, so it persists downloads and configs between runs. However, it does not contain any NVIDIA software - if you leave “Host Machine” checked and allow SDK Manager to install CUDA to the host, these changes will be lost every time you exit. However, you can run a docker commit when the container is running, and this should preserve the changes to a new image.