Installing CUDA as Part of a Automated Build Process

I am trying to setup an automated build process for the root fs. It’s based on the samplefs build process provided with L4T. The build process goes just fine, however I am unable to get CUDA to install.

Here’s what I have done:

  • Modified the nvubuntu-bionic-aarch64-packages file to install Ubuntu packages I need
  • Copied my arm64 deb files (including cudo-repo-l4t-20-2-local_10.2.460-1_arm64.deb) to /tools
  • Run the nv_build_samplefs.sh script

Everything builds fine until the CUDA deb is installed at the end of the script. I get a message telling me the CUDA GPG key is not installed and CUDA can’t be installed. How do I get around this for an automated build? Is a complete automated rootfs build possible?

Any help is appreciated.

Are you using JetPack to install CUDA?