I have just installed the SDK for a Jetson Xavier, and was warned it would require 23GB of disk space, which by doing a df -h before and after, confirms this amount was spot on. The install all went fine and the Xavier is up and running. Now I want to delete all of the files on the host machine and can’t find all the space again. There were 2 nvidia folders in the home directory and downloads, totalling approximately 12GB and 4GB respectively, which leaves about 9GB left on my machine, but I have no idea where to find (and remove) it. Re-loading up sdkmanager it only says files will be installed in /home/user/Downloads/nvidia/sdkm_downloads and /home/user/nvidia/nvidia_sdk. I found it also installed CUDA-10 under /usr/local/cuda-10.0 which is 2.9GB. There is about 6GB that has been used up in the install which I can’t account for and free. Any idea where this would be (Ubuntu 16.04)?
Do you have “~/Downloads/nvidia/”?
Yes, which accounts for about 4GB, (but this was mentioned in the original post).
Under SDK Manager, as far as I know content will be in one of:
- /usr/local
- ~/nvidia
- ~/Downloads/nvidia
- Any local repository from "/var/", e.g., CUDA repos.
Note that if something is in “/var/”, then “apt” can access this. If you were to remove “/var” local respositories, then you’d also want to remove the ability to search for those locations by apt. This might be something listed in “/etc/apt/sources.list” or other apt optional configurations.
There is about 2GB of CUDA-10 things which were installed within /var/. That still leaves a just a few GB (c4GB) somewhere else.
You could easily figure this out on your own by running some snapshottable file system service on your host (or in a VM or whatever.)
Set up your host file system on top of LVM, or use NILFS2 for your file system.
Start with a clean install.
Snapshot the install.
Run the jetpack installer/app.
Re-mount the snapshot as a second read-only file system.
Compare the new state of the file system with the snapshotted state of the file system.
If you have a host system that has lots of customizations that are not easy to re-create, and you don’t have some spare partition space or a spare disk or a spare computer available to run this experiment on, then that’s of course a little harder to do, but in general, these kind of system compares are very easy when you use LVM (and/or good backups :-)
Thanks for the suggestion. Unfortunately I don’t have a spare machine lying around with enough disk-space and the device is already now setup, so repeating is not feasible in this instance. Nonetheless, I see the that the proposed method would do the job.
There would be some cross tools installed in standard directories if samples were installed, e.g., cross compiler, cross linker, perhaps a sysroot. I don’t know where they are, but one way to examine is to type “aarch” on the command line, and then hit tab to see what options there are. As an example you should find “aarch64-linux-gnu-”, and then more tabs would list a series of tools. Then you could for example run “which aarch64-linux-gnu-gcc”. Following that (I’m using Fedora so it will differ somewhat in actual location, but “which” will abstract that out of the question):
sudo dpkg -S `which aarch64-linux-gnu-gcc`
Or in general:
sudo dpkg -l | grep aarch64