Hello,
I am using xavier nx production module (emmc).
Is there a way to install Jetson SDK components (CUDA, CUDA-X AI, Computer Vision, NVIDIA Container Runtime, Multimedia) without using the SDK manager?
Thank you.
Hello,
I am using xavier nx production module (emmc).
Is there a way to install Jetson SDK components (CUDA, CUDA-X AI, Computer Vision, NVIDIA Container Runtime, Multimedia) without using the SDK manager?
Thank you.
If youâve flashed and the system is up and running, is the repo installed?
dpkg -l | grep '\-repo\-'
If the repo is installed, then you have access to those packages via apt-get
. To see a list of interesting packages from another Jetson (which can be a guide to knowing what to install on the eMMC version):
dpkg -l | egrep -i '(cuda|nvidia|l4t)'
Hello,
If there is no output when executing the following command, the repo is not installed, right?
How do I install it from the repo?
Thank you.
You probably have to escape the hyphens (I was surprised to find that I did when using the âabsoluteâ single quote), but keep in mind everything not installed does not necessarily mean the repo is missingâŠthe repo itself is not one of those items on the list, but is a dependency if they are added. I just donât know if the repo is added by SDKM the first time one of those extras are installed, or if the repo is there after a basic flash.
If you go here on your Jetson:
/etc/apt/sources.list.d
âŠdo you have file ânvidia-l4t-apt-source.list
â? If so, then the repo is installed.
If you go to your host PC the downloads which it uses during ssh are stored at:
~/Downloads/nvidia/sdkm_downloads/
Keep in mind that JetPack/SDKM works with a lot of different types of Jetsons, and also works with the host PC, so not necessarily every ârepo
â file in that location is useful for you. On the Jetson only the âarm64
â and âall
â packages matter:
ls ~/Downloads/nvidia/sdkm_downloads/*repo* | egrep -v 'amd64'
Among those, the cross tools are for the PC, so you can eliminate those:
ls ~/Downloads/nvidia/sdkm_downloads/*repo* | egrep -v '(amd64|cross)'
Youâll also find some are for CUDA, and others for libvisionworksâŠit doesnât hurt to install both repos, and if you are doing something with libvisionworks, then youâd need both. Then there are the older ones using CUDA 10.0, and more current ones using CUDA 10.2âŠyouâd only want the version which goes with your JetPack release. Once you have a repo file on the host PC you can copy it to the Jetson any way you want (e.g., SD card, email, scp
, so on), and then use dpkg
to install it. An example:
sudo dpkg -i ./cuda-repo-l4t-10-2-local-10.2.89_1.0-1_arm64.deb
sudo apt update
Do you have the repo files on the host PC? You can tell SDKM to download only, so if the files are not currently there, then you could still create the â~/Downloads/nvidia/sdkm_downloads/
â content.
Hello,
If you do not use the host pc
Is there any way to install the sdk component on the target machine alone?
Thank you.
Hello,
Here, can I move the deb file corresponding to the sdk component to the target machine and install it with dpkg?
Thank you.
Hello,
In this case, can I install nvidia-cuda?
Thank you.
You can download to a host PC this wayâŠthen use any manual copy method to copy to the Jetson. This is designed to give you access to the â.deb
â files related to CUDA, but it would still be up to you to get them to the Jetson. I donât know of another way to get those â.deb
â files in recent releases.
The â.deb
â files you show which are arm64
or all
and not âcrossâ would be the correct files. These can be copied to the Jetson and installed via dpkg
. Note that order generally matters, but often you can name several .deb
âs on the same command line and avoid some install order issues (basically that is what makes apt
so niceâŠit can figure out dependencies).
I donât know about jtop
, but there might be more packages required. Keep in mind that if youâve installed the repo, then sudo apt update
and sudo apt-get
will work to install what is missing if you know the package which is missing.