I’d like to use chroot to install cuda manually on my host side
But I have no idea how to do
Is there any instruction to done this job?
I use Fedora and install manually. Basically it comes down to blacklisting the Nouveau driver after installing the NVIDIA video driver. Once that is done you can install the cuda-repo package, which in turn makes it possible to use your native package mechanism to install the actual CUDA libraries (and dev files if building CUDA apps). All of those details change depending on Linux distribution. Make sure you are set up to be able to boot to text mode before you start the process since you can’t be running graphical for some parts of the install of NVIDIA driver. Browse here:
So far as chroot goes I’m not sure how that would work.
Want to confirm first:
Do you want to install CUDA on Jetson or an x86-based machine?
I want install cuda on sample rootfilesystem
and use ./reflash jetson-tx2 mmcblk0p1 on my target board
In that case you would manually install to a running Jetson, not from the PC host into the rootfs. What you can do is that after you’ve installed to a Jetson to then clone the Jetson and use the clone for future flashes. The clone will be an exact copy.