It is about adding zypper service that refers to NVIDIA repositories.
Ok, various possibilities, but which is best?
@suse_rasputin well the rpms from the repository are signed, so easier with secure boot enabledâŚ
no secure boot here. I only want to use the card for graphics and computing (CUDA). What is the least trouble during install and daily maintenance?
I donât earn my money for playing arround with hard-/software⌠
@suse_rasputin use the rpm and Nvidia repo thenâŚ
PS I donât earn money helping out here eitherâŚ
sorry, you got me wrong. I wanted to point out that Iâm not looking for the most advanced solution, but for the most fool-proof install.
Is it correct that I need to install new drivers manually befor kernel updates if I install âthe hardwayâ?
@suse_rasputin no, you can download/install/update a new Nvidia driver while the system is running, it will then be available and in use on next reboot.
For a new kernel, yes a manual re-install is required of the driver.
For example I use;
systemctl set-default multi-user.target
zypper -vvv dup
systemctl reboot
./NVIDIA-Linux-x86_64-550.54.14.run --no-systemd --ui=none -aq
systemctl set-default graphical.target
systemctl reboot
1 Like
@malcolmlewis Did you test if the way via dkms works?
As dkms is available for Leap and Tumbleweed this may reduce the effort when a minor kernel version update comes.
@hui nope, prefer the finer control
So this is your typical update procedure for TW with G06 âthe hardwayâ, correct?
You do it from a tty with GUI logged out, correct? Or via ssh?
The âsystemctl set-default multi-user.targetâ getâs only applied on reboot AFTER zypper dup, correct?
Then you download the latest and greatest from here:
in this case the 54.14 version of the 550 driver. Why not the 161.07 version of the 535 driver published the day before?
I really want to understand what I am doing/missing here, before starting 
Really enjoy the support, learning curve is steep, as my NVIDIA experience is limited to old stuff (the NVS 310 working flawlessly for years⌠thanks to mrmazda).
This are the explanations for Malcolms terminal commands. And you need to do it in this order. The Nvidia run file needs to be downloaded prior installation and switching to text modeâŚ
systemctl set-default multi-user.target
--> this will change boot target to text mode after the next reboot
zypper -vvv dup
--> this applies upgrades to your machine
systemctl reboot
--> this reboots your machine into text mode
./NVIDIA-Linux-x86_64-550.54.14.run --no-systemd --ui=none -aq
--> this will install the Nvidia driver
systemctl set-default graphical.target
--> this will change boot target to the GUI mode after the next reboot
systemctl reboot
--> this reboots your machine into GUI mode
1 Like
@suse_rasputin yes, after the update reboot to runlevel 3/multi-user (as set) and a VT (tty) login as root user.
I go to https://download.nvidia.com/XFree86/Linux-x86_64/ for the download.
Where do you see for which kernel the .run is adequate? How long do you (have to) wait after a new kernel released befor you dup and start the .run subsequently?
All kernels higher kernel 3.10 are supported by the actual Nvidia drivers. See Nvidia driver documentation which comes with every driverâŚ
https://download.nvidia.com/XFree86/Linux-x86_64/550.54.14/README/minimumrequirements.html
@suse_rasputin I do it after a new kernel or driver is released, no waitingâŚ
And the âhardwayâ for CUDA is how?
@suse_rasputin you can get the latest from here (includes the driver) https://developer.nvidia.com/cuda-toolkit-archive
1 Like
Hi malcolm, many thanks! At the link I get a version for opensuse 15 (.run I would prefer, or?), will work on TW as well? If I install CUDA from repo I have to install a LEAP repo, iirc.
How often is a new CUDA install needed? It should work fine for several driver, or? When is an update required?
@suse_rasputin yes use the run file⌠I would just install the 550.54.14 driver and see if that works for you first as it has the cuda lib, unless you need nvcc etc?
What is nvcc? I want to train some âKIâ (neural networks, some kind of local âLLMâ), what is needed besides CUDA?