Nvidia Update. Which one?

Just installed tumbleweed and this evening with the new snapshot this is what I have with zypper dup:

Problem: 1: The installed nvidia-common-G07-595.84-8.1.x86_64 package requires ‘(nvidia-open-driver-G07-signed-kmp = 595.84 if nvidia-open-driver-G07-signed-kmp-meta)’, but this requirement cannot be met
Removed providers: nvidia-open-driver-G07-signed-kmp-default-595.84_k7.1.8_1-2.6.x86_64

 Solution 1: The following actions will be performed:
  uninstall nvidia-common-G07-595.84-8.1.x86_64
  uninstall nvidia-compute-G07-595.84-8.1.x86_64
  Uninstall nvidia-video-G07-595.84-8.1.x86_64
  Uninstall nvidia-compute-utils-G07-595.84-8.1.x86_64
  Uninstall nvidia-userspace-meta-G07-595.84-26.1.x86_64
 Solution 2: Uninstall nvidia-open-driver-G07-signed-kmp-meta-595.80-29.1.x86_64
 Solution 3: Keep nvidia-open-driver-G07-signed-kmp-default-595.84_k7.1.8_1-2.6.x86_64 as obsolete
 Solution 4: Install nvidia-common-G07-595.84-8.1.x86_64 anyway, ignoring some of its dependencies

Ideas?

Solution 3 is the way to go for the time being.
The open driver has been updated to 595.91.07_k7.1.8_1 in snapshot 20260813, but the matching proprietary drivers have not yet published in the Nvidia repo.

2 Likes

First time I have this problem with tumbleweed (installed three or four Times in the past)…I find strange that it’s getting released a kernel module without the corresponding user space libraries. Not a big deal but I’ve never seen it before

1 Like

Usually the kernel module lags the user-space packages by a day or two due to the release process and you don’t notice since the Nvidia repo keeps the old packages.
This time maybe somebody at Nvidia is on vacation?

1 Like

Guess so…what about the new branch? I mean the 610.x one…Don’t know if you have more “insights” than me.

is there a way to contact NVIDIA repo owner?

Same here, yes, strange situation. Think suse can postpone realising nvidia-open-driver-G07-signed-kmp version 595.91.07_k7.1.8_1-1.2 until nvidia repo is up to date.

2 Likes

No, they must always be breaking NVIDIA drivers. Seriously, this happens so often it’s as if they want the openSUSE experience to be garbage.

4 Likes

suse/nvidia agreements afaik…not ideal for the end user

Just as a warning, if you’ve timed it wrong then (or the resolver takes a different decision?) then you might get an install problem nvidia-common-G07-595.84-8.1.x86_64 requires '(nvidia-open-driver-G07-signed-kmp = 595.84 if nvidia-open-driver-G07-signed-kmp-meta)', but this requirement cannot be provided and picking Solution 3: keep obsolete nvidia-open-driver-G07-signed-kmp-default-595.84_k7.1.6_1-2.5.x86_64 (rather than uninstalling the meta package, deinstalling most of nvidia, or ignoring dependencies) and you can still end up with a broken graphics install.

The problem ended up being that I’ve got 595.84 for kernel 7.1.6 but the update also includes kernel 7.1.8. So even without getting the two parts of the Nvidia driver out of sync, it’s still possible to get the driver and the kernel out of sync (because old packages aren’t retained for some packages).

(I did an update yesterday, which is the one that broke, and before that I’d updated on Friday 7th - so not a huge gap)

1 Like

I have 7.1.6 kernel (default) installed AND longterm kernel 6.18.42 installed as well. My solution sets seems even worse:


Problem: 1: the installed nvidia-common-G07-595.84-8.1.x86_64 requires '(nvidia-open-driver-G07-signed-kmp = 595.84 if nvidia-open-driver-G07-signed-kmp-meta)', but this requirement cannot be provided
deleted providers: nvidia-open-driver-G07-signed-kmp-default-595.84_k7.1.6_1-2.5.x86_64
                   nvidia-open-driver-G07-signed-kmp-longterm-595.84_k6.18.42_1-2.5.x86_64

 Solution 1: Following actions will be done:
  keep obsolete nvidia-open-driver-G07-signed-kmp-longterm-595.84_k6.18.42_1-2.5.x86_64
  keep obsolete kernel-longterm-6.18.42-1.1.x86_64
 Solution 2: Following actions will be done:
  keep obsolete nvidia-open-driver-G07-signed-kmp-default-595.84_k7.1.6_1-2.5.x86_64
  keep obsolete kernel-default-7.1.6-1.1.x86_64
 Solution 3: Following actions will be done:
  deinstallation of nvidia-common-G07-595.84-8.1.x86_64
  deinstallation of nvidia-compute-G07-595.84-8.1.x86_64
  deinstallation of nvidia-video-G07-595.84-8.1.x86_64
  deinstallation of nvidia-compute-G07-32bit-595.84-8.1.x86_64
  deinstallation of nvidia-compute-utils-G07-595.84-8.1.x86_64
  deinstallation of nvidia-video-G07-32bit-595.84-8.1.x86_64
  deinstallation of nvidia-userspace-meta-G07-595.84-26.1.x86_64
 Solution 4: deinstallation of nvidia-open-driver-G07-signed-kmp-meta-595.80-29.1.x86_64
 Solution 5: break nvidia-common-G07-595.84-8.1.x86_64 by ignoring some of its dependencies

none of them seems reasonable, is waiting the only solution?

Kernel 7.1.8 has been pushed because of kernel-7-1-6-causing-graphical-glitching-on-9070-9060-gpu and the Nvidia stack was left behind.
For the time being Nvidia G07 users can:

  • lock the kernel so they are able to boot 7.1.6 until G07 user-space is updated
  • or configure zypp.conf to keep 3-4 kernels (see /etc/zypp/zypp.conf.README) and boot 7.1.6
  • or install the G07 driver “the hard way” with the .run file if they cannot wait.
1 Like

I just locked nvidia-open-driver-G07-signed-kmp-default version 595.84_k7.1.8_1-2.6 to prevent upgrading it to 595.91.07_k7.1.8_1-1.2 - works and updates fine for now.

And the easier option is to simply keep the meta package…

Which requires staying on the 7.1.6 kernel. If you upgrade the kernel without upgrading the driver package then it will not load.

I see that nvidia-open-driver-G07-signed-kmp-default-595.84_k7.1.8_1-2.6 is still available in the OSS repo, so there should be no need to lock the kernel (at least on systems where the -meta packages are installed).
I have no such HW, so cannot check for sure.

I‘m on the latest kernel without any locks. Only keep the meta package.

Thanks for that. I was able to pull it down manually and apply it to get back to the 7.1.8 kernel. I think I fell into a similar timing issue as @IBBoard somehow as zypper would only find the newer release and the one for 7.1.8 isn’t indexed.

Hi All
Depending on your end use requirements, if you don’t need 32bit nvidia libraries I would suggest the cuda repository and the dkms service.

I used the run file a lot, and probably will going forward. I do use the Nvidia rpms on Leap 16.0. But on Tumbleweed I’m giving the cuda repository a whirl.

It builds the open driver fine and I suspect will rebuild if you boot into a slowroll or default kernel…

If you have installed the new 595.91.07 and need the 595.84_k7.1.8_1-2.6 to have a working nvidia stack, you could add tumbleweeds history archive as repo sudo zypper ar -f \ https://download.opensuse.org/history/20260812/tumbleweed/repo/oss/ \ tw-20260812-oss and install the needed version.

1 Like