I have discovered an issue that I can only assume is something to do with dnf, because what used to work, no longer seems to…
Let me explain… I have an offline copy of the nvidia cuda repo (Index of /compute/cuda/repos/rhel9/x86_64), which I could query for all available packages using dnf in the manner:
dnf list available kmod-nvidia* --showduplicates
… and I used to see all the versions in the repo. Now, I only see two!
I’ve reposync’d the entire repo down to a local host and run createrepo, with the same results.
Also, do you have any package version exclusions in your DNF config? An excluded package version will be ignored by all dnf operations, including list.
Side note: in dnf5 it should be --available. The command as written says “list packages whose exact name is ‘available’ or whose name matches ‘kmod-nvidia*’”
But that isn’t the problem here, since you’re specifying neither --available nor --installed, the implicit default is --all.
As far as I can tell, it should. (The docs suggest that --available rather than available is the right switch in DNF4 too, but again this won’t be the cause of your expected packages being excluded.)
Unfortunately it’s difficult on a Fedora forum to answer your original question “Has something changed?” with respect to RHEL. Fedora has moved on substantially from the versions you are working with, so we don’t have a great handle on what might have recently changed on your system.
$ dnf list --all 'kmod-nvidia*' --showduplicates
Not root, Subscription Management repositories not updated
Last metadata expiration check: 0:36:15 ago on Mon 19 Jan 2026 14:37:40 GMT.
Available Packages
kmod-nvidia-open-dkms.noarch 3:590.44.01-1.el9 cuda-rhel9-x86_64
kmod-nvidia-open-dkms.noarch 3:590.48.01-1.el9 cuda-rhel9-x86_64
You have said you are using RHEL and that output shows the cuda-rhel9-x86_64 repo.
Fedora does not use that repo and the versions you are showing are way behind what we see for fedora from the rpmfusion-nonfree repos. I don’t think anyone using fedora exclusively will be able to assist.
Maybe you have to look at the modules: dnf --disablerepo=* --repofrompath test,https://developer.download.nvidia.com/compute/cuda/repos/rhel9/x86_64/ module list nvidia-driver
Hi Joe, Thanks, that gives me 55 rows. I’m still waiting for a colleague to remind me how we listed the kmod-nvidia kernels. I forgot. Old age catching up with me.
Just an update on this, in case any of you are curious, plus I hate forum threads that just peter out with no obvious solution…
To clarify the situation, I have three geographically distributed repositories, let’s call them Tom, Dick and Harry. I download updates to Tom, then they are rsync’d to Dick and Harry.
I deleted repodata on Tom, ran createrepo, did a search and I only get 19 hits on kmod-nvidia
I deleted repodata on Dick, ran createrepo, did a search and I get 361 hits on kmod-nvidia!
I deleted repodata on Harry, ran createrepo, did a search and I get 361 hits on kmod-nvidia!
By way of a test I rsync’d back from Dick to Tom and I still only get 19 hits on kmod-nvidia.
Obviously there’s something different on Tom, but as yet I have not found what!
