OpenMandriva: Lx 26.09 (ROME) rolling · HW: i7-13700K, RTX 5070 Ti, 62 GB RAM
Hey folks 
Quick story about getting ROME working on my RTX 5070 Ti. Everything below happened on my machine and was checked before I wrote it. It’s a rolling distro, so things may have changed since. Hope it saves you some time.
The short version: the Welcome app’s “Install NVIDIA Drivers” button gave me the closed NVIDIA module by default, and I ended up with a black screen. I fixed that by switching to the open module and blacklisting nouveau myself. Later, GRUB couldn’t handle my ~254 MB initrd, so I moved to systemd-boot. It all works now.
1. Welcome app NVIDIA → black screen
New ROME install, RTX 5070 Ti. I clicked Install NVIDIA Drivers in the Welcome app and got a black screen right after the first reboot.
There are actually two NVIDIA kernel module packages in the repos, and the easy path takes the closed one:
dnf repoquery --available 'nvidia-dkms*' # nvidia-dkms-kmod (closed)
# nvidia-dkms-kmod-open (open)
The default installs nvidia-dkms-kmod, and on this very new GPU that one just doesn’t come up clean. So I swapped to the open module:
sudo dnf install nvidia-kmod-common nvidia-dkms-kmod-open nvidia nvidia-32bit
Then double-checked what actually loaded:
lsmod | grep nvidia
journalctl -b -k | grep NVRM # 'loading NVIDIA UNIX Open Kernel Module'
(Userspace libnvidia stays proprietary. That’s true on every distro, not just this one.)
Also: nouveau was never blacklisted
After the NVIDIA stack was installed, nouveau was still around. rpm -qf /etc/modprobe.d/nvidia-blacklist-nouveau.conf says “not owned by any package”. No shipped blacklist exists, so I made one:
printf 'blacklist nouveau\noptions nouveau modeset=0\n' | sudo tee /etc/modprobe.d/nvidia-blacklist-nouveau.conf
sudo dracut --force
One thing that threw me: nouveau_icd.x86_64.json stays behind in lib64dri-drivers even after this. That’s fine. Without the module loaded it does nothing. Don’t worry about it.
2. GRUB, the big initrd, and why I’m on systemd-boot now
The 7.2.6 kernel brought an initrd of about 254 MB:
ls -la /boot/initrd-7.2.6-desktop-1omv2690.img # 265896028
GRUB (2.16-rc1) would just stall and die trying to load it. With TPM enabled it was worse, it ran out of memory during the load. My guess is GRUB’s TPM event logging takes memory that the initrd buffer also needs. I didn’t want to shrink the initrd or disable TPM to work around that, so I switched to systemd-boot. It reads the kernel and initrd straight from the ESP through the firmware, so there’s no size juggling:
sudo bootctl install
sudo efibootmgr # check your own numbers first
sudo efibootmgr -o 2,1,5 # example only, match yours
OM ships kernel-install, so boot entries are generated automatically on every kernel update. I never touched a loader entry by hand. Boot feels faster, and my Windows entry survived.
While staring at boot messages during all this, I saw “No NVIDIA device to probe” in the console output. That was my clue that nouveau still owned the GPU, because nothing blacklisted it. The kernel messages explained everything.
3. The safety net
Before swapping drivers and bootloaders, I set up btrfs snapshots with snapper. On a rolling release this should almost be mandatory:
sudo dnf install snapper
snapper -c root create-config / && snapper -c home create-config /home
systemctl enable --now snapper-timeline.timer snapper-cleanup.timer
Now I have hourly snapshots, and a bad update is a list + rollback command instead of a reinstall.
To the team (no hard feelings): defaulting the Welcome app’s NVIDIA button to the open kmod, or letting people choose, plus shipping a nouveau blacklist file, would have saved me an afternoon. Thanks for ROME, it runs great.