Hello, I own an Nvidia GeForce RTX 5070. I have encountered numerous operational issues with various Mageia ISOs. After extensive testing, I believe I have identified the causes and workarounds to fix these anomalies. This likely only affects recent Nvidia GPUs. I proceeded with the installation of Mageia 10 using the netinstall ISO. During the process, I selected the installation of the proprietary driver. Upon reboot, without any modifications, SDDM fails to start. Analyzing the errors via the command journalctl reveals the following messages: "Feb 07 07:29:27 localhost kernel: NVRM: The NVIDIA GPU 0000:01:00.0 (PCI ID: 10de:2f04) NVRM: installed in this system requires use of the NVIDIA open kernel modules. Feb 07 07:29:27 localhost kernel: NVRM: GPU 0000:01:00.0: RmInitAdapter failed! (0x22:0x56:884) Feb 07 07:29:27 localhost kernel: NVRM: GPU 0000:01:00.0: rm_init_adapter failed, device minor number 0 Feb 07 07:29:27 localhost kernel: NVRM: The NVIDIA GPU 0000:01:00.0 (PCI ID: 10de:2f04) NVRM: installed in this system requires use of the NVIDIA open kernel modules. Feb 07 07:29:27 localhost kernel: NVRM: GPU 0000:01:00.0: RmInitAdapter failed! (0x22:0x56:884) Feb 07 07:29:27 localhost kernel: NVRM: GPU 0000:01:00.0: rm_init_adapter failed, device minor number 0 Feb 07 07:29:27 localhost sddm[13659]: Failed to read display number from pipe Feb 07 07:29:27 localhost sddm[13659]: Attempt 1 starting the Display server on vt 2 failed" After many unsuccessful tests, I concluded that the system does not accept the installation of the Nvidia module in "proprietary" mode. Indeed, the ISO installs the dkms-nvidia-current package by default. However, given the error message, the system requires a MIT/GPL installation. I simply installed the dkms-nvidia-current-open package. This uninstalled the dkms-nvidia-current package. Upon reboot, SDDM started correctly and offered Wayland. Another issue observed is the system suspend with the Nvidia GPU. Upon waking, the display is unusable. The workaround is to modify the /etc/modprobe.d/display-driver.conf file and uncomment the following two lines: "options nvidia-current NVreg_PreserveVideoMemoryAllocations=1 options nvidia-current NVreg_TemporaryFilePath=/var/tmp" Then, force the regeneration of the boot image via the command dracut -f. Note that this is not sufficient; you must also add the following line to the file to disable the framebuffer. By default, it is enabled with recent driver versions: "options nvidia-drm fbdev=0" This point still needs further validation. Finally, it seems it would be beneficial to also install the nvidia-current-cuda-opencl package to take advantage of GPU acceleration in certain applications. This could be a plus. Otherwise, the system is functional and performs well.
Thank you for the detailed report and testing Can you tell exactly which Netinstall ISO you used - name and file date? Assigning to kernel and driver maintainer
CC: (none) => friAssignee: bugsquad => kernel
Hi Here is the link to the relevant ISO file. https://fr2.rpmfind.net/linux/mageia/distrib/10/x86_64/install/images/Mageia-Cauldron-netinstall-nonfree-x86_64.iso
Netinstall files get regenerated now and then in the same name, therefore the date is important.
The open modules are mandatory for Blackwell and facultative for arch GPU from Turing to Blackwell. There is a WIP now for automatic config. Try enabling nonfree/updates_testing and core/updates_testing repos, upgrade ldetect-lst to at least 0.6.69, then (in a previous working installation even with another driver) run drakx11 and let it install the suggested proprietary driver for the Blackwell detected card. Beware you don't have to have installed any nvidia-current* drivers package, since it's not yet handling the Conflicts properly. As for the default values for either using fbdev=0 or 1, it's not modified by default since it was and was causing other troubles so it's left to be modified on demand to the user. Ditto the chnages to NVreg_PreserveVideoMemoryAllocations and NVreg_TemporaryFilePath, which are left to the user to be uncommented on requirement. Note that dracut by default doesn't install any nvidia module in the initrd image. To have them preinstalled in the initramfs you might install the nvidia-<driver>-initramfs package and let it regenerates the dracut image (note that this is outside any drakx11 configuration and might not be detected in case of switching to other driver series, like newfeature or nouveau|modesetting).
CC: (none) => ghibomgx
(In reply to Giuseppe Ghibò from comment #4) > The open modules are mandatory for Blackwell and facultative for arch GPU > from Turing to Blackwell. Correcting, "facultative" from Turing to Ada Lovelace/Hopper arch.