Skip to content

Cerberos is not starting

Rocky Linux 8 + NVIDIA

Failure Recovery & Maintenance Manual

This document covers reliable recovery and maintenance of NVIDIA GPUs on
Rocky Linux 8 / RHEL 8–derived systems, specifically for cases where:

  • The GUI fails to start
  • /dev/dri is missing
  • nouveau reports “unknown chipset”
  • NVIDIA breaks after kernel updates
  • RPM Fusion package names appear inconsistent

This manual is based on real-world recovery and reflects correct EL8 behavior.

1. Core Facts (Memorize These)

Hardware & Drivers

  • Some NVIDIA GPUs are not supported by nouveau
  • Message: nouveau: unknown chipset
  • → nouveau will never work on this hardware

Graphics Stack

  • GUI requires DRM
  • DRM creates /dev/dri
  • On unsupported NVIDIA GPUs:
  • Only the proprietary NVIDIA driver provides DRM

Kernel Rules (Critical)

  • NVIDIA kernel modules must exactly match the running kernel
  • Even minor version differences break the driver
  • EL8 repositories do not keep old kernel graphics modules forever

2. Golden Rules

✅ Disable Secure Boot (mandatory for NVIDIA)
✅ Boot into multi-user.target for repairs
✅ Pick one kernel and stay on it
✅ Use akmod-nvidia, not dkms
✅ Rebuild NVIDIA only for the current kernel

❌ Do not rely on nouveau on unsupported GPUs
❌ Do not chase old kernels
❌ Do not mix kernels and drivers
❌ Do not install random NVIDIA package names from blogs

3. Emergency: Boot Into Text Mode

From GRUB, edit the boot entry and append: systemd.unit=multi-user.target

After login, lock it in:

sudo systemctl set-default multi-user.target

4. Verify Secure Boot (Mandatory)

mokutil --sb-state # verify secure boot state

Must show: SecureBoot disabled

If enabled → disable Secure Boot in BIOS. There is no workaround.

5. Clean NVIDIA Reinstall (Minimal & Correct)

Step 1 — Install a Supported Kernel Stack

sudo dnf install -y \
kernel \
kernel-core \
kernel-modules \
kernel-modules-extra
sudo reboot

After reboot 

uname -r

✅ This kernel is now your only supported kernel

Step 2 — Enable Required Repositories

sudo dnf config-manager --set-enabled powertools

sudo dnf install -y \
https://download1.rpmfusion.org/free/el/rpmfusion-free-release-8.noarch.rpm \
https://download1.rpmfusion.org/nonfree/el/rpmfusion-nonfree-release-8.noarch.rpm

sudo dnf clean all
sudo dnf makecache

Step 3 — Install NVIDIA (Correct Package)

sudo dnf install -y akmod-nvidia

✅ Do not manually install:

xorg-x11-drv-nvidia nvidia-driver *-cuda

RPM Fusion resolves the correct userspace automatically on EL8.

Step 4 — Build NVIDIA for This Kernel Only

sudo akmods --force --kernel $(uname -r)

ls /var/cache/akmods/nvidia/ # verify

Expected: kmod-nvidia-<kernel>.rpm

Step 5 — Install the Built Module Explicitly

sudo dnf install -y /var/cache/akmods/nvidia/kmod-nvidia-*.rpm

sudo modprobe nvidia # load driver

lsmod | grep nvidia # verify

Expected:

nvidia
nvidia_drm
nvidia_modeset
nvidia_uvm

6. Validate Graphics Stack

ls /dev/dri

Expected:

card0
renderD128

Optional verification:

command -v nvidia-smi
nvidia-smi

(Probably not going to work, but that is fine, GUI works even without nvidia-smi, but it’s a good confirmation.)

7. Start the GUI

sudo systemctl set-default graphical.target
sudo systemctl start graphical.target

You should now get:

  • Graphical login
  • Proper resolution
  • Hardware acceleration

8. Make the Fix Permanent

Disable nouveau forever

sudo nano /etc/default/grub

set GRUB_CMDLINE_LINUX="rhgb quiet module_blacklist=nouveau"

Rebuild:

sudo grub2-mkconfig -o /boot/grub2/grub.cfg
sudo dracut --force

Pin the Working Kernel:

sudo grub2-set-default 0
sudo grub2-editenv list

9. Quick Diagnostic Table

Symptom Meaning Fix
/dev/dri missing No DRM Fix kernel / NVIDIA
nouveau: unknown chipset Unsupported GPU Use NVIDIA
modprobe nvidia: key not available Secure Boot enabled Disable Secure Boot (SB)
GUI fails, text works Driver mismatch Rebuild NVIDIA
akmods builds wrong kernel Kernel drift Pin kernel

10. Final Takeaway

On Rocky Linux 8, NVIDIA stability depends on kernel discipline. One kernel, one NVIDIA build, Secure Boot off — and the system is stable.

11. Additional possiblilities

Edit GDM configuration:

sudo nano /etc/gdm/custom.conf

Set: 

[daemon]
WaylandEnable=false

Save, then restart GDM or reboot: 

sudo systemctl restart gdm