Chasing the NVIDIA Ghost: A Linux Desktop GPU Odyssey

Sometimes, fixing a GPU on Linux feels less like tech work and more like a mystical quest. That was exactly my journey with an NVIDIA GTX 1050 on Ubuntu 24.04 LTS.

The Problem

Xorg kept crashing. Logs screamed about missing modules. The system would randomly freeze, then reboot. Despite having NVIDIA 535 installed earlier, only one monitor worked under the fallback modesetting driver. Nouveau was blacklisted, yet the GPU refused to fully awaken.

Attempt #1: Clean slate

• Purged all nvidia-* drivers, blacklists removed.
• Verified that nouveau was not loaded.
• Kernel headers were up to date.

Success? ✅ Partial. The system booted with modesetting, but Xorg was still using VESA, single monitor only, GPU idle.

Attempt #2: Reinstallation Loop

• Installed NVIDIA 535 via apt, purged, reinstalled, repeat.
• Every time, nvidia-smi reported the driver version, kernel modules loaded, but Xorg refused to load nvidia_drv.so.
• System stayed unstable: freezes, crashes, reboots.

Failure? ❌ Absolutely. Circular uninstall/install loops didn’t solve the Xorg integration issue.

Attempt #3: System Health Checks

Ran a custom script to check NVIDIA and GPU health. Observations:

• Kernel modules (nvidia, nvidia_uvm, nvidia_modeset) were loaded.
• Device nodes existed.
• Xorg still fell back to VESA.

Success? ✅ Kernel side only.
Failure? ❌ Userspace / Xorg still broken.

Attempt #4: Diagnosing with Logs

• Xorg logs confirmed: Failed to load module "nvidia"
• Only VESA was being used.
• Missing Xorg NVIDIA driver (nvidia_drv.so) and GLX integration.

Lesson learned: having the kernel driver isn’t enough. Xorg needs its own modules.

Reflection

After multiple rounds of purges, installs, and system reboots, the issue persists. Even with assistance from LLM guidance, the GPU refuses full cooperation. The circular loop of apt purge nvidia-* → apt install nvidia-driver-535 became almost meditative—successive failures teaching patience more than solutions.

Takeaways

1. Always check both kernel and Xorg drivers.
2. Purging NVIDIA drivers is necessary but insufficient; Xorg module installation matters.
3. Legacy boot adds complications; EFI could help with signing and DKMS.
4. LLMs can guide, but real hardware quirks still hold the final say.

In the end, fixing NVIDIA on Linux is not always linear—sometimes it’s a labyrinth. But documenting each success, failure, and circular attempt may be the most useful guide for the next brave soul who dares to wrestle with NVIDIA on Linux.