ticket #0002gpu-blackoutdifficulty: moderate~25 min

Black screen after an update

You updated, you rebooted, and now it's a black screen, a blinking cursor, or a login screen that loops forever. Nine times out of ten that's the NVIDIA driver. And you can fix it without ever seeing your desktop.

A triple-fan graphics card on a workbench, lit by the amber glow of a CRT monitor

01What actually happened

Your graphics driver and your kernel are a dance duo. A system update replaced one partner, and the other missed the rehearsal. The NVIDIA driver didn't get rebuilt for the new kernel. No driver, no picture. Your files are fine, your system is fine; the projector bulb just needs re-seating.

02Get a console

The desktop is gone but the machine is alive. At the black screen (or the looping login), press:

the secret door

Ctrl + Alt + F3. This switches to a text console. You'll get a plain login: prompt. Type your username, then your password (nothing appears while you type it. That's normal, keep going).

If F3 does nothing, try F2 through F6. Once you're logged in, everything below happens right here.

03Ask the card

nvidia-smi

What this does: asks the NVIDIA driver for a status report. Read-only. A neat table = driver is alive (your problem is elsewhere. Try step 5's kernel trick). command not found or couldn't communicate with the driver = keep going, this guide is for you.

04Reinstall the driver

On Ubuntu, Mint, Pop!_OS and friends, the easiest reliable route:

sudo ubuntu-drivers autoinstall

What this does: detects your GPU and installs the recommended driver package for it, including the kernel modules that went missing. Downloads a few hundred MB; needs your password.

Wired internet helps here; if you only have Wi-Fi and it's also down, USB-tether your phone. When it finishes:

sudo reboot

What this does: restarts the machine. Cross your fingers with dignity.

On Fedora the equivalent driver lives in RPM Fusion (akmod-nvidia rebuilds itself per kernel. Give it a few minutes after boot). On Arch, reinstall nvidia or switch to nvidia-dkms.

05The escape hatches

Hatch A. Boot yesterday's kernel

Reboot, hold Shift (or tap Esc on UEFI) for the GRUB menu → Advanced options → pick the second-newest kernel. If the desktop returns, the new kernel and driver just need step 4 to make up.

Hatch B. Check DKMS did its homework

dkms status

What this does: lists drivers that rebuild per-kernel and whether each build succeeded. Read-only. If nvidia shows for the old kernel but not the new one, run sudo dkms autoinstall to force the rebuild.

Hatch C. The clean slate

When the driver install is truly mangled (half-installed .run files, mixed versions), remove every NVIDIA package and start fresh:

sudo apt purge '~nnvidia' && sudo apt autoremove

What this does: uninstalls all NVIDIA driver packages (Debian/Ubuntu). The open-source fallback driver (nouveau) takes over on next boot. Ugly but visible. Then redo step 4.

06Verify and prevent

Back on the desktop, confirm the duo is dancing:

nvidia-smi

What this does: same status check as step 3. This time you want the neat table with your GPU's name in it.

  • Stick with your distro's packaged driver. The .run installer from nvidia.com is how most of these tickets get opened.
  • After big updates, let the machine sit a minute before rebooting. DKMS may still be compiling.
  • Reading this before disaster? Note the Ctrl+Alt+F3 trick somewhere. It's the whole game.
ticket closing note

A black screen looks like a dead computer. It's not. It's a shy one. The console is always there, and now you know the knock.