580 release feedback & discussion

Graphics / Linux Linux
935

As a quick hack you could try enable Nvidia’s compositor pipeline from the control panel – X Server Display Configuration – Advanced. If you already did, never mind :)

936

Since last Linux Manjaro update, the active window blink or not refresh.

For example, at start up, xfce4-panel blink. When I use an xfce4-terminal, I type some char, the last one don’t show, and some time the content of the window disappear and show the background image.

Graphics:
  Device-1: NVIDIA GA107 [GeForce RTX 3050 6GB] driver: nvidia v: 580.119.02
  Device-2: Logitech QuickCam E 3500 driver: snd-usb-audio,uvcvideo
    type: USB
  Display: x11 server: X.org v: 1.21.1.21 driver: X: loaded: nvidia
    gpu: nvidia,nvidia-nvswitch resolution: 2560x1440~60Hz
  API: EGL v: 1.5 drivers: nvidia,swrast
    platforms: gbm,x11,surfaceless,device
  API: OpenGL v: 4.6.0 compat-v: 4.5 vendor: nvidia mesa v: 580.119.02
    renderer: NVIDIA GeForce RTX 3050/PCIe/SSE2
  API: Vulkan v: 1.4.328 drivers: nvidia surfaces: N/A
  Info: Tools: api: eglinfo, glxinfo, vulkaninfo de: xfce4-display-settings
    gpu: nvidia-settings,nvidia-smi x11: xprop,xrandr

nvidia-bug-report.log.gz (439.1 KB)

2 Likes
937

I can confirm the XFCE problems are fixed after reverting to NVIDIA-Linux-x86_64-580.105.08.run … I also had a couple of similar issues with the taskbar and desktop-switching in KDE but haven’t booted back into that just yet … enabling compositor pipeline did NOT work

Edit: So far so good on KDE as well

3 Likes
938

I am also experiencing issues on Xfce; Cinnamon, Budgie, KDE are working as expected. For me, the issue is only visible on a single app - kitty- which works fine on the aforementioned DEs.

1 Like
939

Subject: RTX 5080: System poweroff hangs for ~10 minutes with 580.105.08 open driver (regression from 575.57.08)

System Configuration:
• GPU: NVIDIA GeForce RTX 5080
• Driver: nvidia-driver-580-open (580.105.08-0ubuntu1, open kernel modules)
• OS: Ubuntu 24.04 LTS (Noble)
• Kernel: 6.8.0-90-generic (also tested with 6.8.0-87, 6.8.0-88)
• Display: 3× monitors (2× DP 2560×1440, 1× HDMI 1920×1080), lightdm/X11
• Previous working: nvidia-driver-575-open (575.57.08) — no issues from June-November 2025

Issue Description:
System poweroff (systemctl poweroff or shutdown -h now) hangs with black screen for approximately 10 minutes before completing. Systemd reaches poweroff.target and completes all shutdown tasks cleanly (verified via journal logs), but the final ACPI S5 transition is delayed.

Key Behavioral Observations:

  1. S3 suspend works normally - no delays or issues
  2. Poweroff works immediately after S3 resume - if system is suspended then resumed, subsequent poweroff completes in <5 seconds
  3. ~10 minute timeout: If left alone, system eventually powers off successfully after the timeout
  4. Occurs regardless of display configuration - tested with single monitor and multi-monitor
  5. Systemd logs are clean - all filesystems unmounted, no “stop job running” messages, journal stops normally

Diagnostic Evidence:
• All nvidia-* packages at consistent 580.105.08-0ubuntu1 version (no mismatches)
• Journal logs show clean shutdown up to systemd-shutdown[1]: Shutting down.
• No kernel panics, no hardware errors in dmesg
• Issue persists with various modprobe options tested:
◦ NVreg_PreserveVideoMemoryAllocations=0
◦ NVreg_EnableS0ixPowerManagement=0
◦ Various GRUB/ACPI kernel parameters

Reproducibility:
• 100% reproducible on fresh boot → GUI session → poweroff
• 0% reproducible after suspend/resume → poweroff

Analysis:
The suspend/resume cycle appears to re-initialize GPU power management state correctly, allowing immediate poweroff. This suggests the driver’s initial probe/initialization at boot leaves the GPU in a state where shutdown transitions time out, but the suspend/resume path properly configures the hardware.

The ~10 minute delay suggests a retry loop or hardware polling timeout in the driver’s shutdown sequence waiting for GPU acknowledgment that never arrives from the initial power state.

Workaround:
Suspend system first, then poweroff (immediate completion).

Request:
Please investigate the difference between driver initialization in nvidia_probe() vs nvidia_resume() paths, specifically around power state management registers. The 575 series handled this correctly; this appears to be a regression in 580.105.08.

940

Performance limiters are broken on Blackwell:

941

I role back to driver version 575.64.05 and everything is back to normal.

942

Just tried the v590.48 drivers on Xfce. All of my related problems are now fixed.

1 Like