[580.105.08 Regression] Hibernate (S4) no longer powers off, fans stay on, resume still works

Graphics / Linux Linux
1

Hi, I’m seeing a regression specifically in 580.105.08. Hibernate (ACPI S4) worked on 580.95.05 and stopped working immediately after updating to 580.105.08. Nothing else in my setup changed.

What happens now is:
The system writes the hibernate image, the display shuts off, USB briefly power-cycles, but the system never actually powers down. Fans keep running, power LED stays on forever, and I have to do a manual power button shutdown. After that, the system does resume from the hibernate image just fine, so the snapshot is being created properly - the machine just never transitions into S4.

System details:

  • GPU: RTX 3060 Ti

  • Driver package: nvidia-dkms

  • Working version: 580.95.05

  • Broken version: 580.105.08

  • Kernels tested:

    • 6.17.8-zen

    • 6.12.58-lts
      (same issue across all)

  • Distro: Arch Linux

  • Display server: Wayland

  • Motherboard: MSI PRO Z690-A DDR4

  • Hibernate config:

    • HibernateState=disk

    • HibernateMode=shutdown

  • Swap on disk (not encrypted)

This is the upgrade timeline from pacman:

[2025-11-07] upgraded nvidia-dkms (575.64.03 → 580.95.05)
[2025-11-14] upgraded nvidia-dkms (580.95.05 → 580.105.08)

Hibernate still worked on 580.95.05 and broke only after the 580.105.08 update. Here’s the journal output from a failing hibernate attempt on 580.105.08:

systemd-sleep[1723]: Performing sleep operation 'hibernate'...

PM: hibernation: Creating image:
PM: hibernation: Need to copy 1222466 pages
PM: hibernation: Normal pages needed: 1222466 + 1024, available pages: 2922493
PM: hibernation: hibernation exit

systemd-sleep[1723]: System returned from sleep operation 'hibernate'.
systemd-logind[504]: Operation 'hibernate' finished.

NVIDIA’s hibernate/resume services report success but the system never shuts off:

Starting NVIDIA system hibernate actions...
nvidia-hibernate.service: Deactivated successfully.
Starting NVIDIA system resume actions...
nvidia-resume.service: Deactivated successfully.

Physical behavior:
– screen off
– keyboard LEDs off → immediately on
– fans keep spinning
– system stays powered indefinitely
– manual shutdown required
– next boot resumes correctly

Rolling back from 580.105.08 → 580.95.05 restores correct S4 behavior immediately.

I can attach a full nvidia-bug-report.sh log, PM trace, or ACPI trace if needed. Just let me know what’s useful.

2

Confirming the S4 regression (first seen in 580.105.08) persists in 590.48.01.

Setup: RTX 3060 Ti, Arch Linux, 6.18-zen, Wayland.
Regression: 580.95.05 (Working) → 580.105.08+ (Broken).

Symptoms:
Snapshot and resume work correctly, but the system never reaches ACPI S4 power-off.

  • With hooks enabled: Fans/LEDs stay on indefinitely; requires manual shutdown.

  • With hooks disabled: System resumes immediately after image creation.

Attached are nvidia-bug-report.sh logs (including acpidump) for both hook states. Rolling back to 580.95.05 immediately restores correct behavior.

nvidia-bug-report-590.48.01-hibernate-hooks-enabled.log.gz (3.0 MB)

nvidia-bug-report-590.48.01-hibernate-hooks-disabled.log.gz (3.0 MB)

3

confirming I see this same issue as well on my laptop.

Arch Linux 6.17.1
Nvidia RTX 4050 Laptop GPU
Wayland

4

The issue seems to have finally been fixed for me with the 580.159.03 update.

I had the issue from 580.105.08 (Nov 2025) to 580.142 (March 2026).

I’m only using 580 branch, because my GPU (GTX 970) is not supported by 590.X and newer.

5

595.71.05 on 7.0.3-arch1-1 still failing to power off

Switching nvidia driver to 580.105.08 does work correctly.