RTX 5070 Ti (Blackwell/GB203): Pageflip timeout on resume from s2idle suspend - displays fail to wake

Graphics / Linux Linux
1

After resuming from s2idle suspend, all displays fail to wake. The system remains responsive (accessible via SSH), but KWin reports repeated “Pageflip timed out! This is a bug in the nvidia-drm kernel driver” errors. Restarting the display manager (SDDM) recovers the displays without requiring a full reboot.

Hardware Configuration

GPU: NVIDIA GeForce RTX 5070 Ti (GB203, Blackwell architecture)

  • Vendor: MSI
  • VBIOS: 98.03.58.00.94
  • Bus: PCIe (0000:0a:00.0)

System:

  • CPU: AMD Ryzen 7 5800X
  • Motherboard: Gigabyte X570 AORUS ELITE (BIOS F40g, dated 03/11/2025)
  • RAM: 32GB DDR4
  • Secure Boot: Enabled

Displays (triple monitor):

  • DP-1: ASUS ROG PG279Q (2560x1440)
  • DP-2: ViewSonic VA2756 (1920x1080)
  • HDMI-A-1: ViewSonic VA2756 (1920x1080)

Software Configuration

  • Driver: 590.48.01 (nvidia-open-dkms)
  • Kernel: 6.18.8-3-cachyos (CachyOS/Arch Linux)
  • Desktop: KDE Plasma 6.5.5 with KWin on Wayland
  • Display Manager: SDDM

Kernel parameters:

nvidia-drm.modeset=1 nvidia-drm.fbdev=1 mem_sleep_default=s2idle

Sleep state: s2idle (confirmed via /sys/power/mem_sleep)

Steps to Reproduce

  1. Boot system normally with triple-monitor configuration
  2. Allow system to fully initialize (all displays active, desktop functional)
  3. Initiate suspend via systemctl suspend
  4. System enters s2idle suspend successfully
  5. Attempt to wake system (keyboard/mouse input)
  6. System wakes (SSH accessible, audio works) but all displays remain off

Observed Behavior

The system suspends without errors. On resume attempt, the displays never turn on. The journal shows KWin repeatedly logging:

kwin_wayland[1178]: Pageflip timed out! This is a bug in the nvidia-drm kernel driver
kwin_wayland[1178]: Please report this at https://forums.developer.nvidia.com/c/gpu-graphics/linux

These messages repeat continuously (approximately once per second) until the display manager is restarted.

Recovery

Running sudo systemctl restart sddm via SSH successfully recovers all displays without requiring a reboot. This suggests the GPU itself is functional but the DRM/KMS layer is stuck.

Expected Behavior

Displays should wake and resume normal operation after s2idle suspend.

Additional Notes

I previously ran driver version 580.119.02 on this same hardware without suspend/resume issues. The problem appeared after upgrading to later 580.x versions and persists with 590.48.01. I’m reporting against 590 since that’s where active Blackwell development is occurring.

I’m happy to provide additional logs, test patches, or try different configurations if that would help isolate the issue.

nvidia-bug-report.log.gz (969.0 KB)