NVIDIA Open Driver Power Management#

This article covers power management behavior for the nvidia-open kernel driver on Bluefin laptop systems — specifically suspend/resume reliability, battery drain patterns, SELinux interactions, D3/RTD3 state handling, and the udev rules that control NVIDIA sub-device power. It does not cover driver packaging, Optimus display routing, or container GPU passthrough.

All Bluefin NVIDIA variants use nvidia-open exclusively; the closed driver was deprecated in October 2025. 1

Default Power Management Configuration#

Bluefin ships no custom modprobe.d configuration for NVIDIA power management parameters. Parameters like NVreg_DynamicPowerManagement and NVreg_EnableGpuFirmware are left at NVIDIA driver defaults — they are not set in ublue-os/bluefin, ublue-os/main, or ublue-os/akmods. 2

The one active power-related intervention is a udev rules file in ublue-os-nvidia-addons that removes three NVIDIA PCI sub-devices on add:

3

These removals improve idle power consumption on systems where the GPU audio and USB endpoints are not needed. They do not affect rendering or compute performance.

UseKernelSuspendNotifiers#

The NVIDIA 595.45.04 driver introduced UseKernelSuspendNotifiers=1 in the negativo17 packaging, which switches the suspend mechanism from legacy nvidia-sleep.sh scripts to kernel notifier callbacks. This setting was the upstream trigger for the March 2026 regression. 4

If you need to apply custom NVreg parameters (e.g., NVreg_DynamicPowerManagement=0x02 for fine-grained RTD3 power management), create a file in /etc/modprobe.d/ — changes there persist across OS updates. To bake them into your own image, add the file to your Containerfile.

March 2026 SELinux Suspend Regression#

Build 43.20260312.1 introduced a critical regression that caused complete battery drain on all NVIDIA laptop configurations. 5

Root cause: SELinux denied systemd-sleep the CAP_PERFMON capability (capability 38) in the systemd_sleep_t context. The NVIDIA 595.45.04 driver's new UseKernelSuspendNotifiers=1 mode requires this capability to complete GPU shutdown hooks. Without it, the suspend sequence aborts mid-flight. 4 5

Failure cascade: 5

1. systemd-sleep denied perfmon → emits Failed to put system to sleep. System resumed again: Operation not permitted
2. GPU firmware crashes mid-suspend: Xid 120 / GSP task exception (kern_bus_vbar2.c:346)
3. GPU enters a corrupted state — nvidia-smi, nvtop, btop hang indefinitely
4. nvidia-powerd logs ERROR! Failed to get AC Line status (RM Error Code 98) every second
5. Lid appears closed, system appears suspended — GPU and fans remain fully active
6. Complete battery drain within hours

SELinux audit entry:

AVC avc: denied { perfmon } for pid=50444 comm="systemd-sleep" capability=38
    scontext=system_u:system_r:systemd_sleep_t:s0 tclass=capability2 permissive=0

Affected versions:

5

The upstream fix was a revert of UseKernelSuspendNotifiers — not an SELinux policy patch. The SELinux policy changes are tracked at negativo17/nvidia-driver#196 and require:

allow systemd_sleep_t self:capability2 perfmon;
allow systemd_sleep_t tmp_t:file { open write };

5

GPU State Corruption After Failed Suspend#

Once the NVIDIA GPU enters a corrupted state after an aborted suspend, there is no software recovery path — a full system reboot is required to perform a hardware-level GPU reset. 5

Signs of GPU state corruption: 5

• nvidia-smi hangs indefinitely (does not return)
• nvtop and btop freeze on GPU metrics
• nvidia-powerd logs ERROR! Failed to get AC Line status repeatedly
• HDMI-connected external displays remain dark
• Kernel logs show NVRM: _issueRpcLarge: rpcSendMessage failed errors

On Optimus systems, HDMI output (which is physically wired to the dGPU) will also be unavailable until after reboot.

Other Battery Drain Patterns#

Framework 13 — Spurious Wake Events#

Framework 13 NVIDIA Optimus users have reported spurious wake events causing severe drain (~90% → 10% in a few hours in a bag). 6 The workaround is to disable the lid and trackpad as wakeup sources:

sudo tee /etc/systemd/system/lid-wakeup-disable.service << 'EOF'
[Unit]
Description=Disable lid wakeup
After=sysinit.target

[Service]
Type=oneshot
ExecStart=/bin/sh -c 'echo "disabled" > /sys/bus/acpi/devices/PNP0C0D:00/power/wakeup'
ExecStart=/bin/sh -c 'echo "disabled" > /sys/bus/i2c/devices/i2c-PIXA3854:00/power/wakeup'
RemainAfterExit=yes

[Install]
WantedBy=multi-user.target
EOF
sudo systemctl enable --now lid-wakeup-disable.service

This is safe because it leaves the power button as the only wakeup source. 6

Framework 16 — s2idle Drain#

Framework 16 AMD models are limited to s2idle (suspend-to-idle), not S3 deep sleep — a firmware constraint. NVIDIA dGPU configurations compound idle drain on these systems. 6

Verify your available sleep states with: cat /sys/power/mem_sleep

Workarounds & Configuration#

Update or Roll Back#

The fastest fix for any suspend regression is to update to current stable, or roll back to the last known good deployment:

# Roll back to previous deployment
sudo ostree admin set-default 1

# Or use the interactive helper
ujust rebase-helper

5

Custom SELinux Policy (Temporary)#

For systems stuck on an affected build without an update path:

ausearch -c 'systemd-sleep' --raw | audit2allow -M my-systemdsleep
semodule -X 300 -i my-systemdsleep.pp

5

Important: semodule policies are not guaranteed to survive OS image updates on bootc-managed systems. For a permanent fix, embed the policy module in your own image definition (e.g., via a custom Containerfile), or wait for the policy fix to land upstream.

Custom NVreg Parameters#

Bluefin images do not set NVreg_DynamicPowerManagement or NVreg_EnableGpuFirmware. To apply them yourself:

# Example: enable fine-grained RTD3 power management
echo "options nvidia NVreg_DynamicPowerManagement=0x02" | \
  sudo tee /etc/modprobe.d/nvidia-pm.conf

Files in /etc/modprobe.d/ persist across OS updates on bootc systems. Rebuild your initramfs if needed: sudo dracut --force.

Diagnostics#

# Check which sleep states the kernel offers
cat /sys/power/mem_sleep

# Watch suspend/resume transitions live
journalctl -f | grep -E 'suspend|resume|PM:|systemd-sleep'

# Look for SELinux denials (key for diagnosing suspend failures)
sudo ausearch -m avc -ts recent | grep -i sleep

# Check current NVIDIA driver version
nvidia-smi

# Verify the GPU is not in a hung state
timeout 5 nvidia-smi && echo "GPU OK" || echo "GPU HUNG — reboot required"

# Check current image version and rollback availability
bootc status

Quick Symptom Reference#

Key Source Files#