OPINIONATED OMARCHY
TROUBLESHOOTING CORPUS
790
RECORDS
790
AUDITED
0
UNCHECKED
2388
SOURCES
LIVE
gpu-drivers high occasional DISPUTE

Stop NVIDIA freezing when the display blanks (GSP Timeout / Xid 119)

omarchyarchendeavouroscachyosnvidiahyprlandwaylanddesktoplaptop
CORRECTED An auditor found the fix wrong or incomplete and rewrote it.
AUDIT NOTE

Checked on this workstation, which is Omarchy 4.0.2-1, Hyprland 0.56.2, kernel 7.1.9, nvidia-open-dkms 610.57.04-1 on an RTX 3090, and against the cited pages. The core of the record holds. The Arch wiki's NVIDIA/Troubleshooting section "System freeze when the display powers off or on resume" carries the same Xid 119 GSP timeout log, the same low-clock explanation, the same nvidia-persistenced prerequisite, the same nvidia-smi -lgc and -lmc test values and byte-for-byte the same nvidia-clocks.service unit, so the fix body is a faithful copy of a live source. Confirmed on this machine that nvidia-smi 610.57.04 still offers -lgc, -lmc, -rgc and -rmc, so the commands have not aged out. Three things are wrong. First and worst, the stopgap: "removing or raising the DPMS timeout in your hypridle config (~/.config/hypr/hypridle.conf)" cannot be done on Omarchy 4. Confirmed on this machine that pacman -Q hypridle hyprlock reports neither package installed, that ~/.config/hypr/ contains no hypridle.conf, and that grepping /usr/share/omarchy/shell/, /usr/share/omarchy/bin/ and /usr/share/omarchy/default/ for dpms returns only key_press_enables_dpms and mouse_move_enables_dpms in default/hypr/input.lua, which are about waking from DPMS rather than entering it. Reading /usr/share/omarchy/shell/plugins/services/idle/Service.qml confirms the Omarchy 4 idle cycle is a screensaver window followed by the lock, with no blanking step, driven by idle.screensaver and idle.lock in ~/.config/omarchy/shell.json (150 and 300 seconds on this box) and disabled by the stay-awake marker that omarchy-toggle-idle writes. The old advice is kept as the plain-Arch branch. Second, the verify field reads journalctl -b -1 -k, the previous boot. If the fix works there is no reboot, so that command inspects the wrong boot and would show the old crash forever. Corrected to -b with a note on when -b -1 is right. Third, the record does not scope itself to hardware that has GSP firmware at all. Omarchy ships omarchy-hw-nvidia-gsp, read here, which splits at PCI device id 0x1e00, and install/hardware/nvidia.sh, retrieved from quattro today, installs nvidia-open-dkms above that line and nvidia-580xx-dkms below it. Two additions. The danger field was thin: it covers power and heat but not the trap a reader will walk into next, which is the Arch wiki's own GSP-firmware section recommending NVreg_EnableGpuFirmware=0 while stating that it only works with the proprietary driver. Omarchy's GSP path is the open driver, so that is not available, and with Omarchy's early KMS a driver that will not initialise costs the display. And omacom/omarchy issue 11249, filed on Omarchy 4.0.3 with nvidia-open-dkms 610.57.04, reports an aquamarine 0.15.0 regression where an external monitor powered off never recovers, on nvidia-drm among others, with an NVRM GSP task watchdog timeout noted in one of three attempts. That is a different root cause with an almost identical presentation and it is live on current Omarchy, so the fix now opens by separating the two. On sources: both cited GitHub issues fail to support this record and go in sources_remove. Read in full, issue 2112 is an Omarchy 3 display-wake bug whose only suggested workaround is editing hypridle.conf, and issue 2635 is the same family across Omarchy 3.0 to 3.8, reported on AMD, Intel Iris and NVIDIA alike, whose stated workaround is systemctl restart sddm. Neither contains the string GSP, Xid, or any clock parameter, so neither supports the claim it was cited for. Frequency is dropped from common to occasional: searching the omacom/omarchy tracker for "GSP Timeout" and for Xid 119 returns nothing on point across 63 and 5 hits respectively, and Omarchy 4's own idle path never issues the DPMS off that the record names as the primary trigger, so the exposure on this distribution is narrower than common implies. Severity stays high, since the outcome is a power-button reset. NOT exercised: nothing here was reproduced. Suspend cannot be run on this workstation for this audit, and no hyprctl command that changes DPMS state was issued, because this machine's lock screen cannot be released headlessly. The clock-pinning fix is therefore carried on the Arch wiki's authority plus a check that the nvidia-smi flags still exist, not on a freeze I induced and cleared.

The cause above was rewritten on 2026-09-11 to match this note; the fix was corrected by the audit itself.
RISK

Locking clocks raises idle power draw and temperature. On a laptop that measurably shortens battery life, so do not set the minimum near the card's maximum on a thermally constrained machine. Separately, do not reach for the other workaround you will find for GSP problems. The Arch wiki states that NVreg_EnableGpuFirmware=0 only works with the proprietary NVIDIA driver, and Omarchy installs the open modules (nvidia-open-dkms) on every GSP-capable card, so on Omarchy 4 that parameter is not an available workaround. Omarchy also early-loads the NVIDIA modules from the initramfs, so a driver that fails to initialise leaves you with no display at all and a rebuild through sudo limine-mkinitcpio from a rescue shell. Check pacman -Q nvidia-open-dkms before writing anything into /etc/modprobe.d.

Symptom

The machine hard-freezes seconds after the screen blanks on idle, or right after resuming: the mouse is dead, there is no TTY switch, and only a power-button reset works. journalctl -b -1 -k shows:


NVRM: _kgspLogXid119: ***** GSP Timeout *****
NVRM: Xid (PCI:0000:01:00): 119, Timeout after 6s of waiting for RPC response from GPU0 GSP! Expected function 76 (GSP_RM_CONTROL) sequence 1321

Cause

During DPMS off and suspend or resume transitions the GPU can drop into a very low clock state that the GSP firmware does not recover from. The RPC to the GSP microcontroller then times out, the driver wedges, and the whole machine goes with it. Only GPUs that carry GSP firmware are exposed, which means Turing and newer, and on Omarchy 4 that is the set of cards the installer gives nvidia-open-dkms to. Pre-Turing cards get nvidia-580xx-dkms and cannot hit this.

Fix

Rule out the Omarchy 4 look-alike first, because it is far more common right now and locking clocks will not touch it. Since aquamarine 0.15.0 landed with Omarchy 4.0.3 on 2026-09-09, an external display that is powered off, unplugged or allowed to sleep can come back as "No Signal" and never recover, on nvidia-drm as well as i915, xe and amdgpu. From the chair that looks like this freeze. The machine is not wedged:


pacman -Q aquamarine
hyprctl monitors | grep -E 'Monitor|dpmsStatus'
journalctl -b -k | grep -iE 'GSP Timeout|Xid \(PCI'

If you can still ssh in and hyprctl monitors answers, it is the aquamarine regression and not this record. See https://github.com/omacom/omarchy/issues/11249.

A real Xid 119 GSP timeout needs a GPU that has GSP firmware, which means Turing or newer. Omarchy installs nvidia-open-dkms on exactly those cards and nvidia-580xx-dkms on Maxwell, Pascal and Volta, and ships a detector for the split:


omarchy-hw-nvidia-gsp && echo "GSP capable" || echo "no GSP firmware, this record does not apply"

Then pin a higher minimum GPU and memory clock so the GPU never reaches the unstable state.


sudo systemctl enable --now nvidia-persistenced.service

# find valid clock values for your card
nvidia-smi -q -d SUPPORTED_CLOCKS
nvidia-smi -q -d CLOCK

# temporary test, adjust the upper bounds to your GPU's real max
sudo nvidia-smi -lgc 800,2100
sudo nvidia-smi -lmc 800,10000

# revert with
sudo nvidia-smi -rgc
sudo nvidia-smi -rmc

If that stops the freezes, make it permanent:


sudo tee /etc/systemd/system/nvidia-clocks.service >/dev/null <<'EOF'
[Unit]
Description=Set NVIDIA GPU minimum clocks to avoid GSP timeouts
Requires=nvidia-persistenced.service
After=nvidia-persistenced.service

[Service]
Type=oneshot
ExecStart=/usr/bin/nvidia-smi -lgc 500,2100
ExecStart=/usr/bin/nvidia-smi -lmc 500,10000
RemainAfterExit=yes

[Install]
WantedBy=multi-user.target
EOF

sudo systemctl enable --now nvidia-clocks.service

Lower the 500 floor to reduce idle power draw. Too low and the freeze comes back.

**The idle stopgap, on Omarchy 4.** There is no hypridle and no ~/.config/hypr/hypridle.conf. Neither hypridle nor hyprlock is installed, and the Omarchy shell never issues a DPMS off: on idle it launches a screensaver and then locks. The two knobs are idle.screensaver and idle.lock in ~/.config/omarchy/shell.json, both counted in seconds from when idle began:


{
  "idle": {
    "screensaver": 150,
    "lock": 300
  }
}

To stop the machine idling at all for this session:


omarchy-toggle-idle stay-awake

If the panel still powers off after that, it is the monitor's own standby timer or a system suspend, not a Hyprland setting.

**On plain Arch with hypridle installed**, raising or removing the dpms off timeout in ~/.config/hypr/hypridle.conf is still the stopgap.

Verify

Confirm the unit came up and the floor is applied:


systemctl is-active nvidia-clocks.service
nvidia-smi -q -d CLOCK

Then leave the machine idle past idle.screensaver in ~/.config/omarchy/shell.json, come back and wake it. The desktop returns, and the log of the boot you are still in is clean:


journalctl -b -k | grep -iE 'GSP Timeout|Xid \(PCI'

Use journalctl -b -1 -k instead only when you are reading back a boot that actually froze and had to be power-cycled.

Sources

nvidia-dpms-gsp-timeout-freeze