Stop Chromium's GPU process crashing repeatedly on NVIDIA
Checked on this workstation, which has an NVIDIA RTX 3090 with nvidia-open-dkms 610.57.04-1 and nvidia_drm loaded, omarchy 4.0.2-1, Hyprland 0.56.2, and chromium 151.0.7922.173-1, the exact Chromium build the cited issue names. Three things held. The flag file mechanism is real: /usr/bin/chromium is Arch's launcher and its embedded help text says custom flags are read from /etc/chromium-flags.conf and $XDG_CONFIG_HOME/chromium-flags.conf, so the record is not writing flags to a file nothing reads. The two ANGLE switches exist in the shipped binary (strings on /usr/lib/chromium/chromium matches use-angle, use-cmd-decoder, and the values vulkan and passthrough), and the Arch wiki section Running on Xwayland recommends exactly --use-angle=vulkan --use-cmd-decoder=passthrough for NVIDIA GPU-process crashes. coredumpctl list and journalctl -b -k both run unprivileged here, so the diagnostic commands work as written. Four things were wrong. First, the cause states as fact what the issue reporter offered as a guess. omacom/omarchy#8394 is still OPEN as of 2026-09-13 with no root cause, and the only reproduction comment on the thread (btownsend, on 4.0.2-1 with the same Chromium build) gets the identical SIGTRAP under X11 and with extensions disabled, and isolates the common factor as the --oauth2-client-id and --oauth2-client-secret lines that /usr/share/omarchy/bin/omarchy-install-chromium-google-account appends to ~/.config/chromium-flags.conf. It is not established to be a GPU or Ozone fault. Second, the record presents --ozone-platform=x11 as an alternative that also avoids the crash, which is the original reporter's table but is directly contradicted by that reproduction, so the record must not recommend it as reliable. Third, the record says add to ~/.config/chromium-flags.conf without noting that on Omarchy that file already exists, seeded from /usr/share/omarchy/config/chromium-flags.conf, and already contains --ozone-platform=wayland and --ozone-platform-hint=wayland, so appending --ozone-platform=x11 leaves two contradictory lines (base::CommandLine::AppendSwitchNative does switches_[key] = value so the last wins for lookups, but both stay in argv_ and are passed to child processes, and the honest instruction is to edit the existing line). Fourth, the record does not say that on a stock Omarchy install Chromium is already native Wayland, so the XWayland half only applies once the user has moved it to X11. Added to verify a check that the flag was actually read, because chrome://version lists the full command line. The cited issue URL uses the old basecamp/omarchy name, replaced with omacom/omarchy. NOT exercised: I did not launch Chromium, did not reproduce the SIGTRAP, and did not test the ANGLE flags, because this is the operator's daily workstation.
--disable-gpu turns off hardware acceleration for the whole browser: expect higher CPU use and no hardware video decode. Treat it as a stopgap. --ozone-platform=x11 moves Chromium back onto XWayland, which is the condition the ANGLE flags above exist to work around, and one reproducer on omacom/omarchy#8394 reports it does not stop the crash. Deleting the --oauth2-client-* lines removes Chromium's ability to sign in to a Google account until you re-run omarchy-install-chromium-google-account.
Symptom
Chromium's GPU process crashes repeatedly on NVIDIA. Tabs go blank or show "Aw, Snap!", chrome://gpu reports the GPU process restarting. On Omarchy some users additionally see Chromium die with SIGTRAP (int3) at a fixed offset when a new window is spawned for an OAuth popup, in a ~20-crash loop, while the same flow works on KDE Plasma.
Cause
Two different problems share the symptom.
The XWayland GPU-process crashes are Chromium's ANGLE and GL backend selection failing on the proprietary NVIDIA driver, and they only apply when Chromium is genuinely running on XWayland. Omarchy seeds ~/.config/chromium-flags.conf from /usr/share/omarchy/config/chromium-flags.conf, which sets --ozone-platform=wayland, so on a stock Omarchy install Chromium is native Wayland and this half does not apply unless the user has switched it to X11.
The SIGTRAP (int3) crash loop reported in omacom/omarchy#8394 has no established cause. The reporter suspected Ozone and Wayland GPU initialisation, but the only reproduction on the thread sees the same SIGTRAP under --ozone-platform=x11 and with extensions disabled, and isolates the common factor as the --oauth2-client-id and --oauth2-client-secret lines that omarchy-install-chromium-google-account appends to ~/.config/chromium-flags.conf. The issue was still open on 2026-09-13.
Fix
On Omarchy ~/.config/chromium-flags.conf already exists and is seeded from /usr/share/omarchy/config/chromium-flags.conf. Append to it, and where you are contradicting a line that is already there, edit that line instead of adding a second copy. Chromium's launcher reads /etc/chromium-flags.conf and then ~/.config/chromium-flags.conf into one command line, and a duplicated switch is both confusing and unreliable across child processes.
**XWayland GPU-process crashes on NVIDIA.** Only relevant if Chromium is actually on X11 or XWayland:
--use-angle=vulkan
--use-cmd-decoder=passthrough
The Arch wiki attaches a note to these: they do not prevent all XWayland-related crashes.
**The SIGTRAP (int3) crash loop on Hyprland (omacom/omarchy#8394).** The workaround the issue reports as working is to take the GPU out of the path:
--disable-gpu
Do not rely on --ozone-platform=x11 here. The original reporter listed it as working, but the one person who reproduced the bug saw the same SIGTRAP under X11. If you try it anyway, change the existing --ozone-platform=wayland line rather than adding a second one, and remove --ozone-platform-hint=wayland at the same time.
If you added Google sign-in with omarchy-install-chromium-google-account, that reproduction points at the OAuth flags it appended, and removing them is the narrowest test:
sed -i '/^--oauth2-client-/d' ~/.config/chromium-flags.conf
Flags are only read at process start, so quit Chromium entirely and relaunch:
pkill chromium
Check what actually crashed:
coredumpctl list chromium
journalctl -b -k | grep -i 'trap int3'
Verify
chrome://gpu no longer shows the GPU process restarting, and the OAuth login popup opens instead of crash-looping. Confirm the flag was actually read before concluding anything: chrome://version prints the full command line, and the flag you added must appear in it.