You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[Windows Desktop] Repeated GPU process crash (exitCode=101457950) after update, AppX breaks into NeedsRemediation state #35566
GPU: (user's GPU info from their machine, check dxdiag equivalent)
CLI: codex-cli 0.146.0-alpha.3.1
Symptoms
Codex works for a while, then suddenly the GPU process crashes
After the crash, Codex cannot be relaunched — clicking the icon does nothing
AppX package status becomes \Modified, NeedsRemediation\
The \codex://\ protocol handler becomes broken
Re-registering (\Add-AppxPackage -Register) and resetting (\Reset-AppxPackage) from Windows Settings temporarily fix it, but the crash returns after some use
Before the GPU crash, the log shows many:
\
browser-sidebar-comment-mode-site-status failed to load
\
and occasional \ERR_CONNECTION_TIMED_OUT\ / \ERR_NETWORK_CHANGED\ on /wham/usage\ and /settings/user.
After the GPU process fails to restart, Windows destroys the AppX container (Event ID 217) and the package moves to \NeedsRemediation.
Independent reproduction with the same Store build and exact GPU exit codes. This case broadens the trigger beyond media/WebCodecs or Cloudflare pages: an ordinary persisted Bing search tab in the in-app browser was enough to reproduce it.
Windows destroyed the container at local time 2026-07-27 12:00:11 (UTC+8).
Startup loop / AppX behavior
After the browser-triggered termination, Codex repeatedly created a process but exited during very early initialization. The desktop logs for many launch attempts contained only Launching app and sometimes Appshot hotkey inactive; there was no standard Application Error event and no Crashpad report.
Microsoft Store reported NumberOfApplicableUpdates = 0 before repair. Windows then repaired/re-registered the same package version and logged:
Trying to repair ACLs for ...\WindowsApps\OpenAI.Codex_26.721.4979.0_x64__2p2nqsd0c76g0
ACLs repaired successfully
AppModel Runtime later changed the package status to 0x0 from 0x2. Repair/re-stage operations occurred repeatedly and only restored launchability temporarily.
The persisted in-app browser state appears important to the repeated crash-after-launch behavior: the same browser tab ID and Bing URL were restored before the second GPU crash. The low-level failure is strongly correlated with the Chromium GPU process, while the RDP/multiple-virtual-display environment may be an additional risk factor.
Only redacted excerpts are included; no conversation contents, local project paths, or credentials are attached.
get codex crashes..
when it crashes i have to repair the install using the ms store.
i asked the agent to check the logs and report here is its anaylsis.
Four identical crashes occurred at 11:52:28, 11:55:29, 12:01:13, and 12:05:32 PM:
The in-app browser loaded dash.cloudflare.com.
The GPU process crashed with exit code 101457950.
GPU relaunch failed with exit code 18.
Chromium terminated with: GPU process isn't usable. Goodbye.
The crash telemetry also reported a locked DawnGraphiteCache file. Combined with the VMware/Remote Display graphics stack, this points to an embedded-browser hardware-acceleration/cache conflict. It was not an out-of-memory event or a Windows-wide graphics reset.
Activity
github-actions commentedon Jul 27, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
5y50541 commentedon Jul 27, 2026
Independent reproduction with the same Store build and exact GPU exit codes. This case broadens the trigger beyond media/WebCodecs or Cloudflare pages: an ordinary persisted Bing search tab in the in-app browser was enough to reproduce it.
Environment
26.721.4979.0x640.146.0-alpha.3.130.0.14.7444(2023-05-11)27.20.100.9664(2021-06-01)Reproduction timeline
Reproduction 1
The app restored/created an in-app browser sidebar webview for a normal Bing search URL. The browser reached DOM ready, then the GPU process crashed:
Windows destroyed the Desktop AppX container at local time
2026-07-27 03:06:07(UTC+8), immediately after the GPU restart failure.Reproduction 2
The same persisted browser tab and URL were automatically restored on the next affected launch:
Windows destroyed the container at local time
2026-07-27 12:00:11(UTC+8).Startup loop / AppX behavior
After the browser-triggered termination, Codex repeatedly created a process but exited during very early initialization. The desktop logs for many launch attempts contained only
Launching appand sometimesAppshot hotkey inactive; there was no standard Application Error event and no Crashpad report.Microsoft Store reported
NumberOfApplicableUpdates = 0before repair. Windows then repaired/re-registered the same package version and logged:AppModel Runtime later changed the package status to
0x0from0x2. Repair/re-stage operations occurred repeatedly and only restored launchability temporarily.The persisted in-app browser state appears important to the repeated crash-after-launch behavior: the same browser tab ID and Bing URL were restored before the second GPU crash. The low-level failure is strongly correlated with the Chromium GPU process, while the RDP/multiple-virtual-display environment may be an additional risk factor.
Only redacted excerpts are included; no conversation contents, local project paths, or credentials are attached.
neonglobe commentedon Jul 27, 2026
get codex crashes..
when it crashes i have to repair the install using the ms store.
i asked the agent to check the logs and report here is its anaylsis.
Four identical crashes occurred at 11:52:28, 11:55:29, 12:01:13, and 12:05:32 PM:
The in-app browser loaded dash.cloudflare.com.
The GPU process crashed with exit code 101457950.
GPU relaunch failed with exit code 18.
Chromium terminated with: GPU process isn't usable. Goodbye.
The crash telemetry also reported a locked DawnGraphiteCache file. Combined with the VMware/Remote Display graphics stack, this points to an embedded-browser hardware-acceleration/cache conflict. It was not an out-of-memory event or a Windows-wide graphics reset.
asully-oai commentedon Aug 6, 2026
this is likely a dup of #35352