Windows: GPU process crash (exitCode 101457950 / 0x60C201E) kills the whole app - reproduces idle, single GPU, current driver; correlates with display-mode re-enumeration; Crashpad captures nothing

Environment

• Claude Desktop appVersion 1.26832.0 (isPackaged, win32 x64, Node 24.18.0)
• Windows 11 Pro, build 10.0.26200
• GPU: NVIDIA GeForce RTX 4080 SUPER (single GPU, desktop), driver 32.0.16.1088 = 610.88 Game Ready — current as of filing
• Multi-monitor desktop, Samsung Display Manager installed
• Typical load: 3–5 concurrent local sessions + MCP servers + the in-app Browser pane open

Summary

The app's Chromium GPU process crashes with the same exit code every time and takes the entire app down with it — 4 times in 2 days, killing all running local sessions each time. It reproduces while the app is fully idle (overnight crash with no interaction) and on the current NVIDIA driver, on a single-GPU machine.

Signature in %APPDATA%\Claude\logs\main.log (identical every occurrence):

GPU process gone: {
  type: 'GPU',
  reason: 'crashed',
  exitCode: 101457950,   // 0x60C201E
  serviceName: 'GPU'
}

Occurrences

| GPU crash (local, UTC-6) | App restart | Notes |
|---|---|---|
| 2026-08-10 08:54:30 | 09:07:19 | Windows Display event id 4107 (SetDisplayConfig SDC_FORCE_MODE_ENUMERATION) at 09:00:56 |
| 2026-08-11 00:16:56 | 00:19:46 | App idle overnight. Display event id 4107 at 00:18:36 |
| 2026-08-11 15:11:49 | 15:14:21 | Active use; no display event |
| 2026-08-11 15:26:58 | 15:29:57 | 15 min after the previous one; [process-memory] seconds earlier: tree_rss_sum=2127MB … sys_free=46659MB/65325MB |

Log excerpts

Crash #2 (idle) and #4 (with the memory line ruling out pressure):

2026-08-11 00:16:56 [info] [oauth] using cached token for orgId=…
2026-08-11 00:16:56 [info] GPU process gone: {
  type: 'GPU',
  reason: 'crashed',
  exitCode: 101457950,
  serviceName: 'GPU'
}
2026-08-11 00:19:46 [info] Starting app {

2026-08-11 15:26:58 [info] [process-memory] trigger=interval tree_rss_sum=2127MB electron(12)=2127MB children(0)=0MB top=[electron_renderer:16656:557MB electron_main:13984:295MB electron_gpu:22996:207MB electron_renderer:12832:164MB electron_utility:17464:155MB] sys_free=46659MB/65325MB sys_free_raw=46659MB
2026-08-11 15:26:58 [info] GPU process gone: {
  type: 'GPU',
  reason: 'crashed',
  exitCode: 101457950,
  serviceName: 'GPU'
}
2026-08-11 15:29:57 [info] Starting app {
  appVersion: '1.26832.0',
  isPackaged: true,
  platform: 'win32',
  arch: 'x64',
  nodeVersion: '24.18.0'
}

Full main.log available on request.

Ruled out locally

• Memory pressure: 2.1 GB app / 46 GB system free seconds before a crash (excerpt above).
• Driver TDR: zero nvlddmkm / Dxgkrnl / Event 4101 entries in the System log across the whole window.
• Stale driver: 610.88 is the newest available at filing; #81698 reproduced on adjacent 610.47.
• Native crash with telemetry: no WER report, no Application-log Event 1000, and %APPDATA%\Claude\Crashpad contains only settings.dat — no dumps. If Crashpad is meant to capture these, it isn't; this crash class may be invisible to your telemetry.

Correlations (evidence, not proven cause)

1. Display-mode re-enumeration: 2 of 4 crashes sit within ~2–6 minutes of Windows Display event id 4107 (SetDisplayConfig with SDC_FORCE_MODE_ENUMERATION — monitor sleep/wake or display-management software forcing mode enumeration). I don't believe any existing issue connects these two signals.
2. In-app Browser pane open during at least the two 2026-08-11 afternoon crashes — consistent with reports tying GPU crashes to the browser/preview surfaces (#81664, #83478).

Relationship to existing issues

Same verbatim exitCode: 101457950 signature: #81698 (RTX 5080 Laptop, WebGL INVALID_ENUM pre-crash), #80468 (bug-labeled, hybrid graphics, Turnstile trigger), plus #83478, #68049 (persists across driver reinstall), #83028 (Intel iGPU), #81836, #81159, #82967. This report adds: idle-crash reproduction, single-GPU reproduction (hybrid-adapter renegotiation is not a necessary condition), memory/TDR ruled out with evidence, the display-event correlation, and the empty-Crashpad observation. #77857 (no way to disable hardware acceleration) is the missing mitigation — affected users currently have no supported way to take the GPU process out of the loop.

Asks

1. Is exitCode 101457950 (0x60C201E) a known GPU-process exit signature for this Electron build? It matches no documented Chromium exit code.
2. A supported hardware-acceleration toggle (#77857) would give every issue in this cluster an immediate mitigation.
3. Confirm whether Crashpad should be capturing these — locally it captures nothing, so the crash class may be under-reported in telemetry relative to its real frequency.