GPU process crashes recurring since v2.1.237, still present in v2.1.246 (Windows, hybrid NVIDIA/Intel graphics laptop)
Title
GPU process crashes recurring since v2.1.237, still present in v2.1.246 (Windows, hybrid NVIDIA/Intel graphics laptop)
Environment
- OS: Windows 11 Home (10.0.26200)
- Claude Code desktop app, MSIX install
- GPU: NVIDIA GeForce RTX 5080 Laptop GPU + Intel integrated graphics (hybrid/switchable graphics laptop)
- Affected versions: 2.1.237 through 2.1.246 (current at time of writing)
Summary
Claude Code's Electron GPU process dies (GPU process gone: { reason: 'crashed' } in main.log) roughly every 10–30 minutes, unpredictably, regardless of what's on screen or which feature is in use. Each occurrence freezes/kills the whole app; in several cases the app did not recover on its own for 15–70+ minutes until manually relaunched.
Timing evidence (from local main.log)
- App was updated to v2.1.237 at 11:57:40 AM on the day this started.
- The first
GPU process gone: reason: crashedentry in the log appears 51 minutes later, at 12:48:33 PM the same day. - Crashes continued recurring across multiple days on v2.1.237, at roughly 10–30 minute intervals.
- App was updated again to v2.1.246 mid-investigation — crashes continued afterward too (two more
GPU process gone: reason: crashedentries logged after the update). - No prior working history is visible in the log (log file itself only goes back a few days), but the user reports ~2 months of prior stable daily use with no crashes before this started — consistent with the crash onset lining up with the 2.1.237 update rather than a pre-existing hardware issue.
What was already tried
- Updated the NVIDIA GPU driver from a build over a year old to the current Studio Driver release. Confirmed via
Get-CimInstance Win32_VideoControllerthat the driver version changed and the install succeeded. A crash still recurred ~19 minutes after the driver update completed, so this was not the root cause (though worth keeping as a general improvement). - Disabled Wi-Fi adapter power-saving (unrelated network-drop issue found during the same investigation, not the GPU crash — noted here only for completeness in case it's relevant).
Suspected cause
A regression introduced around v2.1.237 affecting GPU-process stability, possibly specific to hybrid/dual-GPU (integrated + discrete, Optimus-style) Windows laptops.
Request
Please investigate GPU-process crash stability starting around v2.1.237, particularly on hybrid-graphics Windows laptops (integrated + discrete GPU). Happy to provide full main.log excerpts / additional diagnostics if useful — log location: %LOCALAPPDATA%\Claude\logs\main.log.