[BUG] Desktop app (Windows): near-daily GPU process crashes (exitCode 101457950) + RADAR_PRE_LEAK_64 memory leak report

Environment

• Claude Desktop App v1.30096.1 (packaged), Windows 11 Home 10.0.28120
• Hybrid graphics: Intel Arc Graphics (driver 32.0.101.6881, 2025-06-10) + NVIDIA GeForce RTX 4060 Laptop (driver 32.0.15.7283, 2025-03-14)
• 32 GB RAM
• Heavy usage profile: 11 scheduled tasks (cron jobs), up to 8 claude.exe processes in parallel, large sessions (JSONL transcripts up to 30 MB, project folder 263 MB)

What happened

The desktop app freezes / dies almost daily and has to be restarted.

%APPDATA%\Claude\logs\main.log shows a recurring pattern — a GPU process crash, immediately followed by an app restart:

2026-08-15 08:46:52 [info] GPU process gone: {
  type: 'GPU',
  reason: 'crashed',
  exitCode: 101457950,
  serviceName: 'GPU'
}
2026-08-15 08:48:21 [info] Starting app { appVersion: '1.30096.1', isPackaged: true, ... }

Occurrences: 2026-08-11, 08-12, 08-13, 08-14 (3x), 08-15 — always exitCode 101457950.

Additionally, Windows Error Reporting logged a memory-leak report for the app:

• 2026-08-14 21:11 — RADAR_PRE_LEAK_64, P1: claude.exe, P2: 1.30096.1.0 (Windows memory-leak detection)

Once also observed (2026-08-05):
Network Service process gone (reason=crashed, exitCode=-1073741205) — main-process networking is broken until relaunch

Notably, there are no Event ID 1000/1002 entries for claude.exe in the Windows Application event log — the failures are Chromium/Electron GPU-process crashes, not native faults.

Suspected factors

• Hybrid Intel Arc + NVIDIA graphics with hardware acceleration (Electron GPU process)
• Memory growth with very large sessions / many parallel headless sessions

Possibly related

Similar memory-related reports: #45889, #42962, #25545 — this report adds the recurring GPU-process crash signature on hybrid Intel Arc + NVIDIA hardware.

What I can provide on request

• Full main.log excerpts
• Windows Error Reporting details for the RADAR_PRE_LEAK_64 report
• /heapdump diagnostics JSON