My PC occasionally freezes while gaming, so far only while exiting Halo Campaign Evolved. The latest freeze ended with a KMODE_EXCEPTION_NOT_HANDLED (0x1E) BSOD. After restarting, the system entered BIOS because Windows Boot Manager was missing.
The Samsung 990 Pro 2TB was still detected by BIOS and the Windows installer, but DiskPart showed no partitions. From Fedora, TestDisk and gdisk reported that the main GPT header and partition table were invalid while the backup GPT was intact. The backup still contained the correct 100 MB EFI partition and the Windows partition. Rebuilding and writing the main GPT from the valid backup restored the drive, and Windows booted normally.
This has happened multiple times. Samsung's full diagnostic scan found no bad blocks or errors, the SSD firmware is current, and SMART reports the drive as healthy. BIOS and Intel ME firmware are updated. MemTest86 completed four passes both at 4800 MT/s and 6000 MT/s with XMP enabled and found no errors. DISM, SFC, Cinebench, OCCT 3D, and OCCT VRAM testing have also been completed without revealing a problem.
The system uses an MSI B760 Gaming Plus WiFi motherboard, an Intel i7-14700K, an RTX 4080 Super, and the Samsung 990 Pro. I plan to move the SSD from the CPU-connected M2_1 slot to the chipset-connected M2_2 slot to test whether the current PCIe path, slot, or motherboard connection is involved.
Has anyone encountered an NVMe drive intermittently hanging or timing out, corrupting only the primary GPT while leaving the backup GPT usable, even though SMART and vendor diagnostics report the SSD as healthy? What should I investigate next?
3 Answers
The slow-I/O timeout records are more interesting than the clean SMART result. SMART and a full diagnostic scan can miss intermittent controller, PCIe-link, motherboard-slot, power-management, or driver failures because they mostly test the drive itself. Moving the 990 Pro to the other M.2 slot is a worthwhile isolation test. Also check whether the problem follows the SSD or stays with the original slot, and collect the minidump plus any WHEA-Logger events. If it happens again, back up important data immediately and consider testing with a different NVMe drive or replacing the affected slot or drive rather than repeatedly repairing the GPT.
A 100 MB EFI System Partition is smaller than current Windows recommendations. It usually won't explain a GPT header being destroyed, but it can cause trouble during some updates, so expanding or recreating it to around 260–300 MB would be sensible if you are already rebuilding the layout. BitLocker can also complicate recovery after a crash, although it should not be the main suspect if it is disabled.
The most useful next step is to inspect the Windows minidump rather than treating the GPT damage as the original failure. A KMODE_EXCEPTION_NOT_HANDLED crash can be caused by a driver, storage timeout, firmware problem, or unstable hardware. Check C:WindowsMinidump and analyze any dumps with WinDbg. The failing module and storage-related black-box data may help identify whether the NVMe path was already timing out before the crash.
BitLocker is completely disabled. I also checked the crash dump: WinDbg's !blackboxntfs output showed two slow-I/O timeout records, including one on the same thread that crashed. However, Event Viewer has no stornvme, storport, or disk errors, and Samsung Magician still reports the SSD as healthy.