I built this Windows 11 PC around April last year, and it was stable for months before recently developing random blue screens and game crashes. The system has a Ryzen 7 7700X, RTX 5070, MSI B650M motherboard, 32GB G.Skill DDR5-6000 running with EXPO, a 2TB WD Blue NVMe SSD, and an 850W power supply. The crashes began around the time I installed Adobe After Effects from an unofficial source, although that may be unrelated.
I have seen several different stop codes, including IRQL_NOT_LESS_OR_EQUAL, CLOCK_WATCHDOG_TIMEOUT, DPC_WATCHDOG_VIOLATION, KMODE_EXCEPTION_NOT_HANDLED, and KERNEL_MODE_HEAP_CORRUPTION. Some crashes happen while gaming, others while the system is idle, and certain games occasionally close to the desktop without producing a BSOD.
WinDbg analysis has not identified one consistent third-party driver. Several dumps end in ntoskrnl.exe functions such as KiSwapContext, KiIdleLoop, KiTryUnwaitThread, and AuthzBasepDuplicateSecurityAttributes. SFC initially found corruption and DISM had trouble repairing Windows, so I performed an in-place Windows repair installation. SFC now reports no integrity violations. CrystalDiskInfo reports the NVMe drive at 99%, Memory Integrity is enabled, there is no manual CPU or GPU overclock, and Event Viewer has not shown any WHEA-Logger errors. The recurring DistributedCOM and WMI warnings appear unrelated, while Kernel-Power entries only occur after crashes.
The system still experiences game crashes, and I recently received another IRQL_NOT_LESS_OR_EQUAL BSOD after gaming. I have not yet completed extended memory testing, tried running the RAM at standard JEDEC settings instead of EXPO, performed a clean GPU driver installation, or updated the BIOS.
Does this pattern suggest unstable RAM, a CPU or motherboard problem, or a driver issue? Should I disable EXPO first, and are there other important tests I should run?
3 Answers
The NVMe drive is probably not the main suspect, but check its firmware rather than relying on the 99% health number. For NVMe drives, that percentage mostly reflects estimated wear and does not prove the drive is healthy. Look specifically for critical warnings, media or data integrity errors, and any manufacturer firmware updates. Also consider a clean graphics-driver reinstall, but the broad range of crashes and the watchdog timeout still make RAM, EXPO stability, or the CPU's memory controller more likely than a single GPU driver.
The mix of unrelated stop codes makes memory instability a strong possibility. Run a proper extended test such as MemTest86 or TestMem5, and also try the system with EXPO disabled at the motherboard's standard JEDEC speed. If it becomes stable at default settings, that points toward the memory kit, the integrated memory controller, or the board's memory training—not necessarily a permanently defective component. Since there is only one DIMM, testing sticks individually is not an option, so changing memory settings and running longer tests becomes more important.
Thanks, I’ll run an extended memory test and compare the results with EXPO enabled and disabled.
CLOCK_WATCHDOG_TIMEOUT deserves special attention because it usually means a CPU core stopped responding. The memory controller is built into the processor, so a CPU or memory-controller problem can also look like bad RAM. Memory testing is still worthwhile, but it is not always conclusive for intermittent DDR4/DDR5 problems. Try running the machine at stock memory settings, update the BIOS and AMD chipset drivers, and check whether the crashes continue. If possible, testing with another known-good memory kit would be more useful than relying on dump files alone.
The absence of WHEA errors does not completely rule out the CPU or memory subsystem. Intermittent instability can still produce different kernel-level stop codes without a clean hardware error entry.
The timing around the unofficial software is worth considering too. Remove it completely and run a malware scan, since modified installers can add drivers or other software, but the hardware stability checks should happen regardless.