After wiping all of my SSDs and reinstalling Windows 11 from USB, my PC began having intermittent grey-screen recoveries, mainly while using Chrome, Discord, or starting YouTube videos. The display usually recovers without closing my applications. Reliability Monitor recorded AMD watchdog events involving amdfendr.sys, amdkmdag.sys, dxgkrnl.sys, and dxgmms2.sys.
Several days later, the system logged a fatal WHEA 0x124 Machine Check Exception on CPU/APIC 0, MCA Bank 5, with MCi_STATUS 0xBEA0000000000108. The report appears more consistent with a CPU execution-unit, cache, Infinity Fabric, memory-controller, or power-stability problem than a direct PCIe graphics-card error. It did not include a PCIe error section.
The system uses a Ryzen 7 5700X3D, Radeon RX 7900 GRE, MSI MPG B550 Gaming Plus, 32 GB of Kingston DDR4-3200, and a roughly two-year-old 750 W PSU. BIOS settings are stock apart from XMP, with no manual overclock, undervolt, PBO, Curve Optimizer, or Kombo Strike tuning. The system runs Windows 11 Pro with Radeon Adrenalin 26.7.1 and uses two monitors.
The crashes have not happened during gaming so far. Case-fan lighting also occasionally flickers, although I do not know whether that indicates a power problem or is unrelated. Could the Radeon watchdog recovery and later WHEA/MCA Bank 5 error be symptoms of the same underlying instability, and what should I test first to determine whether the CPU, memory settings, motherboard, PSU, or graphics driver is responsible?
1 Answer
The WHEA details are worth investigating separately from the Radeon driver recovery. Ryzen 5000 systems can show voltage or stability problems, and the fact that this is an X3D chip does not completely rule that out. A hardware and event-log report containing the decoded WHEA records and complete system information would help determine whether this resembles known CPU or platform-voltage failures.
I gathered the system and WHEA information. I am worried because some people think the processor may be failing, while others think it could still be a configuration or platform issue. The timing is especially unfortunate because the warranty may have expired only a few days ago.