My custom-built Windows 11 PC has become almost unusable due to frequent blue screens, browser and game crashes, and hangs during restarts. The errors include MEMORY_MANAGEMENT, IRQL_NOT_LESS_OR_EQUAL, PFN_LIST_CORRUPT, SYSTEM_SERVICE_EXCEPTION, UNEXPECTED_STORE_EXCEPTION, PAGE_FAULT_IN_NONPAGED_AREA, and STATUS_BREAKPOINT.
The system often freezes on the crash screen or restarts to a black screen while the fans and power lights remain on. A red DRAM diagnostic LED appears during these restart hangs, although cold boots sometimes work normally.
The hardware is a Ryzen 7 7800X3D, Gigabyte B650 Gaming X AX revision 1.3, 32 GB Corsair Vengeance DDR5 using two 16 GB modules, an RTX 4080, and BIOS version FC4c. I have disabled EXPO and all overclocking, updated the BIOS, cleared CMOS, reseated the memory, and cleaned dust from the DIMM area.
My testing has been inconsistent. Both modules produce MemTest86 errors in the A2 slot, but the number varies dramatically between runs. One module produced thousands of errors, while another test of the same module and slot completed four passes with zero errors. Windows still crashed shortly afterward. The second module produced hundreds to thousands of errors in A2 and dozens of errors in B2. Neither module would POST in A1 or B1, with the DRAM LED immediately lighting up.
Because both modules show errors and the primary slots sometimes fail to POST, I am unsure whether the problem is defective memory, a DIMM-slot or motherboard fault, poor CPU/socket contact, or the Ryzen memory controller. I do not currently have known-good DDR5 memory available and would prefer not to replace parts without better evidence. What would be the best way to identify the failed component and proceed with warranty service?
4 Answers
The different BSOD codes are consistent with corrupted memory, but dump files can help confirm whether a driver or storage device is also involved. If Windows stays up long enough, copy C:WindowsMinidump and configure small memory dumps for future crashes. Still, software analysis cannot explain thousands of MemTest errors, so resolve the hardware instability first. Avoid using the machine for important work until backups are verified, because repeated crashes can corrupt the Windows installation or files.
MemTest errors are not normal, even if the count changes between runs. A pass followed by failures does not clear the memory because intermittent faults can depend on temperature, training, contact, or the exact test pattern. Test one module at a time in the motherboard's recommended single-DIMM slot, normally A2, and let it run for several complete passes or overnight. If a module produces even one repeatable error at stock settings, treat the memory subsystem as unstable and pursue an RMA rather than trusting the occasional clean run.
The fact that neither module can POST in A1 or B1 is worth investigating, but do not assume those slots are defective until checking the manual and testing with a known-good module. A single DDR5 module normally belongs in A2, and a two-module kit normally belongs in A2 and B2; the other slots may have different training behavior, but they should not automatically be treated as proof of failure. Since both DIMMs show errors in A2/B2, the motherboard, CPU memory controller, socket contact, or a damaged pin could also be responsible.
Remove the CPU cooler and inspect the AM5 socket carefully for bent or contaminated pins, then reseat the CPU and reinstall the cooler evenly without excessive pressure. Check that the 24-pin motherboard and CPU EPS power connectors are fully seated. AM5 memory-channel problems can come from socket contact or mounting pressure, and that can look exactly like bad RAM. If the board is still under warranty, document the MemTest errors, BIOS defaults, slot behavior, and DRAM LED symptoms before opening an RMA.
That is what makes this difficult: each module has produced errors, but the count changes drastically and one run passed completely. I am concerned that replacing only the RAM could miss a board or CPU memory-controller problem.