Cleaning Up a Mystery AM4 Build on a Day Off
A colleague handed me his son's desktop on my day off with a classic set of symptoms: no video output over HDMI or DisplayPort, and the keyboard LEDs would flash for a second or two before going completely dark. The optical sensor on the mouse stayed lit, the case fans spun up, and the RGB bar on the Sapphire Radeon RX 6600 XT powered on, but the system refused to pass POST.
When you open up a rig that someone else assembled, you usually find a mix of obvious physical configuration errors and deeper hardware unknowns. The goal is to strip away every layer-1 mistake you can verify by hand without crossing the line into invasive teardown work when you don't have spare bench parts on hand.
Initial Inspection and Layer-1 Fixes
Opening the side panel on the ASRock B450M/ac R2.0 revealed two immediate hardware mismatches that would stop a clean boot on their own:
- Improper Memory Channel Population: Both sticks of
XPG DDR4RAM were installed side-by-side in slotsA1andA2, putting both modules on the same memory channel. On Ryzen AM4 boards, populatingA1/A2instead of the primaryA2/B2pair (the second and fourth slots from the CPU cooler) frequently halts memory training before video initialization. - SATA M.2 Drive in an NVMe-Only Slot: Seated in the board's
Ultra M.2(M2_1) slot was a 64GBVaseky V820M.2 drive. Pulling the drive out confirmed it was aB+Mkey SATA M.2 module. On theB450M/ac, theUltra M.2slot is wired directly to the CPU's PCIe Gen3 x4 lanes and does not support SATA M.2 drives at all. Leaving a SATA M.2 drive inside an NVMe-only slot will never register as a boot device and can hang PCIe bus initialization.
Alongside the tower, my colleague had a boxed 1TB Predator GM7 PCIe Gen4 x4 NVMe SSD waiting to go in.
The Troubleshooting Progression
To give the board a clean baseline, I worked through the hardware systematically:
- RAM Reconfiguration and Single-DIMM Isolation: I pulled both XPG modules out of
A1/A2and locked them intoA2/B2using the board's single-latch top clips. When the dual-channel pair still failed to POST, I tested each stick individually inA2andB2to isolate a potential dead DIMM or dead memory channel. - Storage Upgrade: I dismounted the incompatible 64GB Vaseky SATA M.2 stick and installed the 1TB
Predator GM7NVMe drive into theUltra M.2slot. - CMOS Reset and Battery Replacement: Because failed memory training on AM4 boards can lock the NVRAM state, I pulled the old
CR2032coin-cell battery from beneath the GPU shroud and installed a freshToshiba CR2032to force a complete CMOS reset. - Power, PCIe, and Front-Panel Isolation: I reseated the Sapphire GPU into the primary
PCIE1slot, verified the 8-pinATX12V1CPU power and 24-pin motherboard connectors, and disconnected the thick 19-pin front-panel USB 3.0 header (USB3_5_6) to rule out a shorted case port dragging down the 5V rail.
The Roadblock: Locked Super I/O State and Factory BIOS Limits
Despite correcting the RAM layout, swapping in the proper NVMe drive, and clearing the CMOS, the exact same symptom persisted. Flipping the rear switch on the Aerocool KCAS 850W PSU to I sent 5VSB standby power to the rear I/O, flashing the keyboard lights for three seconds and lighting the mouse sensor. Pressing the case power button spun up the 12V fans and GPU lighting, but the keyboard stayed dark and the monitor never received a signal.
More importantly, holding down the case power button for four to ten seconds wouldn't force the system to shut down. Every power-off required flipping the hard switch on the back of the PSU.
When a motherboard ignores the four-second hardware power-button override, the Super I/O controller (Nuvoton) and chipset are locked out of their normal ACPI state machine. Looking closely at the BIOS chip next to the M.2 standoff explained the most likely culprit: a factory P2.70 sticker. On the ASRock B450M/ac R2.0, BIOS P2.70 supports earlier Ryzen chips, but newer Ryzen 5000 revisions (such as the Ryzen 5 5500, 5600, or 5600G) require BIOS P2.90 or higher to initialize. Without a compatible BIOS, or if the case's front-panel reset line is shorted, the PSU's PWR_OK signal is failing, or both RAM sticks are dead out of the box, the CPU never executes POST code.
Knowing When to Hand It Off
At that point, the next step would have been pulling the AMD Wraith Stealth cooler to inspect the CPU model and AM4 socket pins, or swapping in a known-good bench PSU and spare DDR4 stick. Doing unpaid work on a friend's machine on a Friday is one thing; pulling the heatsink off someone else's mystery build without spare parts on hand is a fast way to inherit ownership of a dead board.
Even though the rig didn't wake up on my desk, the triage did its job. The RAM is finally in the right channels, the unusable 64GB SATA stick is replaced with the 1TB Predator NVMe drive, the CMOS battery is fresh, and my colleague has a clear, three-point checklist to hand to a local repair shop so they can drop in a test CPU to flash the BIOS and bench-test the board in ten minutes instead of billing him for two hours of guesswork.