I'm troubleshooting a memory overclocking issue on a Ryzen 9 3900X system with an ASRock B550 Taichi running BIOS 4.00. The system has a 2x16 GB G.Skill Trident Z RGB DDR4-3200 CL16 kit, a GeForce RTX 2070 Super, an NVMe bifurcation card, and a 360 mm Cooler Master AIO. The memory is stable at default JEDEC settings, but enabling the motherboard's DOCP/XMP profile at DDR4-3200 produces inconsistent results.
With one DIMM installed, both sticks can pass in B2, but one stick fails almost immediately in A2 with thousands of errors. Testing both sticks in A2 and B2 has also produced contradictory results: sometimes the system completes several Memtest86 passes without errors, while other times it reports around 10,000 errors within seconds without any hardware or BIOS changes. The failures appear immediately when they occur, and the error patterns are repeatable.
I inspected the Ryzen CPU for bent or contaminated pins, cleaned the DIMM slots and memory contacts with high-purity isopropyl alcohol, reapplied thermal paste, and reseated the cooler. The behavior remained the same. At JEDEC speeds, both modules are stable in either slot configuration.
Interestingly, setting DOCP while forcing the Infinity Fabric to an asynchronous 1567 MHz appears to make the system stable with both DIMMs installed. What is the most likely cause of this behavior, and what other troubleshooting steps should I try without access to another motherboard, CPU, or memory kit?
2 Answers
Because one stick consistently fails in A2 while both sticks pass in B2, the first suspects are the A2 slot, the memory channel, CPU socket contact, or the CPU’s integrated memory controller. The fact that a later test can pass and then suddenly fail also suggests a very small stability margin rather than a clearly dead DIMM. Removing the CPU and checking for bent or contaminated pins, cleaning and reseating the memory, and reinstalling the cooler evenly are reasonable steps. Also make sure the DIMMs are installed in the board’s recommended A2/B2 positions and that the cooler is not applying excessive or uneven socket pressure.
The asynchronous Infinity Fabric result points more toward Ryzen memory-controller or fabric stability than a completely defective memory module. DDR4-3200 normally runs with a 1600 MHz FCLK in a 1:1 ratio, so forcing 1567 MHz makes the fabric and memory clocks asynchronous and may avoid a marginal FCLK or IMC limit. That can appear stable while the normal 1600 MHz synchronized setup fails. Try manually setting memory speed and timings instead of relying entirely on DOCP, then test FCLK values around 1500–1600 MHz in small increments. Keep memory voltage, SoC voltage, and other supporting voltages within conservative motherboard defaults, and validate with multiple full Memtest86 passes plus an in-OS memory stress test.
The system appears stable with DOCP enabled and Infinity Fabric set asynchronously to 1567 MHz, whereas the normal synchronized configuration fails. I’m running a complete multi-pass test to confirm whether that workaround is reliable.
I checked the CPU pins and found no damage or debris, cleaned the DIMMs and slots with 99.99% isopropyl alcohol, reapplied MX-4, and reseated the cooler. The results were unchanged: stick A fails immediately at DOCP in A2, while stick B passes there; both sticks pass in B2. At JEDEC speeds, everything remains stable.