I built a system with a Ryzen 9 9900X3D, RTX 5080, MAG B850M Mortar WiFi motherboard, 32GB of Corsair Vengeance RGB DDR5-6400 CL32, an 850W power supply, and a 1TB NVMe SSD. During an initial dry test, the system booted normally, Windows installed, the BIOS was updated, and the memory profile worked. After installing the water-cooling loop and completing a leak test, the system stopped producing video output and the motherboard began showing CPU and memory debug LEDs.
The GPU, cables, RAM, and connections have all been checked and reseated. A single memory module works in DIMMB2, but neither module works by itself in DIMMA2, and installing both modules fails in either A2/B2 or A1/B1. Clearing CMOS, changing DRAM voltage slightly, loosening and reinstalling the CPU water block, and testing with the block backed off did not resolve the issue. Loosening the block initially changed whether even one stick would boot, which makes CPU socket, memory-controller, or mounting pressure a possibility.
The system draws roughly 490W from the wall while running a GPU stress test, so the 850W power supply seems unlikely to be the cause. XMP is currently disabled, although the kit previously worked at its rated speed before the rebuild. Before draining the loop and removing the motherboard for an RMA, what other troubleshooting steps should I try?
2 Answers
The behavior may not mean that three DIMM slots suddenly failed. On AM5, memory support depends heavily on the CPU's integrated memory controller, the socket contact, and pressure from the cooler. The fact that changing the water-block tension affects whether a single stick boots points more toward socket contact or mounting pressure than toward the power supply.
For testing, use one module at a time in the board's recommended A2 slot, with the memory profile disabled and all settings at defaults. Also inspect the AM5 socket carefully under bright light for even a slightly bent or contaminated contact. A socket contact associated with a memory channel can cause exactly this sort of partial-memory detection problem. If the socket looks good, test with another known-good DDR5 module or kit if possible. The CPU's memory controller could also have been damaged, though that is less common than a socket or mounting issue.
DDR5-6400 is an overclocked setting on this platform, but it should not prevent the machine from posting at safe default speeds once the profile is disabled. A higher DRAM voltage is unlikely to fix a slot or channel that is not being detected. The power draw you measured also makes an undersized PSU a poor fit for these symptoms.
The next useful isolation test is to remove the GPU and all unnecessary devices, connect the display to the CPU's graphics output, and run the board with one DIMM at stock settings. If one stick works only in B2, focus on the CPU socket, the memory channel wiring, and cooler pressure. If you can borrow another CPU or motherboard, swapping one component at a time will identify whether the board or the processor's memory controller is at fault.
I had already disabled the memory profile and tried clearing CMOS. I also backed the water block off and reinstalled it from scratch, but the system still only starts with one stick in B2. I’ll inspect the socket closely and try to test with another kit before pulling the board.