My PC hangs during a restart whenever A-XMP is enabled. This happens whether I restart from Windows or Linux: the system gets stuck with the DRAM debug LED lit, and I have to hold the power button—or sometimes switch off the PSU—to shut it down. A normal shutdown works correctly, and the next cold boot is fine. Disabling A-XMP completely fixes the problem, but then the memory runs at 4800 MT/s instead of its rated 5600 MT/s.
The issue began unexpectedly a few months ago. I have updated to the latest motherboard BIOS and tried common memory-training and voltage adjustments. My system is otherwise stable. The build is a Ryzen 7 7700X, MSI PRO X670-P WiFi, 32 GB of Corsair Vengeance DDR5 rated for 5600 MT/s, and an RX 7900 XTX. I currently use PBO Curve Optimizer at -30 and have tested the memory at 6400 MT/s and 1.35 V, but the restart problem also occurs with PBO disabled and the memory at its standard 5600 MT/s profile. It only disappears when A-XMP is disabled and the memory falls back to 4800 MT/s.
I have run extended CPU, GPU, memory, and gaming stress tests, and the computer remains stable apart from this restart-specific failure. What could cause this, and is there a way to keep the memory profile enabled while fixing restarts?
2 Answers
The 5600 MT/s rating is an overclocked memory profile rather than a guaranteed default for every Ryzen 7000 CPU and motherboard combination. Cold boots can succeed while warm restarts fail because the firmware handles memory training differently during a restart. That does not necessarily mean the system will crash during regular use, but it does suggest the current training or memory-controller settings are marginal. Try loading optimized defaults, updating or re-flashing the BIOS, reseating the DIMMs, testing each module separately, and manually using a slightly lower speed instead of disabling the profile completely. If only 4800 works reliably, the CPU’s memory controller, the kit, or the board’s training behavior may be the limiting factor.
A restart failure is still a sign that the memory or memory-controller settings are not fully stable, even if the system passes normal benchmarks and games. A -30 Curve Optimizer value and DDR5-6400 are both fairly aggressive on AM5, and 6400 may also force a less favorable memory-controller ratio depending on the chip. Clear CMOS, return the CPU and memory to stock, then test only the memory profile. Use tools such as MemTest86, OCCT memory testing, and y-cruncher. If the problem remains at the rated 5600 profile, try manually setting a lower speed such as 5200 or 5400, or increase memory-controller-related settings conservatively according to the board’s guidance.
I also tested with PBO disabled and the memory at the standard 5600 MT/s profile, with extended stress testing. The restart issue still occurred there; it only stopped when A-XMP was disabled entirely and the memory ran at 4800 MT/s.
That makes sense, although I’m confused why the system can remain stable under normal workloads if the profile is marginal. I’m mainly trying to find a fix for the restart-specific memory-training failure without giving up all the extra memory speed.