I've been dealing with severe stability problems on my AM5 system for about a year, including random AMD Adrenalin driver timeouts, black screens, GPU fans suddenly ramping to 100%, complete freezes, and occasional POST failures after a cold start. My system uses a Ryzen 7 9800X3D, ASUS TUF GAMING B650-PLUS WIFI, ASRock Radeon RX 7900 XTX, 1000W be quiet! Pure Power 12 M, and 64GB of DDR5 running at 6000 MT/s with EXPO enabled. The memory consists of two different 2x16GB kits: GoodRAM modules rated at 1.35V and Kingston modules rated at 1.40V, although both are CL30. The modules occupy all four DIMM slots. When the system fails to boot, the BIOS reports that it has entered safe mode after a failed POST; restarting then usually lets Windows load normally at 6000 MT/s. I've already tested with EXPO disabled, updated the BIOS and chipset drivers, and clean-installed the current graphics driver. The GPU is properly seated and uses three separate PCIe power cables. Could the mixed memory kits or using four DDR5 sticks be causing both the memory-training failures and the apparent GPU crashes? Would replacing them with one matched 2x32GB kit be the best route to reliable operation?
3 Answers
The graphics crashes may be a separate issue. Since the GPU is using separate power cables and the drivers have been clean-installed, try an older Adrenalin release instead of assuming the newest one is best. Some users have found earlier releases more stable. If an older driver still crashes with only two matched RAM sticks at conservative settings, then test the GPU in another system or investigate the motherboard and graphics card separately.
The POST warning strongly points toward memory configuration or memory training rather than proving that the graphics card is defective. Make sure the BIOS is current, but also avoid cutting power completely with a switched power strip while troubleshooting, since a full cold start can expose marginal training settings. Test one kit at a time with default settings, then gradually increase the memory speed. If two sticks are stable and four are not, the practical fix is a matched 2x32GB kit rather than trying to force 6000 MT/s with all four slots populated.
Four DDR5 sticks are much harder for the memory controller to run, especially at 6000 MT/s. Mixing kits with different voltage and memory characteristics makes it even less predictable, even if the advertised timings match. Try removing one complete kit and test with only two matched modules. If the crashes stop, replace the current memory with a single matched 2x32GB kit. You can also temporarily test at a much lower speed, around 4400 MT/s, to confirm whether memory training is involved.
I've already tried disabling EXPO, but the crashes still happened. Testing with only one of the two kits is probably the next step, even though one set is difficult to remove because of the CPU cooler.
Every driver version I've tried so far has produced problems, but I'll test an older release after isolating the memory configuration.