Seems like lowering the speed of the memory in bios fixes the problem. doing some testing now.
Ok, first do not run 96GB, please. So the 64GB at 3200 sets.
Second, mixing and matching speed can effect timings, as each set may need different amounts to work as well as different voltages.
Third, if stable at lower MHz, you may just have a weaker memory controller on one of the dies. You also could just have a situation of incorrect timings for Ryzen gen 1 in the XMP settings, so it trips up once in quad channel.
How long do you have that you can take the machine out of commission? What knowledge do you have for ram overclocking? How familiar are you with ram stability testing?
It could even be the SOC voltage set wrong (sometimes lower is better; like if I set over 1.05V on LLC level 2 SOC, I lose stability on my system with my 32GB at 3466 CL14).
So it really is a situation of needing to know timelines, familiarity with the process, etc.