It's been awhile since I posted, so I thought I'd update you. I managed to fix my motherboard myself. A couple of the CPU socket pins were bent, so I bent them back into place (yes, I was nervous!) and it seems to be fine. Anyways, prior to all that, I was having random lockups/freezes and Geekbench 5 wouldn't complete. I decided to reinvestigate and it turns out that it was always stalling on the last test: machine learning. It turns out that the AVX and AVX512 offsets (the machine learning instruction sets) were too low. I jacked those numbers way up (12 and 9, respectively) and now GB5 runs to completion reliably. I'm in the process of incrementally lowering the offsets, but I'm getting GB5 scores on par with kgp's original post. I'm getting ~71k on GB5, and ~8500 on Cinebench R20.
I swear I'm not obsessed with benchmarks. I just figured that if a short benchmark was enough to crash the system, then maybe something was configured incorrectly, and the system was truly unstable. Now that I've identified the problem and workaround, I feel much more confident that my workloads won't crash.
PS: I'm still on Mojave and Clover.