As I found out today, it was the motherboard being unsteady on power. Replacing the motherboard has fixed the problem. FWIW, I've had COTS hardware running 24/7 for close to 20 years. From time to time I've also had some SuperMicro hardware and while it is nice to have, "gaming" hardware as you put it does the job just fine and lasts almost as long as the enterprise level hardware. I'm willing to accept the "gaming" hardware differences. It's worked out just fine for me.
Glad You found it, let's hope it stays ok.
I'm not opposing such an approach, rather the contrary; I myself do mix&match whatever I happen to get, curious for the outcome. Only, one needs some willingness to investigate and figure things out, and not being frowned upon an occasional bug-hunting.
I usually go with the cheap stuff, then watch where the weak points are, and replace just that with improved equipment, as needed - and learn a lot in the process. With computers I had mainly three areas of issues:
- failing memory, not easily detectable with memtest. ECC capable equipment is worth the premium, it reports exactly when and what failed.
I'm still looking for a way to monitor temperature of my memory chips - they should provide that data, but FreeBSD cannot find it. - cooling problems. There is AFAIK no packaged solution to monitor temperature of a bunch of disks, and then switch individual fans appropriately. But it's not too difficult to script+build one.
- power wiring. Putting lots of drives onto Y-cables does not work, the connectors have too much resistance. Sometimes even the original wiring does not work.