Unable to resolve sudden hard reboots

[some3t1m3s]

Since weeks I'm trying to resolve the following issue: My Proxmox Server randomly reboots for no apparent reason. It's a 'hard' reboot as if someone presses the reboot button. This happens every 1 hours to 3 days.

The systemd journal shows nothing noteworthy before the reboot, usually there aren't entries directly before the reboot happens anyway, just "-- Boot [..]". The reboots are completely unrelated to load. I'm unable to provoke reboots, eg. when loading a VM with Prime95. The BMC also doesn't log anything, especially not the "Vcore 0.0 V" error that the MC12-LE0 produces with a Ryzen 5950X under certain circumstances.

My initial configuration:

• Gigabyte MC12-LE0
• Ryzen 5950X
• 4x Samsung 32GB DDR4-3200 CL-22-22-22
• 3x Crucial MX500 250GB
• Mellanox Connect-X 3 CX311A
• Seasonic Gold Focus 450W

The case is well ventilated. When I push the CPU with normal workloads it eventually reaches 75°C, that seems to be the limit. Other temperatures from 'sensors' are well below that. I can touch the Connect-X3 heatsink with my hand and it doesn't even feel warm.

Things I tried so far:

• Reading basically every thread on the internet mentioning "Ryzen" "Proxmox" and "Reboot"
• Removing all 'optional' hardware (not listed here as obviously unrelated to the problem)
• Updated Proxmox to the current version
• Updated the BIOS and BMC
• Replaced the mainboard with a new ASRock X570D4U
• Updated the BIOS and BMC
• Installed amd64-microcode
• Replaced the PSU with a new bequiet! Pure Power 12 M 750W
• Stresstested the RAM with memtest86+
• Disabled Core Watchdog in BIOS (and re-enabled it again after it didn't help, same with all other BIOS options mentioned)
• Enabled Eco Mode in BIOS
• Disabled PBO
• Manually lowered the TDP and the boost limits even further
• Disabled Global C-States in BIOS
• Changed the Power Supply Idle Control setting
• Set all Windows VMs CPU type to x86-64-v2-AES
• Disabled all Windows VMs altogether
• Underprovisioned the remaining VMs so vCPUs < actual CPU cores (16)
• Installed optional 6.11 kernel
• Considered having a breakdown and applying for the job with the least amount of responsibility

Things I haven't tried so far:

• Exchanging the CPU
• Exchanging the RAM despite the fact that it passes memory tests
• Replacing the Connect-X 3
• Disabling all VMs altogether

I'd like to point out that I have a similar setup with TrueNAS Scale running on Gigabyte MC12-LE0, Ryzen 5600, Connect-X 3 CX311A, 4x 32GB Kingston Server ECC RAM that runs rock-solid to such a degree that I didn't even bother to update the BIOS.

Do you have any idea or direction how I might be able to narrow down what the culprit is? Do you need any additional information?