ROME2D16-2T and dual Epyc 7H12 - Can't install Linux

[mullenba]

I've tried OpenSUSE and Ubuntu Server, but can't get Linux to work on this system. The best I've been able to do is limit 'maxcpu' to 64 in grub, which allows one slot to be recognized. Trying to use both causes the system to restart during boot. I'm completely lost on what to try or check!

From my research, here are some of the things I've tried.
- Disabled Core Performance Boost (CPB) and fixed Global C-States Control to rule out voltage drop failures when all cores initialized at once.
- Disabled IOMMU and SR-IOV to bypass faulty virtual IO mappings during storage device probing.
- Tried Ubuntu 26.04, 24.04, and OpenSUSE Tumbleweed
- Tried forcing NUMA NPS1 and NPS2
- Verified all memory is in the right slots and recognized.

 

[RageBone]

Roughly sounds like a degraded and instable CPU.
I have multiple different AMD cpus including epyc who crash some point on the way to the OS.
Windows is the most obviously picky.

Have you tried single CPU setups to validate the CPUs as fine and working individually?

 

[TrevorH]

Have you tried turning off HT in the BIOS so that it limits the number of cores reported? And where does it "restart during boot"? With what symptoms? Have you removed 'quiet' from the kernel command line so you can see progress?

 

[mullenba]

Have you tried turning off HT in the BIOS so that it limits the number of cores reported? And where does it "restart during boot"? With what symptoms? Have you removed 'quiet' from the kernel command line so you can see progress?

Turning HT off in the BIOS doesn't seem to have an effect. This is what it shows before rebooting.

 

[TrevorH]

There's an awful lot going on there in parallel so there are several candidates for problems, ixgbe, nvme, lots of USB probing, all going on at the same time. It did boot past the point where it brought all processors up so I'm not sure it's just a "too many cores" problem. Also from "systemd 255-4" that sounds like Ubuntu 24.04, are the symptoms the same with a newer distro? I would try disabling things in the BIOS to narrow down what the problem is as well, see if you can turn off the 10G interfaces for example. On Fedora or RHEL-alike distros you can use dracut to blacklist kernel modules - e.g. rd.driver.blacklist=<drivername>[,<drivername>,... appended to the kernel command line. You could try blacklisting ixgbe and nvme (it won't see the drives but you'll know if it boots) and/or xhci-hcd (USB3).

It's also possible it's a power problem and it gets to a certain power usage and something browns out.

 

[RageBone]

what does the syslog in the impi say? Could be uncorrectable memory issues as well.

 

[mullenba]

what does the syslog in the impi say? Could be uncorrectable memory issues as well.

I checked the post codes and everything looks good. IPMI log just has time sync and a high temp alarm from this morning (I forgot to tighten down the cooler)

 

[TrevorH]

What PSU are you using and are all possible power connectors on the motherboard connected?

 

[mullenba]

What PSU are you using and are all possible power connectors on the motherboard connected?

Is this a bad CPU? Post codes don't show an error, but I made a video of the boot and went frame by frame to see everything it displayed. I found this:

 

[TrevorH]

It's definitely not healthy. I believe you can get chatgpt to decode that hardware error info and that might give you more of a clue as to whether the processor itself is dodgy or if it's doing it due to something else.

 

[RolloZ170]

Turning HT off in the BIOS doesn't seem to have an effect.

needs a AC power cycle to turn off.

 

[RageBone]

MCE (Machine Check Exceptions) point at this being a hardware issue.
As it sais, Processor 2 on Socket 1.

https://forums.servethehome.com/index.php?resources/machine-check-exception-mce-workaround.55/

Its for intel so maybe not relatable. I was also told that the MCA Bank 27 points at something not core related.
PIE for Power Interrupts E(something).


Try single socket configs to validate a single CPU as fine.
Try swapping cpus, cpu0 as cpu 1, 1as 0. If the error stays where its at, its maybe a board fault.
Maybe try that guide anyhow once you know which CPU is bad and if its bad.