SuperMicro H12DSi-NT6 / Epyc 7763 Dual CPU boot failure

[KBentley57]

Hi all,

New to the site, but I seem to hit this forum quite a bit with my googling. I've got a SM H12DSi-NT6, two Epyc 7763's, and 256 GB of LRDIMM ram (2 x 128 2933 ECC Micron). With one CPU and both both dimms in the C/D channel of CPU 0, the system boots. When we try both cpu's, the system never gets to the first stage of anything. The power light doesn't stay on, and it wont turn off or reset with the front panel buttons, only by turning off the power. Any advice?

We've swapped CPU's with only one installed, and it works fine, so we know the CPU's are good. I've tried putting one dimm in each C channel for cpu 0 and 1, but no luck there either. The manual shows (page 32) that if I've got two dimm's, they need to go in cpu0 slot c1/d1, but there are other problems with the manual that make me suspicious of the accuracy of information. We've got more RAM on the way, it just hasn't made it here yet.

Thanks in advance for any help!

Kyle

 

[jpmomo]

do you have all of the psu connectors connected? How long did you wait for the system to try and boot with the dual cpu setup? Can you take a close look at both sockets without the cpus installed? Does everything look ok? ex. no bent pins or pieces of broken plastic bits in or around the socket. Make sure that both cpus are properly tightened and seated correctly. In the long run, you will want to populate all 8 channels for each cpu. that might cost a lot with 128GB dimms. You should at least try some other ram chips to rule that out as well. good luck and keep us updated.

 

[KBentley57]

Hello JP,

Thanks for the feedback. We have a 1300W PSU running the system, both 8 pin CPU power cables connected. I waited a good amount of time, just to make sure it had a chance to initialize, at least 10 minutes in one case so that I could definitively rule it out. It's funny you mention the pins. There's def. a section that looks non-uniform over a ~8 pin spot, but it was this way from the factory, and I'm hoping that it was just some odd reflections. There doesn't seem to be any missing pins, but you can def. see that the area is different than the surrounding pins.

What I didn't mention in the first post, in hopes to keep the problem focused, is that this board is already a little borked. The two 10Gb ethernet ports are dead, as are all of the usb2 ports on the back. I've been chatting with a supermicro tech back and forth for a few days, and we're going to RMA the board anyways. We were really hoping to run a few tests before we did as more of our hardware came in.

I'll keep posting back with what I find. Hopefully it's just a lemmon.

On a side note - you're right about the ram cost. Is mixing ram size between different channels asking for more headaches? We needed a TB, but not really much more. If we were to populate the second 8 slots with say 32 GB modules, would that be ok, or do the P0 and P1 channels have to match in some way?