Fail to POST with 2nd CPU riser

[_A_T_]

Good day,

I have a Z640 workstation that I have recently acquired, and upgraded from 2xE5-2650v4 to 2xE5-2690v4.

Prior to doing so, I upgraded the BIOS to (M60) 2.61

On first boot with the new CPUs, I received two warnings at post:

932-Warning one of QPI links is not operating
932-Warning one of QPI links is not operating

At the time, all CPUs, cores and memory were identified in SuSE Linux.

Subsequent boots fail to POST. The power LED illuminates steady white, but otherwise no activity. After several minutes the fans run at high speed and no other activity is evident.

If I remove the 2nd CPU riser board, the system will boot normally.

I have tried:

1. Swapping CPUs. Same result. Either CPU as CPU0 works as normal, either CPU in socket 1 and board installed results in the above described behavior.
2. Swapped memory from the riser to motherboard, and attempted boot without riser, resulting in a successful boot. With riser board, fails to POST as above.
3. Removed all but one DIMM on each. Same results.
4. Replaced original CPUs. Same results with either CPU in socket 0.
5. CMOS reset via reset button, removed 2032 button cell.
6. All fans power up - board is seemingly powered.
7. Inspected CPU1 socket. Pins appear clean and none bent.
8. Inspected QPI connector pins. All appear in good condition.
9. Same for riser board power connector.
10. With the original Quadro K220 removed and the riser board installed, same non-POST activity. White LED, fans at speed after a short time. With the riser board and Quadro removed, 6 beeps and red LED as expected.
11. As a last ditch effort, I tried Crisis Recovery, perhaps a corrupt BIOS. With the riser installed, 8 beeps and RED flashing LED, otherwise no activity. USB activity LED remains dark. Without riser, recovery ran and BIOS recovery was performed as expected.

Any suggestions greatly appreciated. Any tests I can do on the riser or motherboard?

[DGroves]

I ran into this years ago on a z820 where the client was adding a sec cpu

and he took the time to match/source a identical second cpu,

the problem was the installed cpu was a eval/engineering sample not retail

when i saw the error and non retail cpu i suspected what was going on.

as a quick test i swapped the cpu's and the system booted

further inspection confirmed the original cpu (which has dual "QPI" links)

did indeed have one defective qpi link which rendered the systems dual cpu's

unable to talk to each other  it was also discovered the "qpi" links

were slower than a retail model cpu

what's a "QPI" link read the link below

https://www.intel.com/content/www/us/en/docs/programmable/683888/current/qpi-overview.html#:~:text=T....

bottom line, we replaced both cpu's with retail models

and all issues on that system were resolved

bottom line never use QL or Engineering samples on dual socket systems

or any production system, if gaming on a single cpu system then you might

want to take a chance, but otherwise stay away from non retail cpu's

[DGroves]

i didn't mention it on the first reply but damage to the cpu socket pins can also cause a non functional qpi link

same for damage to the riser board (bad/defective riser board) or a bad connection between the socket for the motherboard and riser

[CharlotteYoung]

Thank you, you saved my day.

Spoiler

I’ve seen similar POST issues when hardware compatibility isn’t optimal. If you're adding a second riser, make sure your system supports the specific configuration. For similar multi-CPU applications, a component like the CQ2332 can handle advanced loads more reliably in demanding workstation environments.