Question 5950X Cache hierarchy error

[pakotlar]

Got a random reboot when starting windows, event logger says WHEA cache hierarchy ID for APC ID 31. This corresponds to core 15 I believe, which is perf #12/13, so middle of pack. First reboot in about 4 months since enabling CO.

I have PBO enabled with curve optimizer set to -30 on that core. Additionally I had Fclk at 1866mhz and ran overclocked to 3733, 64GB (32 x 2, planning on 128 longer term for ML stuff) that is rated for 3600.

Anything to be concerned about (could CPU be faulty or is this clearly just effect of overclocking)? Is this more likely to be caused by memory, fclck, or CO? Before updating to Agesa 1.2.0.3 patch B I was able to do 1866 fclck no problem (but no curve optimizer).

So far have backed down to 1800 fclck and 3600mhz memory clock, but have left curve optimizer untouched. No reboots, but so sporadic I don’t expect them for many months.

 

[Cali3350]

I had been getting this error on my 5800x while playing Diablo 2 Resurrected. Would work perfectly fine but crash maybe once every ~10 hours. Was running PBO with a -20 offset. Moved it to -15 and it seemed to help, but still crashed eventually. Now at -12 and no crashes so far.

So it seems, anecdotal, it might have been a voltage issue for me.

 

[Makaveli]

Got a random reboot when starting windows, event logger says WHEA cache hierarchy ID for APC ID 31. This corresponds to core 15 I believe, which is perf #12/13, so middle of pack. First reboot in about 4 months since enabling CO.

I have PBO enabled with curve optimizer set to -30 on that core. Additionally I had Fclk at 1866mhz and ran overclocked to 3733, 64GB (32 x 2, planning on 128 longer term for ML stuff) that is rated for 3600.

Anything to be concerned about (could CPU be faulty or is this clearly just effect of overclocking)? Is this more likely to be caused by memory, fclck, or CO? Before updating to Agesa 1.2.0.3 patch B I was able to do 1866 fclck no problem (but no curve optimizer).

So far have backed down to 1800 fclck and 3600mhz memory clock, but have left curve optimizer untouched. No reboots, but so sporadic I don’t expect them for many months.

I would lower the -30 on that core.

 

[PJVol]

So far have backed down to 1800 fclck and 3600mhz memory clock, but have left curve optimizer untouched

Why don't you loosen the magnitude for that core (which seems most logical to do right away) instead of messing with interchip and memory subsystem?
This is one of those rare cases when system clearly points to an issue.
Most of the time you see something like "kernel power event id 41" and put it however you want.

 

[pakotlar]

Yeah, I backed off to -25 on that core. No further issues. For safety I backed off tdc and edc as well, don’t need the extra few %

 

[Shmee]

Yeah, sounds like CO isn't giving enough voltage. Even a -25 offset sounds very optimistic to me, but it depends on the chip and core, of course. Have you tested your CPU config with core cycler?