i9-7980XE OC and XMP issues

Oct 9, 2018
My spec:
Asus Rampage VI Apex
G.Skill Trident Z DDR4 4000MHz F4-4000C18Q-32GTZKW
GPU: 1070ti
PSU: Seasonic Prime Ultra Titanium 850W
Custom liquiq cooling with ekwb monoblock

My tasks:
OC (not extreme overclocking) 4.1-4.4 all cores for stable work.

In all cases, the Aida CPU stress test passes without problems.
After disabling USB Legacy support, the memory test also passes without problems.
But the cache test always fails: sometimes the test just stops with an error, sometimes with BSOD.
After starting the Cinebench r15, the computer just turned off.

Memtesting it For five hours and no errors.
Set the default settings and there were no more problems. Conclusion: problems in overclocking.

What I've tried:
XMP profile
XMP profile and manul DRAM timings and voltage
Same as above + little offset for cpu and cache voltage
Manual ratio for each core + manual DRAM settings
lower DRAM freq.
and some other things..

but nothing helped. With option SyncAllCores I still can not pass the test, with option ByCoreUsage it turns out.

I do not ask to make fine tuning for me, tell me in which direction to look. Not enough voltage? Maybe power supply unit? Or do I need to edit the frequency? Maybe some advice or a hint? Many thanks.
Oct 9, 2018

Yes, chipset drivers and BIOS latest version.

I did some more tests and here are the results:

All tests were performed with the following settings:
LLС: 1
Intel Speedstep: disable
spread spectrum: disable
SVID Support: disable
CPU and DRAM capability: 140%

1. XMP profile, Core Ratio: Auto with option Sync All Core which actually means 4400 for all cores.
Cinebench was successful, but the render test is not, because when the render started, avx instructions were used and the temperature jumped to 105, which apparently led to the triggering of protection and reboot.
2. Same as section 1 plus AVX offset 3 = 4100-4400 depending on the task.
Cinebench and Blender passed the test, but Aida64 cache test stopped with an error after 12 seconds.
The memory test stopped with an error after 6 minutes.
3. Max Core Ratio: 42, AVX offset 3 = 3900-4200, cache and system agent voltage: 1.1
Cinebench and Blender passed the test, cache test stopped with an error after 11 minutes.
4. Increased RAM or cache voltage did not bring results.

I checked the processor with a processor diagnostics utility from Intel. All tests were successfully passed. I checked the memory for 5-6 hours with memtest - no errors.

Can this behavior be a prerequisite to a problem? Or just forget about this test?

I can give you another example.
Everyone probably knows streaming software such as OBS or XSplit.
The settings of the encoder are the same, however, XSplit sometimes leads to blue screens, especially with a CPU multiplier of 44, but no problem with OBS. Streaming is also something from the rendering area. Maybe there will be some kind of hint.

Oct 9, 2018
I would like to add information to the previous message. I received more or less stable results with the settings of 3900-4200 for all cores. For the purpose of the experiment, I left the frequencies the same, but returned all voltages except memory to auto. There was no time for long tests, but more than 20 minutes there was a test and no errors. It turns out that errors begin when I try to raise the multiplier or reduce the offset (increase the frequency for avx).