Originally Posted by Falkentyne
The first setting should have definitely worked. Can you test that again with XMP disabled?
Also, try testing with CPU multiplier set to x43 and cache ratio at x40. Is this stable? Also you can set VCCIO to 1.15v and VCCSA to 1.20v. Keep PWM switching frequency at 400khz.
x47 is stock turbo boost on all cores. 100% of CPU's should be *rock stable* with 160 AC Loadline (Intel maximum) and Auto voltage at 4.7 ghz.
On auto voltage, Loadline calibration should not be set at low when ACLL is set to 160. It must be left at standard (or Normal) or too much overvolting will occur. If VR VOUT is still too low with ACLL=160 and LLC=Standard, you should use a DVID offset instead, not loadline calibration.
AC LL of 1 is ok for Vcore LLC=turbo or high, 40 is ok for Vcore LLC low or medium (on auto voltages).
If x47 is not stable at all, and XMP disabled doesn't help, you may need to RMA that CPU, as it qualifies as being defective.
I've been testing these suggested settings using 2133 MHz memory speed only with XMP disabled. None of these fix the issue. Even after downclocking core to 4.3 GHz and cache to 4.0 GHz using ACLL=160 and LLC=standard, I still get incorrect residuals.
The mismatching residual is always the same order of magnitude as the correct residual. For example, if the correct residual is around 9.026e-10, an error may offset that to 9.8e-10 or 1.1e-9. I ran prime95 blend for 6 h and small fft for 1h without errors. Perhaps the calculation errors are so small that prime95 can't detect them.
I'm going to check the motherboard and CPU for physical damage and reseat the CPU.
EDIT: I found a post by the developer of Linpack Extreme (another interface to MKL benchmark). The developer says:
"Different residual values are normal on modern Intel CPUs.
Abnormal on AMD and legacy Intel CPUs (Sandy Bridge and older)."
I'm not sure what to think of this. The residuals I get are always in the same ballpark. The dev also says that linpack performs its own residual check already.