I received yet another crash with no error, and with error´including whea internal parity error. So i went back to LLC 6 from 8 at 4.9 ghz, 1.3 volt. Started from 1.2v and worked upwards. My load lines did not cause any heat issues or instability. Maybe a developer could explain if its a coding issue or cpu i would be glad. If this continues i might RMA my cpu and motherboard. Spectre and meltdown mitigation has been disabled, could this be a factor that brings up the error in event viewer, and coreparking disabled?
@Ariuc It's some sort of Intel microcode bug. RMA won't do anything.
I did some tests last night. I actually *disabled* loadline calibration completely (leaving it on the standard value of 1.6 mOhms of vdroop) and used Auto vcore, letting the Internal VR AC loadline do the work (which Intel specifies as 1.6 mOhms maximum also). This worked out fine (5 ghz) but idle voltages were much too high (1.40v) and low to medium load voltages were also too high, but Apex had no problems. 15k small FFT AVX prime95 ran fine here at 1.240v (VCC_Sense voltage) and 184 amps, and 96C max temps after 6 minutes, so this was ok but again not what I suggest. The reasaon I disabled Loadline Calibration and let the AC Loadlines do the work, is this improves transient response massively and stops excessive unmeasured voltage dips and spikes (Important for something like Prime95 on very high amps load).
I then enabled SVID offset, which kept idle voltages the same, and raised small FFT avx disabled prime95 by 30mv, but this raised AVX enabled prime95 small FFT by *90mv!*...leading to 212 amps, and 105C temps in FIVE seconds! Ouch....uh....
So SVID offset seems to boost certain load voltages.
Ok so I kept SVID offset enabled but set AC loadline to 1.0 mOhms (I kept DC loadline at 1.60 mOhms to match the CPU Vcore loadline calibration standard preset setting of 1.6 mOhms also). This reduced idle voltages significantly--down to 1.30 to 1.32v. Tested 0.9 and 1.0 mOhms for AC loadline were both rock stable in the harshest stress test on the planet (15K small FFT AVX prime95), with VR VOUT vcc_sense voltage reported at 1.230v (0.9 mOhms) and 1.250v (1.0 mOhms), and everything on the planet ran fine.
Except Apex Legends.
0.9 mOhms of AC loadline with SVID offset enabled caused random CPU parity errors. 1.0 mOhms didn't (most likely would have eventually; just didn't play long enough) but the game crashed anyway after awhile (with that typical "execute code before its ready" hexadecimal code crash). Probably 1.05 or 1.1 mOhms would be enough here but the idle voltage is getting too much for me to like.
I did these tests to take 'transient response' issues that occur when using CPU Vcore Loadline Calibration (aka VRM loadline) out of the picture. Since 15K AVX small FFT passed at very high amps, the transients were splendid. But the problem is, at a certain load voltage that would have no problem passing prime95, Apex Legends runs into this strange timing bug.
On my own system, it seems like, at 5 ghz, this "bug" happens if the CPU on-die sense voltage drops below 1.270v.
So far, going back to "manual" voltages and tweaking the usual CPU Vcore loadline calibration (instead of auto or SVID offset enabled and messing with the AC loadline value), I found the following settings so far, seem to make Apex Legends stable:
5 ghz: 1.290v, Loadline Calibration=Extreme.
5 ghz: 1.310v, Loadline Calibration=Turbo
5.1 ghz: 1.360v, Loadline Calibration=Turbo.
Hopefully @OrioStorm manages to code a fix in for this issue soon, and someone at Intel makes a microcode update that fixes this permanently (If Oriostorm can convince the right people at Intel that this is an internal processor bug).
