z370 Taichi/8700k OC voltages too high

delatroy wrote: Looks like I have the same issue: Are you also seeing differences in voltages between CPUz, Core Temp and A-Tuning at idle and under load? I'm on the 8700k, Taichi @5ghz LL1 @ 1.32v in bios and I get the following peak voltages under load: * CPUz 1.328v * Core Temp 1.4191 * A Tuning 1.318v That's huge variance over bios even with ASRock LLC1's negative vdroop. Is this a bug? What's going on? Is any correct Why are they all different?

First, VID is NOT the actual VCore. They are two different things.

Core Temp labels the reading as the VID. VID is the easiest thing to read from an Intel CPU, since it comes from an internal register whose ID/location is known and easily read. VID is the voltage requested by the processor from Intel's built in VID table, for the core speed, CPU load, etc. Note there is a VID for every core, but whichever core requests the highest VID voltage, that is what will be provided to the processor.

VID voltage is usually if not always more than the actual VCore required by the cores in a CPU. Intel over compensates the VID voltage values to insure the processor will always operate.

The actual VCore voltage being used by the board is normally always lower than the actual VCore voltage being provided to the CPU. If your board has CPU VRM voltage sensor outputs, only shown by a few programs (HWiNFO64 will show CPU VRM output voltage if the board provides it), you'll see that voltage is below that of the VID, and very close to the actual VCore.

Reading the actual VCore provided to a CPU takes specific programming for EVERY make of board, and possibly different between different models of the same make of board. That's why you see just the VID being displayed by some programs, it's easy to do.

Facts about monitoring programs:

Is the VID or "VCore" reading a continuous, analog meter type of reading? No, it is a sample reading done at specific programmed intervals, such as once every two, three, or five seconds. Each monitoring program has its own VID/VCore sample interval programmed into it. HWiNFO64's sample interval can be set by the user, the default is 2,000ms/2 seconds.

When running multiple monitoring programs at the same time, do they have their sample intervals synchronized? How could they be synchronized, when they can have different intervals between samples?

Next, if two monitoring programs try to read a sample at the same time, for some reason that results in one or both programs receiving bad data. That is just a reality of the sensor chips used in mother boards to provide readings like VCore. So monitoring programs that are programmed correctly use a read instruction that waits its turn to get the data. So we are assured that the sample read by each program is not read at exactly the same time.

Adding these three things together, results in the differences between readings of multiple monitoring programs. Include a VID reading instead of the true VCore, we have another difference.

Now consider the CPU itself. Is it using the same VCore constantly? It may or may not. If any of the CPU performance and/or power saving options are enabled (Turbo, Power Saving Mode, SpeedStep, C1E, C States, etc), and the Windows Power Plan is not set to High Performance, the VCore will change constantly, except when under a constant load. In this scenario, the VCore will change thousands of times a second. Multiple monitoring programs each reading this dynamic VCore at different times will result in different readings. This does not take into consideration the CPU Cache Ratio, Long and Short Term Power and Current option settings, etc, which all affect the VCore.

How do we get a static (or as close as possible), non-changing VCore? Only by using Fixed Voltage as the VCore mode, none of the CPU power saving features enabled, Turbo not active by an all core OC at or beyond the max rated Turbo frequency, LLC set to the highest level, and the Long and Short Term current and power setting set high enough, etc. With  the Windows Power Plan set to High Performance, of course, and a high quality PSU whose +12V rail will not droop much under high loads.

Which is correct? They all probably are, for the reasons I just explained. When each program read the VID or VCore, that's what it was when it took its sample, each at a slightly different time in human terms, but in CPU terms, very far apart. Core Temp read the VID, the other two the actual VCore, each in turn.

We'd like to believe that a VCore reading is simple just like the speedometer reading in our cars. But it is for more complicated than that. By the time we see it on the program's display, it has changed many, many times, unless we have it locked down statically.

Used Conductanaught on both die and lid, yes. Using same cooler D15 in a Carbide Air 540 case... basically the best airflow case ever made.

I can go x49 with 1.36v (temps in high 80s) and x50 with 1.40v but temps are in the mid to high 90s. Voltages any lower than this fail in Prime95 torture test about 8 minutes in.

I am starting to think my 8700k is a dud... I have a spare 8600k, maybe I should be trying that instead.


Edited by N.I. - 07 Nov 2017 at 4:07am

Thanks japau, appreciate the info.
I stopped using HW64 because the vcore voltages were always ".665" volts so I could never tell what they really were. I have switched to CPU-Z and that seems to give accurate readings.
I am on 1.17 BIOS.
Right now I am stable at x48 with 1.315v and highest temp is 81c with Prime95 torture test.

I will try your above settings and see what I get, thanks again.




Edited by N.I. - 07 Nov 2017 at 1:52am

I reset the BIOS back to default, disabled multi core enhancement, set Windows to performance mode and HW64 is reporting an idle voltage of 1.395. Prime965 torture test spikes that to 1.414.

Went back into BIOS and set multiplier to 47 and voltages are idle 1.395 and spike to 1.519.

Can anyone tell me how to reduce voltages? Just going in and setting a manual voltage to 1.35 or whatever has no effect whatsoever.

Screen shot of 47 locked speeds with default voltages.  This is identical to the voltages if I go in and manually set the voltage to Fixed and 1.whatever...

N.I. wrote: BIOS version 1.17 Z370 Taichi 8700k dellided/metal TIM Win 10 set to Balanced power mode If I set the CPU to 45 locked on all cores and set LLC to 1 and a fixed voltage of 1.200v I get min voltages of .665 but sustained voltage under load of 1.46.  If I set the LLC to anything other than 1 the voltages go higher (level 3 got to 1.54).  Load temps are in the mid 60s. If I lock the CPU at 4.8 on cores, LLC 1, voltage at 1.200 I get the exact same results... .665 to 1.46 65c. If I lock the CPU to 50 with everything the same I get same voltages but temps in the low 90s. Is there some setting I am not seeing to get the voltages down under 1.4? This is really frustrating. Forgot to mention... I reset BIOS back to default and I am getting VID .665 to 1.414. So basically this board seems to have its factory voltages set WAY too high... this is with Multicore Enhancement set to Disabled

If you're monitoring VIDs, that is not the true VCore. VIDs are normally much higher than the true VCore, and are not the actual voltage being used by the CPU. What are you using to monitor VCore? Is that the just in the UEFI, or in Windows?

VID is what the processor's VID table believes it needs, which is not necessarily (or at all) what the VCore needs to be. From experience with an i7-7700k, once you OC above 4.7GHz, the VCore requirement takes off. BTW, the highest VID programmed into these processors is 1.52V, and I've seen users run them at that voltage, suicide runs of course.

I suggest HWiNFO64 to monitor both VIDs and the actual VCore:

https://www.hwinfo.com/download.php