ASRock Extreme 4 Vcore Reading

FreeStream · 13 Jul 2018 at 19:05

I've now delidded my 8600K thanks to @Awahwah lending the tool out to forum members. Sadly I need to be AVX stable as Camera Raw (Lightroom) uses AVX and AVX2 in the rendering engine.

Net result is I can pass IBT for 10 loops using 8GB memory now without throttling due to one core dropping 30C(!) and on average 20C. (~200W load)

Board: ASRock Z370 Extreme 4
Memory: Corsair CMK16GX4M4A2133C13 Vengeance LPX 16 GB (2x8 GB).
Cooler: Noctua NH-D15 (Twin fan config)

Settings in the BIOS are:

CPU Multi: 48
Cache Multi: 40
Vcore: Dynamic +60mv
LLC: Level 1 (Highest)
VCCIO: 1.11V
VCCSA: 1.11V
Mem: 3200MHz, 15-17-17-34-540, 1.35V. (2 passes Memtest 86+)

CPU-Z/HWINFO shows Vcore of 1.41V but the multimeter on the PWM shows 1.49-1.50V(!).
Load temps are ~75C.

I'm now testing at 4.9Ghz (Prime95 non-AVX) with +70mv which shows 1.34V in CPU-Z and 1.41V on the multi-meter.

I'm slightly concerned at the differential between what the motherboard is reading Vcore to be and what it is actually being given.

So the question is (might be one for @8-Pack):

A) Are people giving their CPUs more V than they think?
B) People with 5GHz+ CPUs running 1.4+V Vcore, have you actually measured your physical Vcore?

The measurement points on a Asus Z77 board I have reads +-15mv of the CPU-Z reading, the ASRock Extreme4 is WAY off.

Thoughts??

LuckyBenski · 13 Jul 2018 at 22:31

I'd speculate that moving to physical voltage readouts will have to be considered separate from the wealth of data on overclocking online. Basically, almost all data other than from extreme overclockers was gleaned via motherboard readings in software. If there is some bias, variation or inaccuracy - we have to let it be what it is. It's risky to assume that the "standard" known Vcore etc ranges are real-world numbers, and then apply that to physical readings. Just my 2p as an electronic engineer who's only dabbled in PC overclocking.

I'm quite surprised you're pulling 200W through that i5! Is that typical of this generation?

FreeStream · 14 Jul 2018 at 10:24

LuckyBenski said:
I'd speculate that moving to physical voltage readouts will have to be considered separate from the wealth of data on overclocking online. Basically, almost all data other than from extreme overclockers was gleaned via motherboard readings in software. If there is some bias, variation or inaccuracy - we have to let it be what it is. It's risky to assume that the "standard" known Vcore etc ranges are real-world numbers, and then apply that to physical readings. Just my 2p as an electronic engineer who's only dabbled in PC overclocking.

I'm quite surprised you're pulling 200W through that i5! Is that typical of this generation?

I agree that physical vs software readings are different but at the end of the day people are basing their overclocks on the software figures, so may unwittingly be pushing 1.5v+ thorough the chip.

Especially as Intel deam the max to be ~1.52V.!!

As an engineer I was just curious when I was testing what the PMWs were outputting. I was really concerned that a ‘safe 24/7’ Software voltage of 1.41 was actually 1.5V!!

Now I’ve changed my LLC and Vcore offset such that I’m stable at 5GHz (non-AVX prime95).

Prime95 Software/Physical Voltage: 1.36/1.44
AVX Software/Physical Voltage: 1.41/1.49.

And yes these chips at ~5GHz under intense AVX draws ~210W.

Prime95 about 165W
Lightroom about 140W

It seems even under “light” AVX load my chip gets physically 1.5V.

However temperatures are below 60 and the current draw is ‘low’ so maybe it’s OK?

FreeStream · 14 Jul 2018 at 16:30

Dropped back to 4.9Ghz, LLC2, +100mv (Physical 1.386V, HWINFO sweeps between 1.312 and 1.328V.) until it is determined what voltage I should actually be overclocking to...

acemastr · 16 Jul 2018 at 13:33

Remember that voltage doesn't kill chips, temperature does. Personally I'd go by what the motherboard reads as that's what everyone uses.

Why are you using LLC2 by the way? Generally most use LLC6 (myself included)

FreeStream · 16 Jul 2018 at 18:03

acemastr said:
Remember that voltage doesn't kill chips, temperature does. Personally I'd go by what the motherboard reads as that's what everyone uses.

Why are you using LLC2 by the way? Generally most use LLC6 (myself included)

Because from my testing, lower levels of LLC enables me to push the 1/2/3 core loaded multi up compared to all 6 cores and lower the stable 6-core load voltage from 1.5 to 1.48V (multimeter).

Note that Level 1 is highest on ASRock which is the other way round to others it seems.

Currently running:

Ring Ratio: 42
Offset: +160mv
LLC: Level 4
CPU Ratios: 51/50/50/49/49/48

IBT Thread Count - 1/2/3/4/6 (30 mins minimum stable)
Load Voltage - HWINFO: 1.39/1.39/1.38/1.39/1.39
Load Voltage - Physical: 1.41/1.42/1.43/1.45/1.48
10 Run Average GFlops: 37.8/74.0/109.5/143.3/204.5

Essentially now whatever the load on my chip, be it 1 or 6 threads its running as fast as it can.

Temps hit 70C (220W) AVX load, but generally in lightroom 55C so I'm happy leaving it loading to 1.5V (Physical).

Rubbish chip really as even if I just do non-AVX prime95 it falls over at 4.9ghz, 1.41V (software), 60C.

acemastr · 16 Jul 2018 at 18:16

It might be worth emailing ASRock to confirm, that would be a serious issue if the actual voltage is significantly higher than the motherboard reads.

Does the ASRock have dedicated points on the motherboard to check voltage?

FreeStream · 16 Jul 2018 at 18:36

acemastr said:
It might be worth emailing ASRock to confirm, that would be a serious issue if the actual voltage is significantly higher than the motherboard reads.

Does the ASRock have dedicated points on the motherboard to check voltage?

Nope - Reading the output from the PWM on the back of the board.

Have contacted them now though...

LuckyBenski · 16 Jul 2018 at 18:45

FreeStream said:
Nope - Reading the output from the PWM on the back of the board.

How well calibrated is your meter?

FreeStream · 16 Jul 2018 at 18:50

LuckyBenski said:
How well calibrated is your meter?

Never calibrated a MM in my life - I would not expect a device designed to measure DC voltage to go 7% out of spec...

However, according to the spec sheet, accuracy is to "±1% + 2 Digits".

Besides, the motherboard records roughly the same voltage across a load range whilst the multimeter doesn't.

For the record it measures the 3.3V, 5V and 12V rail to within 0.01 of what the motherboard says they are. In addition, on a Asus Maximus Gene V I've tested recently it, using the supplied reading points it gives a difference +-0.02 from what the software says.

It *may* be that the place I'm measuring from is incorrect, but it is the output side of the PWM so I don't see how it can be that wrong!

LuckyBenski · 16 Jul 2018 at 18:59

I'm wondering whether it's down to a PWM voltage being read and averaged by a meter across several samples. But then I don't know what meter you're using

I often use cheap and cheerful meters for most work but for a PWM controlled CPU supply accurate to millivolts, I might we'll look to something more well made.

Plus I/we don't really know the topology of a CPU and its regulation circuitry. The output of the MOSFET might have some distribution to the cores, and we don't know where it's read from, etc. Just speculating

FreeStream · 16 Jul 2018 at 19:26

I agree that PWM could be throwing it off, but again the CPU should be receiving a fairly clean PWM (300khz?) so I don't think my MM would be sampling so badly as to get the reading wrong :confused:

This is essentially a toss up between what a device designed to measure voltages says and the chip and software combo on a motherboard.

I'd be inclined to believe the multimeter personally. I'll ask around to see if I can borrow a "better" MM.

LuckyBenski · 16 Jul 2018 at 21:14

You can believe the meter for a real voltage but if every motherboard ever made has relied on software readings... ultimately those are the relevant numbers even if inaccurate. One of the dangers of knowing how to measure things!