Confusion about overclocking voltage
Confusion about overclocking voltage
I am adjusting the text to maintain the original length and structure while changing the wording. The revised version preserves all key details and intent.
Which BIOS version are you running? What aftermarket cooler do you have? Ideally, include full details such as: CPU, Motherboard, RAM, SSD/HDD, GPU, PSU, Chassis, OS. Also let me know your airflow configuration—I believe your voltages are high, so using the BIOS voltage numbers as a reference would be best. You might want to check HWMonitor for confirmation.
BIOS version is P3.30 for AsRock mobo (think its the latest as I updated a few weeks ago).
The airflow in my case is set up with two Noctua NF-A14 chromax fans as intake on the front, and two Fractal Design Dynamic X2 GP-12 120mm stock fans (one at the back end, and one directly on top of the cooler - both as exhaust).
CPU: Intel i5 8600k
CPU Cooler: be quiet! Dark Rock Pro 4
Motherboard: AsRock Z370 Extreme4
Ram: HyperX Fury 2x4GB 2400MHz DDR4
SSD: Samsung 850EVO 120GB for OS drive
HDD: ST3500620AS (I really don't know the hdd, it's really old from a computer ages back)
GPU: RTX 2080 ASUS Strix Advanced Edition
PSU: Corsair RM750x (2018)
Chassis: Fractal Design Meshify C
OS: Windows 10 Pro
I just don't know if its a RealBench problem, but it seems that with offset setting, Realbench draws less voltage to run than Prime95. VID sometimes jump to 1.319 while the highest vcore with the offset setting stagnates at 1.3v12.
VID represents the voltage the CPU instructs the motherboard VRM to provide. Vcore is the actual voltage consumed by the CPU. Apply offset at idle, not load voltages. For instance, if your idle voltage is 1.256V and your VID is 1.3461, your offset is -0.0901. Rounded to the nearest 0.005V, it turns into 0.10V. This indicates a negative offset, so you need to switch to the negative offset mode sign to subtract 0.10V from 1.3461V.
It's usually preferable to select the lowest LLC setting possible, particularly when using offset. High LLC settings may lead to large voltage overshoots during load changes (the significance of this depends on interpretation), and more importantly, they can result in insufficient offset, causing crashes during idle or light tasks. Therefore, a lower LLC with a slightly higher offset is typically optimal.
Various applications can adjust voltages dynamically. RealBench is built to handle this by simulating different loads for CPU, RAM, and GPU, making it suitable for stability testing. P95 performs less well here because its loads are more static, especially during small FFT tasks, and it doesn't penalize the entire system—just the CPU and some RAM usage.
Ensure the motherboard chipset drivers are up to date. Windows 10 CE updates may struggle with older or lower version driver versions.
The HDD in use is a Seagate Constellation ES.2, which is an enterprise-grade drive, offering solid performance.
Wouldn't both vid and vcore behave identically when idle? I attempted the change but didn't notice any variation (also, numerous background elements cause frequent jumps whereas under load it remains stable). I assumed you wanted vcore above VID to achieve a positive offset, but I'm still unclear about this point.
I understand your concern, and I agree on this aspect. Asrock's LLC appears distinct from ASUS or Gigabyte's models. Their top tier doesn't seem robust and doesn't exceed voltage limits as much—actually, it still shows some voltage droop even during stress tests. The second tier (level 2) seems to have excessive droop, requiring a higher voltage increase than desired.
I rely on RealBench for my primary stress testing, but I'm having trouble locating an offset that would pass this test, despite knowing a fixed setting succeeded after 8 hours of testing.
Under loads voltage typically decreases. Aim for vid and vcore to stay nearly identical, ideally within 0.05v. With a high vid offset you essentially waste power on the VRM. If vid was 1.3461 and vcore 1.256, applying a -0.1v offset would lower vid to 1.2461v while keeping vcore at 1.256v. Offset values can be negative and idle vid is usually higher than load vid. With LLC it behaves differently: when first connected a load, voltage drops below the set vcore level. This can be mitigated by raising vcore, though it increases heat. LLC provides a small extra voltage when needed to preserve vcore. 0% adds minimal voltage, 100% tends to overvolt aggressively, slightly at load and more so during idle. When idle is disabled, aggressive LLC works well; with idle enabled it can disrupt stable idle voltages. For most systems, 50% is a reasonable starting point, adjusting as needed. Adding unnecessary voltage only raises temperatures unnecessarily.
It appears you're focusing on incorporating high voltage in various areas. You're clarifying that this isn't necessary for all components. The readings should reflect idle conditions without loads running, not under heavy stress. Adjusting the voltage reference helps stabilize performance when sudden changes occur. Higher operating voltages mean more offset is needed to maintain balance. Excessive offset only helps during intense loads, increasing overall voltage and causing larger corrections. Aim for close matching between voltage and core voltage to minimize unnecessary adjustments. Depending on the situation, a negative or positive offset can be applied accordingly.
Are you happy with BIOS 3.30 with regard to overclocking?
Did you use a previous version?
Is 3.30 more stable with same OC settings?
I have the same board, delidded 8086k.
I'm on BIOS 3.10, and to preface this, I won't likely buy another AsRock board.
They have the worst customer service, they don't answer in their own forums and even sub forums like here and tweaktowns, they ignore their customers.
Their adaptive voltage is absolute trash when you are looking to push a CPU to it's max (delid etc), and when you are pushing 5.3-5.4Ghz (what I'm running on the 8086K, all cores, no AVX offset), so fixed LLC1 is really the only way to get stable, at least on BIOS 3.10.
I use small 40mm fans to actively cool the VRM/MOSFET's, which is important when you water cool, you have Noctua air cooler so you are good
That being said, there is something you might want to look at in your OC quest.
Using HWInfo64 you will see that this board has 2 entries (in sensors) for Renesas ISL69138 controller.
One of these sensors is wrong, you can see that it's wrong, so hide it.
You can easily tell which one is correct as the first value 'VR Loop1' will correspond to a CPU Value 'VR VCC Temperature(SVID)' - this is essentially the VRM/MOSFET temperature sensors.
If you have a 'fixed' vcore in BIOS, the value below 'VR Loop 1', a value named 'VR VOUT' will be very close to the 'vCore' setting.
This 'VR VOUT' value is essentially a more accurate reading of the real voltage that the core is requesting from the VRM/MOSFET's - there's a lovely technical reason for this which involves resistance in circuitry that means 'vCore' is not the most accurate reading, and that's really all you need to know.
The author of HWInfo and AsRock have confirmed this sensor is indeed correct and should be trusted as a more accurate reading of core voltage.
As other's have pointed out (kinda in a roundabout way), if we are looking at fixed vCore, due to vdroop it's impossible to have the same fixed idle voltage as under load voltage.
And the higher the OC, the more voltage you need to push.
With Fixed vCore and LLC1, you will typically see idle voltage about 30mv higher than full load, such as stress test with Realbench (also my favorite).
So let's say my idle vCore at 5.3Ghz is 1.415v, under load the vCore value in HWInfo will pretty much stay the same, with occasional dips to 1.406v.
BUT, the VR VOUT value will be in the 1.386v area, which shows just how much variance there can be, and why even the tightest LLC on this board is still trash.
High voltage at idle is NOT dangerous in terms of killing your CPU - within reason of course. If you have nice cooling, you can sit at 1.45v at idle, with 1.41-1.42 under full load.
High idle voltage will barely draw any current (amps) or power (watts) - Coffee Lake processors are good for 138A/180W continuous, which even under stress testing conditions is very unlikely. You might see spikes, but not constant.
On the other hand, what can damage or degrade a CPU is High Voltage combined with High Current and High Power,.
This is why I've always used fixed voltages, and always turn off all C-States, Speedstep(EIST)/Speedshift - my PC's are primarily for high performance gaming, so I could care less about saving a few dollars a year, and quite frankly the environment has WAYYY bigger issues than me drawing a few more amps per month than another PC with max power saving settings.
There's many who have the opinion that higher idle voltages kill CPU's, but if you have good cooling and keep those temps in check (I'm at max 72C after 4 hours Realbench) then you will replace your PC long before your CPU degrades.
Now, if you are running vCore at 1.5v, then you will likely see degradation during the life of the PC, but even then it's still not a fixed science.
Stay under 1.45v idle and good cooling and if you replace your PC every 3-5 years you will most likely have zero issues.
I have never had a CPU die on me, 30 years of building PC's - and I'm not alone
So, this may help you feel a little more comfortable about raising the 'vcore' value in BIOS, because your under load value on this board is going to be quite a bit lower, and that's the one you really need to care about.
BTW - would love to hear if you like the new BIOS better.
I'm thinking of going back to 1.20 before all this Spectre microcode crap that has impacted performance, but most likely tossing this board in the trash.
Cheers!
5.3 out of 5.4 all core performance is quite impressive, particularly at 72°C after four hours of handbrake use. Yes, I think it would be better if the mobo community agreed on standard terms. Using consistent labels like vcore (voltage used by core), VID (voltage in demand), VRM temperature, and CPU temperature would make things much clearer. If we just talked about VR loop 1, VR vout, or temporary values, it would be confusing unless you were directly addressing someone.