thread # 5 billion about Intel Skylake/Coffeelake OC Voltage.

[K0NG]

Hi!
i got a really.. really good deal on amazon.
i Run a 8086k (delidded and with a h115i Pro 280 AiO) since january.
on my old cheap z370 tuf pro gaming was a 5.1 Ghz OC at 1.376v stable but the VRMs were at a 100C under load.

now i got a brand new "open box" ASRock Z390 Extreme4 on Amazon for 94€ (it costs around 160-180 normally.)

with the 10+2 Phase design i can OC without any Vdroop. (from idle to Cinebench R20 with exactly 0mv vdroop)

now is my question.  which OC i should use with that board.

5.1 Ghz and 4.9 Ghz Cache at 1.376v
Or
5.2 Ghz and 5 Ghz Cache at 1.408v

both is absolutely stable in realistic enviroments (not like pushing 280 watts out of 6 cores in Prime95...) but Aida64 Stresstests, long term gaming etc. (no BSOD since the OC was applied 10 months ago.)

i don't think that these 30mv would make a huge difference. (i don't need 1.376v for 5.1 but the problem is i can reach 1.344 or 1.376 and the cpu is only stable down to 1.36v which does drop to 1.344 and 5mv does 1.376v out of it.)

the CPU Power draw increases by around 15W and the temperature by 5°C (from around 55 to 60°C in Games).

 

[Fakmykak]

6 minutes ago, K0NG said:

with the 10+2 Phase design i can OC without any Vdroop. (from idle to Cinebench R20 with exactly 0mv vdroop)

Yeah "no VDroop". You getting your voltages from where, cpu-z? 

[K0NG]

2 minutes ago, Fakmykak said:

Yeah "no VDroop". You getting your voltages from where, cpu-z? 

HWInfo64
the Vcore is 100% stable and does not change.
from desktop to Aida64 Stresstest with CPU,Cache,FPU and Memory.

it sticks to 1.408v even after 30 minutes. (Bios voltage is 1.42v LLC1 (max))
 

[Jurrunio]

I'd just do 5.2GHz for the sake of it. Temperature of the CPU cores seem low most of the time

[Falkentyne]

41 minutes ago, K0NG said:

Hi!
i got a really.. really good deal on amazon.
i Run a 8086k (delidded and with a h115i Pro 280 AiO) since january.
on my old cheap z370 tuf pro gaming was a 5.1 Ghz OC at 1.376v stable but the VRMs were at a 100C under load.

now i got a brand new "open box" ASRock Z390 Extreme4 on Amazon for 94€ (it costs around 160-180 normally.)

with the 10+2 Phase design i can OC without any Vdroop. (from idle to Cinebench R20 with exactly 0mv vdroop)

now is my question.  which OC i should use with that board.

5.1 Ghz and 4.9 Ghz Cache at 1.376v
Or
5.2 Ghz and 5 Ghz Cache at 1.408v

both is absolutely stable in realistic enviroments (not like pushing 280 watts out of 6 cores in Prime95...) but Aida64 Stresstests, long term gaming etc. (no BSOD since the OC was applied 10 months ago.)

i don't think that these 30mv would make a huge difference. (i don't need 1.376v for 5.1 but the problem is i can reach 1.344 or 1.376 and the cpu is only stable down to 1.36v which does drop to 1.344 and 5mv does 1.376v out of it.)

the CPU Power draw increases by around 15W and the temperature by 5°C (from around 55 to 60°C in Games).

 

You should never use a 0 mv vdroop.  All it does is raise your peak to peak voltages and cause your minimum voltage (a transient that you can't see on sensors) to drop farther below the "RMS" voltage that you get on your sensors.  And it raises your heat output since the average voltage is higher also.  This issue gets far worse at heavier loads (programs that put a "heavier" 100% load on your chip, like Blender, Y-cruncher, Prime, CB R20, etc) than at lighter loads (x264/x265 stress test, Cinebench R15, etc).

 

The Dark (Maximus XI Apex has not been tested) gets the lowest transients of any board, meaning it can run stable at lower vcore than other boards)

 

Visual representation of a 0 mOhm vdroop here.

https://elmorlabs.com/index.php/2019-09-05/vrm-load-line-visualized/

But this was done on a Gene.  Apex would do better here.

 

Transients test #1 (hardware readings):

 

 

Transient test #2:

 

What people don't understand is that it's your minimum voltage floor that determines your stability, not your RMS.  If your RMS is too low you will insta-crash fast.  "Random instability" (like, being stable for 30 minutes then suddenly getting a CPU L0 error) is because of transients dips.  That's what people don't understand.

[K0NG]

10 minutes ago, Falkentyne said:

You should never use a 0 mv vdroop.  All it does is raise your peak to peak voltages and cause your minimum voltage (a transient that you can't see on sensors) to drop farther below the "RMS" voltage that you get on your sensors.  And it raises your heat output since the average voltage is higher also.  This issue gets far worse at heavier loads (programs that put a "heavier" 100% load on your chip, like Blender, Y-cruncher, Prime, CB R20, etc) than at lighter loads (x264/x265 stress test, Cinebench R15, etc).

 

The Dark (Maximus XI Apex has not been tested) gets the lowest transients of any board, meaning it can run stable at lower vcore than other boards)

 

Visual representation of a 0 mOhm vdroop here.

https://elmorlabs.com/index.php/2019-09-05/vrm-load-line-visualized/

But this was done on a Gene.  Apex would do better here.

 

Transients test #1 (hardware readings):

 

 

Transient test #2:

 

What people don't understand is that it's your minimum voltage floor that determines your stability, not your RMS.  If your RMS is too low you will insta-crash fast.  "Random instability" (like, being stable for 30 minutes then suddenly getting a CPU L0 error) is because of transients dips.  That's what people don't understand.

Wouldn't any overshoot would at least be visible after 3 hours of benching and gaming in HWinfo64?

The VCore is at Current/Minimum/Maximum/Average at exactly 1.408v and does not change or overshoot at least in these monitoring softwares.

if i use LLC 2 instead of one the Vcore drops to 1.376 and crashes within Seconds.

btw i use a fixed voltage and not adaptive/offset.
 

[K0NG]

at least this is what HWINFO spits out after over 3 Hours of different utilization. (Gaming, Benching, Idle.)

[Jurrunio]

1 hour ago, K0NG said:

at least this is what HWINFO spits out after over 3 Hours of different utilization. (Gaming, Benching, Idle.)

he's talking about the voltage ripple you need an oscilloscope to see.

 

2 hours ago, Falkentyne said:

You should never use a 0 mv vdroop. 

I'm almost certain that it's the overshoot LLC settings that shouldn't be used and 1.4v recommendation for max voltage is based on 0mv vdroop

[K0NG]

56 minutes ago, Jurrunio said:

he's talking about the voltage ripple you need an oscilloscope to see.

 

well that couldn't be so bad that it would kill or degrade a CPU fast enough at 1.42v ish.

i mean a friend of mine runs his 8600k at 1.48v fixed with maxed out LLC that actually overshoots to 1.49/1.52v (on a Z390 Gigabyte gaming X) to hold his 5.3 Ghz since it launched 2 years ago.

his rig runs 24/7 and gets rebooted once a week and he games like 3-4 hours a day.
the CPU seems to be running absolutely fine.. never complained about instability over the years or bsods/crashes.

[Falkentyne]

23 minutes ago, K0NG said:

well that couldn't be so bad that it would kill or degrade a CPU fast enough at 1.42v ish.

i mean a friend of mine runs his 8600k at 1.48v fixed with maxed out LLC that actually overshoots to 1.49/1.52v (on a Z390 Gigabyte gaming X) to hold his 5.3 Ghz since it launched 2 years ago.

his rig runs 24/7 and gets rebooted once a week and he games like 3-4 hours a day.
the CPU seems to be running absolutely fine.. never complained about instability over the years or bsods/crashes.

You are correct.  The ripple wouldn't degrade a CPU, as they only last for microseconds.  But if the ripple (the proper word is transients) would cause repeated dips below the minimum voltage needed for stability (at those temps, as lower temps =more stable), then this can cause random crashes or "CPU Cache L0 errors", or worse, a BSOD if hyperthreading is enabled or a BSOD (clock watchdog timeout) with HT disabled (L0 errors rarely happen if HT is disabled).

This is the reason why some people wind up stress testing and then even though temps haven't changed, they're "stable" for 5 hours doing the same instructions, then it randomly crashes in hour 6.  It's because of transients.  The transient issue gets worse at heavy current.  Trying to run small FFT AVX Prime95 with a 0 mohm loadline if you are not --rock stable-- (your load vcore far above the minimum needed) will often require a -higher- "RMS" load vcore to be stable, than if you reduced the loadline calibration and raised the bios voltage!

 

Here is proof of this on a Maximus XI Gene.

This is comparing LLC6 (0.4 mOhms LL) with LLC8 (0 mOhms loadline).

 

Notice LLC8 requires a much higher "VR VOUT", or voltage measured on the CPU die, than LLC6?

 

https://rog.asus.com/forum/showthread.php?109211-Question-about-Transient-Response-(to-Shamino-and-Raja)

 

Quote

 

I fired up the my Maximus XI Gene + 9900K to see if I could replicate your behavior.

Core = 4.7G
Cache = 4.4G

P95 29.1 FMA3 Small FFTs 15K

LLC=6, Vcore set = 1.130V, Vcore read = 1.066V: 1 thread failed after 6 minutes
LLC=6, Vcore set = 1.140V, Vcore read = 1.074V: pass 20m+

LLC=8, Vcore set = 1.075V, Vcore read = 1.074V: 1 thread failed after 2 minutes
LLC=8, Vcore set = 1.085V, Vcore read = 1.083V: 1 thread failed after 4 minutes
LLC=8, Vcore set = 1.095V, Vcore read = 1.092V: 1 thread failed after 2 minutes
LLC=8, Vcore set = 1.105V, Vcore read = 1.101V: 1 thread failed after 9 minutes
LLC=8, Vcore set = 1.115V, Vcore read = 1.110V: 1 thread failed after 6 minutes
LLC=8, Vcore set = 1.125V, Vcore read = 1.119V: 1 thread failed after 2 minutes
LLC=8, Vcore set = 1.135V, Vcore read = 1.137V: pass 1h+

I repeated it again with LLC=6, Vcore set = 1.140V, Vcore read = 1.074V and 1 thread failed after 14 minutes. Probably 10-20mV extra would pass for 1h+.

This seems even worse than what you reported. One aspect to think of is with higher voltages the temperatures will increase and worsen stability, resulting in even higher voltage required. It would be interesting with a direct comparison with the same CPU/cooling but different boards. I could possibly try to get an Aorus Master next week. I've got a Maximus XI Apex, but I doubt that would perform much better for this specific test case. Primarily the additional VRM components should yield lower VRM temperature.
 

 

 

Why does this happen?
Transients.  Dem transients.

 

Again it's not as big of an issue if you're gaming.  The transient problem gets -worse- the higher the amps draw is.

[Falkentyne]

2 hours ago, Jurrunio said:

he's talking about the voltage ripple you need an oscilloscope to see.

 

I'm almost certain that it's the overshoot LLC settings that shouldn't be used and 1.4v recommendation for max voltage is based on 0mv vdroop

The "recommended voltage" came from Asus, as a general consideration to avoid giving end users too much math to deal with under workloads most users experience.

But it's FAR more complicated than this.

 

Look at the Intel specification sheets.

https://www.intel.com/content/www/us/en/products/docs/processors/core/8th-gen-core-family-datasheet-vol-1.html

 

Go down to electrical specifications.  Do you see where it says "loadline slope within the VRM loop capability?".  That's the default VRM Loadline, or "Intel default" Loadline calibration.  Notice they don't specify anything for minimum.  That's because it's intended to be 1.6 mOhms (2.1 for 6 core processors).  A lower value means less vdroop, but they design their processors specification to be used with a certain vdroop.  

Raja (the Asus guy) told me in a PM that loadline calibration is a motherboard hack, made to cater to overclockers, and they didn't even want to give users high levels of LLC, but customer pressure made them go "let them eat cake", since that's what people wanted.  Intel's own answer for vdroop is the 'AC Loadline' setting and "Thermal Velocity Boost" voltage optimizations (which drops VID 1.5mv every 1C temp drop, starting at 100C, down to -150mv at 0C).

 

You'll also see max amp and VID there too.  All those values are related to each other.  It's a simple formula also.

These are for absolute maximum ratings (where the processor won't suffer long term or functional damage).

 

And this is based on auto voltage being used as well.

 

Max VID - ( R * I ) = maximum safe voltage (up to 100C).  We leave current as a variable since this will change at load.  And VID and resistance as fixed values.

Where max VID is in millivolts, R is resistance in milliohms, and amps is the current being used. So R is 1.6 mOhms and max VID is 1520mv and max amps: 193A.

So: VIDMAX - (R * I)=max load VR VOUT.

Or

1520 - (1.6 * I)= max load VR VOUT.

 

Since all values are known, you can simply plug that the amps value to get the load voltage that shouldn't be exceeded:

 

So let's say your average videogame draws 100 amps.

1520 - (1.6 * 100) = 1360mv (1.360v) <--intel spec, is the load voltage that should not be exceeded.

 

Using a little more logic, you can see where the "1.35v" recommendation comes from.

Since Level 6 LLC On Asus is 0.4 mOhms (also Turbo on Gigabyte) on 8 core CFL:

 

Let's substitute the max VID for the bios voltage and the "new" loadline and see what happens.

1.350v, LLC6, or LLC Turbo, or LLC2 (Asrock).

 

1350 - ( 0.4 * 100) = 1310mv = 1.310v.

So at 100 amps and a 0.4 mOhms VRM loadline, you're going to be using 1.310v VR VOUT.

 

Going back to the above formula, 1.310v is below 1.360v, so you're good.  But let's say your game uses 125 amps (Battlefield 5, etc).

1520 - (125 * 1.6) = 1.320v. <--Intel spec.

 

Cool, let's put in our 1.350v bios set with LLC6.

1350 - (125 * 0.4) = 1.300v.  <--just beyond intel spec.

 

I think this is self explanatory.  Now you see where "1.35v bios set+LLC6 came from" without making it complicated.  But now do you see how many people hate math?
No one wants to read all this stuff.  But Intel's engineers are not stupid.  They know full well what they're doing (ignoring Netburst).

[Jurrunio]

@Falkentyne The point of overclocking is to break specs doesn't it? Voltage behaviour you mentioned is pretty much what Intel CPUs do ever since they start getting high turbo clocks since Coffee Lake. the 1.4V number probably comes out from the logic that "ok we know you dont need this CPU to last 10-15 years, so this voltage is fine for the CPU to do at least 5 years and still be resellable".

 

My 2600k is evidence for that. The previous owner (my uncle) ran it on 1.47V I think and gave it to me for free after 5 years of use. Was capable of 5.05GHz, now would only do 4.75GHz at that voltage... I still kept it that way tho, I only use it 3 months per year so I dont mind if I'm putting a whole year worth of wear in 3 months

[Falkentyne]

1 hour ago, Jurrunio said:

@Falkentyne The point of overclocking is to break specs doesn't it? Voltage behaviour you mentioned is pretty much what Intel CPUs do ever since they start getting high turbo clocks since Coffee Lake. the 1.4V number probably comes out from the logic that "ok we know you dont need this CPU to last 10-15 years, so this voltage is fine for the CPU to do at least 5 years and still be resellable".

 

My 2600k is evidence for that. The previous owner (my uncle) ran it on 1.47V I think and gave it to me for free after 5 years of use. Was capable of 5.05GHz, now would only do 4.75GHz at that voltage... I still kept it that way tho, I only use it 3 months per year so I dont mind if I'm putting a whole year worth of wear in 3 months

Yes, but the reason it degraded was NOT because it was ran with a bios voltage of 1.47v.  It degraded because loadline specifications were violated at that voltage set.  If Intel vdroop (whatever it was for that platform) were respected, it would not have degraded at all.

The problem is you have to know what the specifications are, and what the "Motherboard hack" loadline calibration is doing to skew the default curves.

 

I mean I degraded two 2600k's.  I know what I'm talking about.

[Jurrunio]

17 minutes ago, Falkentyne said:

Yes, but the reason it degraded was NOT because it was ran with a bios voltage of 1.47v.  It degraded because loadline specifications were violated at that voltage set.  If Intel vdroop (whatever it was for that platform) were respected, it would not have degraded at all.

I know that with lowest LLC settings all boards run CPU vendor spec vdroop, so that means I should keep lowest LLC, and raise the base voltage all the way up to compensate for the vdroop? That will be 1.6V+ for my board I think

[alatron978]

8 hours ago, Falkentyne said:

The "recommended voltage" came from Asus, as a general consideration to avoid giving end users too much math to deal with under workloads most users experience.

But it's FAR more complicated than this.

 

Look at the Intel specification sheets.

https://www.intel.com/content/www/us/en/products/docs/processors/core/8th-gen-core-family-datasheet-vol-1.html

 

Go down to electrical specifications.  Do you see where it says "loadline slope within the VRM loop capability?".  That's the default VRM Loadline, or "Intel default" Loadline calibration.  Notice they don't specify anything for minimum.  That's because it's intended to be 1.6 mOhms (2.1 for 6 core processors).  A lower value means less vdroop, but they design their processors specification to be used with a certain vdroop.  

Raja (the Asus guy) told me in a PM that loadline calibration is a motherboard hack, made to cater to overclockers, and they didn't even want to give users high levels of LLC, but customer pressure made them go "let them eat cake", since that's what people wanted.  Intel's own answer for vdroop is the 'AC Loadline' setting and "Thermal Velocity Boost" voltage optimizations (which drops VID 1.5mv every 1C temp drop, starting at 100C, down to -150mv at 0C).

 

You'll also see max amp and VID there too.  All those values are related to each other.  It's a simple formula also.

These are for absolute maximum ratings (where the processor won't suffer long term or functional damage).

 

And this is based on auto voltage being used as well.

 

Max VID - ( R * I ) = maximum safe voltage (up to 100C).  We leave current as a variable since this will change at load.  And VID and resistance as fixed values.

Where max VID is in millivolts, R is resistance in milliohms, and amps is the current being used. So R is 1.6 mOhms and max VID is 1520mv and max amps: 193A.

So: VIDMAX - (R * I)=max load VR VOUT.

Or

1520 - (1.6 * I)= max load VR VOUT.

 

Since all values are known, you can simply plug that the amps value to get the load voltage that shouldn't be exceeded:

 

So let's say your average videogame draws 100 amps.

1520 - (1.6 * 100) = 1360mv (1.360v) <--intel spec, is the load voltage that should not be exceeded.

 

Using a little more logic, you can see where the "1.35v" recommendation comes from.

Since Level 6 LLC On Asus is 0.4 mOhms (also Turbo on Gigabyte) on 8 core CFL:

 

Let's substitute the max VID for the bios voltage and the "new" loadline and see what happens.

1.350v, LLC6, or LLC Turbo, or LLC2 (Asrock).

 

1350 - ( 0.4 * 100) = 1310mv = 1.310v.

So at 100 amps and a 0.4 mOhms VRM loadline, you're going to be using 1.310v VR VOUT.

 

Going back to the above formula, 1.310v is below 1.360v, so you're good.  But let's say your game uses 125 amps (Battlefield 5, etc).

1520 - (125 * 1.6) = 1.320v. <--Intel spec.

 

Cool, let's put in our 1.350v bios set with LLC6.

1350 - (125 * 0.4) = 1.300v.  <--just beyond intel spec.

 

I think this is self explanatory.  Now you see where "1.35v bios set+LLC6 came from" without making it complicated.  But now do you see how many people hate math?
No one wants to read all this stuff.  But Intel's engineers are not stupid.  They know full well what they're doing (ignoring Netburst).

What kind of degradation may you experience if you were to run a 1 mOhm loadline at 193A giving you 1327mv on an 8 core intel part?

[Falkentyne]

7 hours ago, Jurrunio said:

I know that with lowest LLC settings all boards run CPU vendor spec vdroop, so that means I should keep lowest LLC, and raise the base voltage all the way up to compensate for the vdroop? That will be 1.6V+ for my board I think

No, not at all.  Just saying a very aggressive LLC will make a chip degrade faster than a less aggressive LLC, when using higher voltages.

And using 1.6v with intel spec vdroop will make your chip degrade regardless, unless you use sub ambient cooling.

1.350v with a medium amount of LLC is a fair balance.

1.40v with maximum LLC is a very bad thing.

[Falkentyne]

4 hours ago, alatron978 said:

What kind of degradation may you experience if you were to run a 1 mOhm loadline at 193A giving you 1327mv on an 8 core intel part?

Probably pretty substantial if you are on ambient cooling.  Plus anyone without direct die and double or triple 360 rads would not be able to cool a 1.327v load VR VOUT chip at 193 amps and keep it under 100C to begin with.

[Falkentyne]

Anyway here is a very easy to understand video on why a too high loadline calibration is bad and why vdroop is your friend.

 

 

[K0NG]

5 hours ago, Falkentyne said:

Anyway here is a very easy to understand video on why a too high loadline calibration is bad and why vdroop is your friend.

 

 

but what is better for me?
1.45v in idle that drops to 1.408 with the second highest LLC

or 1.408 in idle which never drops at all (atleast in software readings)

btw here is my HWINFO after 2 hours of Ground War in COD MW without Framecap

[Falkentyne]

57 minutes ago, K0NG said:

but what is better for me?
1.45v in idle that drops to 1.408 with the second highest LLC

or 1.408 in idle which never drops at all (atleast in software readings)

btw here is my HWINFO after 2 hours of Ground War in COD MW without Framecap

0 mOhm loadline is just plain bad regardless.

And I still wouldn't use the second highest LLC on any board without top tier VRM's, especially at that voltage.  The transients are going to be almost as awful.

I would only trust 1.45v as a bios voltage without *ANY* LLC at all.  You're risking slow degradation with any LLC besides one of the lowest levels.  I explained the math before (that 6 core CFL was designed for a 2.1 mOhms loadline so there is a voltage/amps curve which shouldn't be deviated too much from on the high amps/vcore side (lower sides are always fully safe).

 

Since Asrock's LLC is inverted (compared to Asus' levels), I would not go above LLC3 with that high bios voltage.  And I would only use LLC2 at a much lower bios voltage (not higher than 1.30v).  I'm not even fully sure if Asrock's LLC1 is "0 mOhms" or not.  Check the buildzoid video to find out.  But it's your CPU.  You can do what you want with it.  And not all CPU's are built equal.  Some can handle more abuse than others.

[K0NG]

9 hours ago, Falkentyne said:

0 mOhm loadline is just plain bad regardless.

And I still wouldn't use the second highest LLC on any board without top tier VRM's, especially at that voltage.  The transients are going to be almost as awful.

I would only trust 1.45v as a bios voltage without *ANY* LLC at all.  You're risking slow degradation with any LLC besides one of the lowest levels.  I explained the math before (that 6 core CFL was designed for a 2.1 mOhms loadline so there is a voltage/amps curve which shouldn't be deviated too much from on the high amps/vcore side (lower sides are always fully safe).

 

Since Asrock's LLC is inverted (compared to Asus' levels), I would not go above LLC3 with that high bios voltage.  And I would only use LLC2 at a much lower bios voltage (not higher than 1.30v).  I'm not even fully sure if Asrock's LLC1 is "0 mOhms" or not.  Check the buildzoid video to find out.  But it's your CPU.  You can do what you want with it.  And not all CPU's are built equal.  Some can handle more abuse than others.

Uhm.. my Vdroop without LLC at 1.408v goes down by over a 160mv that would be 4.7-4.8 GHz for me instead of 5.2.. then i could run my CPU at stock either...  that's not the point of overclocking.
 
i totally agree that a 0 mOhm LLC is stupid but why should i use nothing above 1.3v with LLC at level 2? That sounds paranoid. It's Called overclocking.. not running at stock voltages and hope the CPU does not blow up before we've built our first 2 Cities on Mars. There are several 2 digit Millions of Coffeelake CPUs out there and i never heard something about any kind of degradation or dead CPUs through running the CPU at stupidly high voltages and LLC. i've seen one post on reddit a while ago where a dude is running his 3770k at 2V! since a half decade. and the CPU runs perfectly fine at around 4.8 GHz.

All i can do now is 1.425v at Level 2.
That does drop to 1.376v under load and stays there. (50mv vdroop is pretty fine instead of nothing)

i'll stick to LLC level 2 at 1.425v and save this thread.
If the CPU Degrades in the next years or even dies.. i will post it here.