CPU Overclocked - VCore voltages varies depending on the load

[katulen]

Hey guys.

 

I've managed to get my overclock stable with adaptive voltages, however I have experienced some weird behavior which im trying to find an answer to.

I hope you can help me.

 

Here is a quick brief of my OC settings:

Adaptive voltages: 1.350v

Offset 0 (Auto)

AVX: Auto

LLC: Level 5

 

When running Realbench, my VCore is 1.350v (no vdroop).

When running Prime95 nonAVX my VCore is 1.332v (vdroop).

When playing games, e.g. CS:GO or PUBG, my VCore is 1.368v (vboost)

 

So is the reason behind this voltage behavior, caused by the type of load that is put on the CPU ?

E.g. stresstesting using Realbench, which uses AVX instructions, is causing my CPU to need more voltages, and therefor sits on 1.350v.

Where running Prime95 nonAVX causes my CPU to be fine with 1.332v, due to no AVX instructions.

And with games, where my CPU load is 40-50%, needs even less voltages to be stable, and therefor makes a slight vboost.

 

Or perhaps the answer is due to having Offset set to auto ?

 

I don't know if above is the cause of this, I do feel it could be a solution, but i am absolutely clueless on this tho.

 

Please correct me if I am wrong. Hoping to find an answer to this behavior.

[Jurrunio]

7 minutes ago, katulen said:

caused by the type of load that is put on the CPU ?

Yes

 

7 minutes ago, katulen said:

E.g. stresstesting using Realbench, which uses AVX instructions, is causing my CPU to need more voltages, and therefor sits on 1.350v.

Where running Prime95 nonAVX causes my CPU to be fine with 1.332v, due to no AVX instructions.

And with games, where my CPU load is 40-50%, needs even less voltages to be stable, and therefor makes a slight vboost. 

Er, the opposite. More stressful load = more power draw = more voltage drop. That's what loadline calibration is for: compensating for the voltage drop

[katulen]

1 minute ago, Jurrunio said:

Er, the opposite. More stressful load = more power draw = more voltage drop. That's what loadline calibration is for: compensating for the voltage drop

Make sense.

Well that applies with the game part, where its not really under any load, and therefor a slight vboost occur. But it doesn't really apply to Realbench vs. Prime95 nonAVX then ? Cause its the straight opposite in my situation right. Or is Prime95 just more stressful vs. Realbench, even tho the Prime95 i use is nonAVX ?

[Jurrunio]

Just now, katulen said:

Make sense.

Well that applies with the game part, where its not really under any load, and therefor a slight vboost occur. But it doesn't really apply to Realbench vs. Prime95 nonAVX then ? Cause its the straight opposite in my situation right. Or is Prime95 just more stressful vs. Realbench, even tho the Prime95 i use is nonAVX ?

Not sure. Maybe you're already using some level of LLC so it's not significant.

[katulen]

Just now, Jurrunio said:

Not sure. Maybe you're already using some level of LLC so it's not significant.

Hmm okay. But for the gaming part of the "issue", it is not something I should be worried about ?

And could it potentially be "fixed" by using a negative voltage offset ? It iches my eye to see i'm having a vcore at 1.368v x'D

 

And add to that, I assume using a fixed voltage instead, would proberly fix all the "issues" combined.

[Jurrunio]

1 hour ago, katulen said:

Hmm okay. But for the gaming part of the "issue", it is not something I should be worried about ?

No, just watch the temperatures

 

1 hour ago, katulen said:

And could it potentially be "fixed" by using a negative voltage offset ? 

Use more LLC to get rid of the drop under load. This will lead to higher voltage all the time, so lower offset after that. If you have time you should use adaptive+offset.

 

1 hour ago, katulen said:

assume using a fixed voltage instead, would proberly fix all the "issues

No, the problem will just get worse as it uses high voltage at all times

[katulen]

26 minutes ago, Jurrunio said:

Use more LLC to get rid of the drop under load. This will lead to higher voltage all the time, so lower offset after that. If you have time you should use adaptive+offset.

Okay. So the thing is, with LLC 5 i get vdroop, with LLC 6 i get vboost. So your suggestion is to e.g. use LLC 6 to get vboost, but adjust offset accordingly to that? Problem is tho, if im on 1.368v when playing games, with LLC 6 i will probably have around 1.380-1.390v. So I will need to use a negative offset of 0.030-0.040v. Is that how you mean or? 

If so, would it change anything to do it the otherway around ? 

 

So e.g. using LLC 4 (bigger vdroop) but adjust with a positive offset ?

[Jurrunio]

1 hour ago, katulen said:

So I will need to use a negative offset of 0.030-0.040v. Is that how you mean or? 

the fast way, yes. The exact negative offset value depends on your Realbench test stability though

 

1 hour ago, katulen said:

If so, would it change anything to do it the otherway around ? 

LLC to level 6, remove overclock and find out the number for offset mode, then apply the overclock and use the adaptive+offset mode, with the same offset used before. Raise voltage until stable.

[katulen]

1 hour ago, Jurrunio said:

the fast way, yes. The exact negative offset value depends on your Realbench test stability though

Problem is, if i do that i wont be stable in Realbench / Prime95 etc. Only for my games. I already know my min. voltage is 1.332v for being stable at 5.0GHz. Which is why i am currently using 1.350v with LLC 5, because that gets me to 1.332v-1.341v.

 

1 hour ago, Jurrunio said:

LLC to level 6, remove overclock and find out the number for offset mode, then apply the overclock and use the adaptive+offset mode, with the same offset used before. Raise voltage until stable.

As said above. Already clarified whats my min. voltage for being stable is. Can't I work out from that?

The issue is, if I put a negative offset of 0.018. It will probably work fine in games, since i get a slight vboost. However if I then do Prime95 i wont be stable there.

I hope you understand what i say here

[Mister Woof]

What's the problem here?

 

This is just LLC doing it's thing.

 

As long as your temps are fine, 1.36v isn't going to hurt you

[Falkentyne]

7 hours ago, katulen said:

Hey guys.

 

I've managed to get my overclock stable with adaptive voltages, however I have experienced some weird behavior which im trying to find an answer to.

I hope you can help me.

 

Here is a quick brief of my OC settings:

Adaptive voltages: 1.350v

Offset 0 (Auto)

AVX: Auto

LLC: Level 5

 

When running Realbench, my VCore is 1.350v (no vdroop).

When running Prime95 nonAVX my VCore is 1.332v (vdroop).

When playing games, e.g. CS:GO or PUBG, my VCore is 1.368v (vboost)

 

So is the reason behind this voltage behavior, caused by the type of load that is put on the CPU ?

E.g. stresstesting using Realbench, which uses AVX instructions, is causing my CPU to need more voltages, and therefor sits on 1.350v.

Where running Prime95 nonAVX causes my CPU to be fine with 1.332v, due to no AVX instructions.

And with games, where my CPU load is 40-50%, needs even less voltages to be stable, and therefor makes a slight vboost.

 

Or perhaps the answer is due to having Offset set to auto ?

 

I don't know if above is the cause of this, I do feel it could be a solution, but i am absolutely clueless on this tho.

 

Please correct me if I am wrong. Hoping to find an answer to this behavior.

Vboost doesn't exist

A negative loadline is NOT possible on any current hardware!

Transient spikes (which is not this 1.368v) are NOT vboost! Nor can they be measured in onboard sensors (requires an oscilloscope).

The reason you think you have vboost is because of the Super I/O sensor.

https://www.overclock.net/forum/27686004-post2664.html

[Jurrunio]

3 hours ago, katulen said:

Problem is, if i do that i wont be stable in Realbench / Prime95 etc

Then you're doing it wrong. LLC level 6 from what you described should give you voltage boost in realbench comparing to games. Just make sure your offset lets the CPU get around 1.34v in realbench will do, games will be fine with voltage below that. If that crashes, go a bit higher.

3 hours ago, katulen said:

I hope you understand what i say here

I would be grateful if you can ditch this wrong concept

8 hours ago, katulen said:

E.g. stresstesting using Realbench, which uses AVX instructions, is causing my CPU to need more voltages, and therefor sits on 1.350v.

Where running Prime95 nonAVX causes my CPU to be fine with 1.332v, due to no AVX instructions.

And with games, where my CPU load is 40-50%, needs even less voltages to be stable, and therefor makes a slight vboost.

 

for good

[katulen]

2 hours ago, Plutosaurus said:

As long as your temps are fine, 1.36v isn't going to hurt you

I know. As said, it is just itching my eye, and wanted to see if it was normal and if I could fix it

 

1 hour ago, Jurrunio said:

Then you're doing it wrong. LLC level 6 from what you described should give you voltage boost in realbench comparing to games. Just make sure your offset lets the CPU get around 1.34v in realbench will do, games will be fine with voltage below that. If that crashes, go a bit higher.

Okay, so did very little testing.

Made my adaptive voltage 1.315v and 0 offset - During RB vcore was 1.385v.

I made a negative offset of 0.053v, ran RB again, and now its at 1.341v, and in games sitting on 1.332v. So now its the opposite from what I had before, however better - I would assume. But if I at some point play a game which is more CPU heavy, and it crashes - I assume I just adjust the voltage or offset accordingly, correct?

 

Another thing I made notice of. Due to my negative offset 0.053v, it made my minimum idle vcore a bit lower. Lowest is 0.568. Before it was +0.600v. Is that a problem?

 

I only did like 15-30mins of RB testing, so obviously have to do more thoroughly testing, and will so when I get this tuned in.

[Jurrunio]

5 hours ago, katulen said:

I assume I just adjust the voltage or offset accordingly, correct? 

Just raise offset

 

Think of the ideal situation without Vdroop or overcompensated vdroop. The motherboard will give voltage according to the multiplier following Intel's recommended numbers, otherwise known as the VID table.  It goes up exponentially(roughly), from idle clock to max turbo clock. Go past that when overclocking, the motherboard will stay with the VID with the max turbo clock because it has no idea what to do outside of the VID table.

 

Adaptive mode affects what the board will do when overclocking past the max turbo clock. It will run to the voltage you set in the BIOS on your max frequency. Otherwise follows the VID table

 

Offset mode just adds or deducts a certain value from whatever the VID table says

 

Override mode just ignores the VID table

In other words, a flat line from idle to max frequency

 

With adaptive+offset mode, you could lower the offset to effectively undervolt the CPU at stock clocks, then use the adaptive part to make it stable with the overclock.

 

5 hours ago, katulen said:

that a problem? 

If it doesnt crash at idle it's fine

it can be hard to test tho

[katulen]

8 hours ago, Jurrunio said:

Adaptive mode affects what the board will do when overclocking past the max turbo clock. It will run to the voltage you set in the BIOS on your max frequency. Otherwise follows the VID table

 

Offset mode just adds or deducts a certain value from whatever the VID table says

I've got Manual Mode, Adaptive Mode(Optional: Offset) and Offset Mode. So when you refer to "Offset mode" you don't think about Adaptive Mode + an offset, but Offset Mode as its own option, yeah?

 

Is there any information on the specific VID tables available out there? Can't seem to find any. Found an Intel Article from years back, but I assume its different from chipset to chipset.

 

FYI, I made the current OC this morning, and everything worked fine. But tonight i noticed that I had a "WHEA Error" listed in my HWiNFO. Didn't have any BSODs, crashes or anything all day. But I assume when getting WHEA Errors there is something, somewhere, which isn't stable. So upped the voltage by 0.005

[Jurrunio]

1 hour ago, katulen said:

So when you refer to "Offset mode" you don't think about Adaptive Mode + an offset, but Offset Mode as its own option, yeah?

yup

 

1 hour ago, katulen said:

Is there any information on the specific VID tables available out there? Can't seem to find any. Found an Intel Article from years back, but I assume its different from chipset to chipset.

it's CPU specific. Check intel's spec sheet for your generation of CPUs. Dont think it's available anymore for new CPUs, but you could still see VID from software like HWinfo64.

[katulen]

18 hours ago, Jurrunio said:

it's CPU specific. Check intel's spec sheet for your generation of CPUs. Dont think it's available anymore for new CPUs, but you could still see VID from software like HWinfo64.

Alrite, thanks.

 

Btw. atm. i trying to finetune my OC by lowering the voltage as far as i can, using adaptive.

Ran 8hr Realbench last night, and everything seemed ok. However, today while playing PUBG i get a BSOD, with the error message:

IRQL_NOT_LESS_OR_EQUAL

 

I've read different opinions to this error. Some say its Driver related, some says its Overclock related.

So it makes me wonder, whether it's a driver issue, windows issue (reinstall windows) or a voltage issue?

I've also finetuned my VCCIO from 1.300v to 1.100v and had it like that for the past 3 days, but perhaps it needs more juice ?

 

Any thoughts on this?

[Jurrunio]

14 minutes ago, katulen said:

Btw. atm. i trying to finetune my OC by lowering the voltage as far as i can, using adaptive.

Ran 8hr Realbench last night, and everything seemed ok. However, today while playing PUBG i get a BSOD, with the error message:

IRQL_NOT_LESS_OR_EQUAL

voltage issue from my experience. Driver issue will lead to random error codes

 

14 minutes ago, katulen said:

Because i've also finetuned my VCCIO from 1.300v to 1.100v and had it like that for the past 3 days.

what is the memory frequency?

[katulen]

1 hour ago, Jurrunio said:

what is the memory frequency?

3000 MHz. DRAM Voltage is 1.300v (didnt touch this - default after enabling XMP)

I got told my vccio was high when it was on 1.3XXv, so tryid decreasing it.

 

I then got the same BSOD twice within 30 minutes.

First time, i'd upping vcore by 10mv. Then got the same BSOD, so upped VCCIO to 1.150v.

 

Trying to figure out what is causing this.

[Jurrunio]

Just now, katulen said:

3000 MHz. DRAM Voltage is 1.300v (didnt touch this - default after enabling XMP)

I got told my vccio was high when it was on 1.3XXv, so tryid decreasing it.

 

I then got the same BSOD twice within 30 minutes.

First time, i'd upping vcore by 10mv. Then got the same BSOD, so upped VCCIO to 1.150v.

 

Trying to figure out what is causing this.

1.3V VCCSA and VCCIO is stupid. On Intel that's what I'd use for testing 4000MHz. Wont go past 1.2V for daily use.

[katulen]

36 minutes ago, Jurrunio said:

1.3V VCCSA and VCCIO is stupid. On Intel that's what I'd use for testing 4000MHz. Wont go past 1.2V for daily use.

Yeah thats what I got told as well. The thing is, i did several 8hr realbench testing and all passed. With 1.315v (offset -0.053v) vccio 1.100v.

I am currently at 1.340v (offset -0.053v), VCCIO 1.150v.

 

So you'd assume its a voltage issue. But if it was, how come I can pass 2x Realbench 8hour stresstest with no issue, but playing PUBG makes it crash ? Both is also AVX.

 

Had "IRQL_NOT_LESS_OR_EQUAL" 3 times today, while playing PUBG. 1 time while i was ingame, and 2 times while i was ALT+TAB'd and 

 

EDIT:

So I did some research, and I have no clue if this could be a potential cause.

CAMAPP (NZXT app) asked me for 3 days to update, which I didnt. However, I did update yesterday.

When googling IRQL_NOT_LESS_OR_EQUAL camapp, there is several people back in the days, whom had issues with CAMAPP causing BSOD IRQL_NOT_LESS_OR_EQUAL. Thoughts on this?

[Jurrunio]

25 minutes ago, katulen said:

Yeah thats what I got told as well. The thing is, i did several 8hr realbench testing and all passed. With 1.315v (offset -0.053v) vccio 1.100v.

I am currently at 1.340v (offset -0.053v), VCCIO 1.150v.

 

So you'd assume its a voltage issue. But if it was, how come I can pass 2x Realbench 8hour stresstest with no issue, but playing PUBG makes it crash ? Both is also AVX.

 

Had "IRQL_NOT_LESS_OR_EQUAL" 3 times today, while playing PUBG. 1 time while i was ingame, and 2 times while i was ALT+TAB'd and 

 

EDIT:

So I did some research, and I have no clue if this could be a potential cause.

CAMAPP (NZXT app) asked me for 3 days to update, which I didnt. However, I did update yesterday.

When googling IRQL_NOT_LESS_OR_EQUAL camapp, there is several people back in the days, whom had issues with CAMAPP causing BSOD IRQL_NOT_LESS_OR_EQUAL. Thoughts on this?

last time I had driver issue I get many different errors, not just IRQL. Maybe it's different for yours? You can try updating software, not stopping you from that

[katulen]

2 minutes ago, Jurrunio said:

last time I had driver issue I get many different errors, not just IRQL. Maybe it's different for yours? You can try updating software, not stopping you from that

Well I would like too. But im pretty confident all my drivers are up to date. The one I didnt do was CAMAPP, but did that yesterday as written above. Therefor had some suspicion that could be the culprit. Also my NVIDIA driver is the 2nd latest. And reason for that, is that the newest NVIDIA driver is causing issues, and therefor i rolled back.

 

But again, if it was a voltage issue, how come I can pass 2x Realbench 8hour stresstest with no issue, but playing PUBG makes it crash ?

[Jurrunio]

1 minute ago, katulen said:

 

But again, if it was a voltage issue, how come I can pass 2x Realbench 8hour stresstest with no issue, but playing PUBG makes it crash ?

because they are different. I can pass P95 with AVX yet crash Assassin's Creed Origins.

[katulen]

1 minute ago, Jurrunio said:

because they are different. I can pass P95 with AVX yet crash Assassin's Creed Origins.

Hmm i see. I just thought the reason for doing stresstest, was to stress the CPU as hard as possible, and therefor it made a good reference to whether your OC was stable or unstable for anything else

 

So out of curiousity. What did you do with your OC to fix crashing in Assassins Creed?