14900KS - LLC Levels question

[0121danwilliams84]

Hi all,

I am currently running the 14900KS at stock with XMP of 7200MHz.

I have read a lot into LLC and the basics:

Lower LLC levels = higher idle voltage and higher vdroop.
Higher LLC levels = lower idle voltage and less vdroop.

I have been reading a few forums and watched a few YT videos where it is suggested to increase the stock LLC (level 3) to a higher value to prevent 'overshooting' voltage.

This would seem to me to be counter intuitive and increase the 'unseen by software' voltages and temps.

Would someone happen to know the theory behind such a setting and why it might be useful?

This video kind of sums up some of the LLC settings theory I have seen:

Intel i9 14900KS 6.3GHz Benchmarks & DDR5 8533MHz Asus Bios Settings

ENEBA - great place to buy games (PS5, Xbox, PC)! https://ene.ba/Bang4BuckAll games, DLC! https://ene.ba/Bang4Buck-StoreGift Cards! https://ene.ba/Bang4Buck-...

youtu.be

I should also mention that I have tried undervolting, but this chip seems to be at the limit as a fairly small undervolt is not stable.

Thanks!

 

[Proto]

You shouldn’t be using LLC unless you’re trying to run voltages higher than your board allows.

 

[0121danwilliams84]

You shouldn’t be using LLC unless you’re trying to run voltages higher than your board allows.

I was fairly sure that LLC is used in conjunction with raising vcore to stabilise overclocks as well. Is that not true? Pretty sure LLC is always used to some extent.

 

[Tetras]

I have been reading a few forums and watched a few YT videos where it is suggested to increase the stock LLC (level 3) to a higher value to prevent 'overshooting' voltage.

This would seem to me to be counter intuitive and increase the 'unseen by software' voltages and temps.

Would someone happen to know the theory behind such a setting and why it might be useful?

When the demand on the CPU increases, the VRM has to deliver more power, which causes a drop in voltage when the load spikes and an increase in voltage when the load decreases.

LLC compensates by pushing the voltage higher to maintain a more consistent profile and that reduces both the voltage drop under load and the increase when it returns to idle.

Lower LLC levels have lower idle voltage which is good (because they're generally compensating less in anticipation of the drop), but, they're also less effective at preventing the increase (the overshoot).

This is only what I understood from watching buildzoid's video though, so I can't confirm if I'm right:

Explaining Load-Line Calibration (LLC) & Not Killing Your CPU

Buildzoid joins to define load-line calibration in motherboard BIOS, then talks about how to ensure you don't kill your CPU with LLC.Ad: EVGA 1080 Ti SC2 w/ ...

www.youtube.com

 

[Proto]

You’re feeding your CPU more voltage that what’s being displayed.

The CPU is designed to run safely up till a given voltage, and the physical materials will naturally resist up to a certain point. All this is completely normal.

LLC will go above your set voltage, so potentially damaging your CPU. It will also try to reduce the affects of resistance, so placing even more unnecessary strain on your CPU.

LLC will make a certain displayed figure appear lower than it would be in some situations. However, this is only artificial, and the real impact isn’t clearly visible using monitoring tools.

It will also not stabilise your system, but have the opposite effect, putting more strain on your system than necessary.

Fortunately, some side effects that fall outside the scope of LLC can still be seen using monitoring software. A good example is higher temperature when using LLC, so you know there’s something wrong.

Understanding this requires a lot of experience, and this makes it hard to explain to people.

 

[0121danwilliams84]

When the demand on the CPU increases, the VRM has to deliver more power, which causes a drop in voltage when the load spikes and an increase in voltage when the load decreases.

LLC compensates by pushing the voltage higher to maintain a more consistent profile and that reduces both the voltage drop under load and the increase when it returns to idle.

Lower LLC levels have lower idle voltage which is good (because they're generally compensating less in anticipation of the drop), but, they're also less effective at preventing the increase (the overshoot).

This is only what I understood from watching buildzoid's video though, so I can't confirm if I'm right:

Explaining Load-Line Calibration (LLC) & Not Killing Your CPU

Buildzoid joins to define load-line calibration in motherboard BIOS, then talks about how to ensure you don't kill your CPU with LLC.Ad: EVGA 1080 Ti SC2 w/ ...

www.youtube.com

Thanks for the input.

It is hard to understand. I thought lower LLC levels had higher idle voltage? And I thought that was based on some of the theory to increase LLC to get a lower idle voltage.

 

[Tetras]

Thanks for the input.

It is hard to understand. I lower LLC levels had higher idle voltage? I thought that was based on some of the theory to increase LLC to get a lower idle voltage.

I think it depends on your motherboard.

From what Buildzoid said, different boards have different labels/terminology and what I'm describing is:
- If you have a low compensation then the general voltage given to the CPU will be lower, but the drops/increases will be higher.
- If you have a high compensation, then the general voltage given to the CPU will be higher, but the drops/increases will be lower.

I don't know if you're referring to an LLC level that is actually less than stock and if so, that was outside of the scope of Buildzoid's video, so by low and high I'm only talking about applying more voltage than stock.

 

[0121danwilliams84]

I think it depends on your motherboard.

From what Buildzoid said, different boards have different labels/terminology and what I'm describing is:
- If you have a low compensation then the general voltage given to the CPU will be lower, but the drops/increases will be higher.
- If you have a high compensation, then the general voltage given to the CPU will be higher, but the drops/increases will be lower.

I don't know if you're referring to an LLC level that is actually less than stock and if so, that was outside of the scope of Buildzoid's video, so by low and high I'm only talking about applying more voltage than stock.

Just regarding regular llc levels (1-8 for ASUS in the bios)

I guess I was just wondering if there was a benefit to increasing the llc level (even if stable anyway) just like the guy in the video does I linked in the first comment. He’s barely overclocking the cpu but seems to think it will give better voltage regulation.

 

[Tetras]

Just regarding regular llc levels (1-8 for ASUS in the bios)

I guess I was just wondering if there was a benefit to increasing the llc level (even if stable anyway) just like the guy in the video does I linked in the first comment. He’s barely overclocking the cpu but seems to think it will give better voltage regulation.

Yes, I believe that's the case, but if the voltage is generally raised higher to compensate, it'll also increase power consumption and temperatures under load, so I don't know if that would be worthwhile if the CPU is generally stable.

It would maybe depend on the quality of the PSU, the VRM, how many phases it has, etc. I'd guess that a lower quality VRM would benefit more from doing that than a higher quality one.

 

[0121danwilliams84]

Yes, I believe that's the case, but if the voltage is generally raised higher to compensate, it'll also increase power consumption and temperatures under load, so I don't know if that would be worthwhile if the CPU is generally stable.

It would maybe depend on the quality of the PSU, the VRM, how many phases it has, etc. I'd guess that a lower quality VRM would benefit more from doing that than a higher quality one.

He does have a pretty insane Mora watercooler so perhaps it makes sense to run tighter voltages if his temps are of no concern. I still think that wouldn’t be worth it if voltages only seen with an oscilloscope would be possibly way higher. He seems to know what he’s doing but I don’t get it.

 

[Tetras]

He does have a pretty insane Mora watercooler so perhaps it makes sense to run tighter voltages if his temps are of no concern. I still think that wouldn’t be worth it if voltages only seen with an oscilloscope would be possibly way higher. He seems to know what he’s doing but I don’t get it.

Yeah, I'd assume that if the cooling is good then the downside would be minimal, so long as you don't go too crazy.

 

[Proto]

You can push a wall gently all day with little effect, but a very quick hit with a large hammer will instantly shatter it.

This can also show up as cracking that eventually leads to shattering.

 

[0121danwilliams84]

Vcore regulation on the ASUS Maximus Z790 Apex.

My Patreon: https://www.patreon.com/buildzoidTeespring: https://teespring.com/stores/actually-hardcore-overclockingBandcamp: https://machineforscreams.bandca...

www.youtube.com

Buildzoid seems to recommend higher LLC levels in this video to reduce idle voltages and this is the motherboard I am using.

 

[Proto]

The higher idle voltage is to be expected.

Maybe use a tiny little bit of LLC?

 

[Tetras]

Vcore regulation on the ASUS Maximus Z790 Apex.

My Patreon: https://www.patreon.com/buildzoidTeespring: https://teespring.com/stores/actually-hardcore-overclockingBandcamp: https://machineforscreams.bandca...

www.youtube.com

Buildzoid seems to recommend higher LLC levels in this video to reduce idle voltages and this is the motherboard I am using.

My understanding of what he is saying in the video, is that when you're using a static voltage, the lower levels of LLC lead to the board having higher drops, which means if you want to maintain a higher average voltage at load (which is important to OC stability) then you have to push up the static voltage.

If you use a higher level of LLC, then you can use a lower static voltage (which gives you lower idle voltage), because the level of drop is lower at load, which means the overall stability is better even though your idle voltage can be lower than at lower levels of LLC.

He clarifies that even though higher levels of LLC potentially push more voltage overall (at high load) to maintain a stable average, this is not important for short benchmark runs, because the VRM can handle this fine (especially when you're not scooting around the desktop and causing frequent load spikes), but you probably wouldn't run a high level of LLC if you wanted to spend 5 hours rendering.

 

[0121danwilliams84]

My understanding of what he is saying in the video, is that when you're using a static voltage, the lower levels of LLC lead to the board having higher drops, which means if you want to maintain a higher average voltage at load (which is important to OC stability) then you have to push up the static voltage.

If you use a higher level of LLC, then you can use a lower static voltage (which gives you lower idle voltage), because the level of drop is lower at load, which means the overall stability is better even though your idle voltage can be lower than at lower levels of LLC.

He clarifies that even though higher levels of LLC potentially push more voltage overall (at high load) to maintain a stable average, this is not important for short benchmark runs, because the VRM can handle this fine (especially when you're not scooting around the desktop and causing frequent load spikes), but you probably wouldn't run a high level of LLC if you wanted to spend 5 hours rendering.

That makes sense thanks for summing that up. I didn’t think about that. So does the test he did just apply to a static voltage or would a similar thing happen with auto/adaptive voltage I wonder?

 

[Tetras]

That makes sense thanks for summing that up. I didn’t think about that. So does the test he did just apply to a static voltage or would a similar thing happen with auto/adaptive voltage I wonder?

I think a higher LLC would allow you to use an offset to lower the voltage, because of the higher stability at high load, but I'm not sure it would make sense to do that, because it would also raise the average voltage under load higher than stock.

I suppose if your board was highly effective at a particular LLC level or your CPU is a lemon that is unstable at low voltages, then maybe it would make sense, but it would need a lot of testing for the impact on stability.

 

[momo56]

You would need to also reduce and mess about with the AC_LL to reduce voltage/ power.

The z790 guide on oc.net has this in great detail with diagrams and formulas if you’re really interested.

What’s the SP score on your CPU ?

 

[0121danwilliams84]

You would need to also reduce and mess about with the AC_LL to reduce voltage/ power.

The z790 guide on oc.net has this in great detail with diagrams and formulas if you’re really interested.

What’s the SP score on your CPU ?

Not amazingly confident and don't particularly understand the AC_LL although I have read a lot about it.

I have read the guide on oc.net.

I have an SP score of 102 (not great from what I have seen) for the 14900KS

I have tried undervolting with the global core svid voltage (ASUS mobo) but it gives WHEA errors at even just -0.01000v

I read about undervolting with the V/F point curve and looked at the Skatterbencher guide.

From what it look like, I have to undervolt the points 6,7,8 and 9 that control the higher boost frequencies. Maybe that would be more likely to be stable?

Strange I can't undervolt using the global setting as I always had good luck with that in the past.

 

[momo56]

Not amazingly confident and don't particularly understand the AC_LL although I have read a lot about it.

I have read the guide on oc.net.

I have an SP score of 102 (not great from what I have seen) for the 14900KS

I have tried undervolting with the global core svid voltage (ASUS mobo) but it gives WHEA errors at even just -0.01000v

I read about undervolting with the V/F point curve and looked at the Skatterbencher guide.

From what it look like, I have to undervolt the points 6,7,8 and 9 that control the higher boost frequencies. Maybe that would be more likely to be stable?

Strange I can't undervolt using the global setting as I always had good luck with that in the past.

Have you tried AI OC ? It’s surprisingly good. Less of a hassle too.

LLC levels is just default DC_LL on asus so you will need to adjust AC_LL if volts or power is a concern.

Also if -0.01 isn’t stable then there’s a good chance stock is not stable either and you might have a duff cpu.