BuildZoid analyzes 0x123 versus 0x125 microcode using his Oscilloscope

[FlyingScot]

While there is nothing revolutionary in these notes, there might be some value in summarizing what we already know, assume, or suspect. It might also be worth stating that the primary focus of these notes is the issue of High Voltages. In other words, Temperatures and Power Usage are not the specific focus, although some actions that can be taken by the average user can help all three metrics.

All testing was done on a 14900K with 256W as the “power limits” and CEP enabled. The motherboard in question was a Gigabyte this time.
My commentary = [ ]

Based upon what he saw on the oscilloscope, he recommends the 0x125 microcode over the 0x123 even though it’s far from perfect. On his 14900K, average voltage dropped by 50mV (due to lower low workload voltages with the new microcode). Single core boost voltages are also lower with 0x125 than 0x123, which is a good thing.

Typical voltages were around 1.4V, but he couldn’t find an explanation for momentary spikes as high as 1.59V. And this was with the 0x125 microcode. EDIT: These spikes last as little as 50ms and therefore are unlikely to be captured by HWInfo.

[I think this issue is something Intel will try to mitigate in the August 2024 microcode/BIOS update.]​

​

AC_LL is higher with 0x125. But for some unknown reason, the average voltages dropped on his 14900K.

[Incidentally, Gigabyte’s default Auto LLC has always been the droopiest according to one user, so BuildZoid’s testing mirrors what happens when choosing the new MSI CPU Lite Load Mode=Intel Defaults option, i.e. Auto LLC drops from the equivalent of mode 3? to equivalent of mode 8.]​

On his 14900K, E-core VID voltage requests can sometimes exceed P-Core VID voltage requests (which affects the voltage for the entire CPU). This condition was observed when certain workloads allowed the P-Cores to remain at 5.0Ghz or lower.

There is a constant voltage fight going on between the E-cores and the P-Cores for who controls the voltage. This can lead to erratic behavior if you are near the power limits. “To mitigate this condition, users could manually lockdown the E-cores to 4.0Ghz, rather than letting them try to boost to 4.4Ghz, etc.”

[Perhaps Raptor Lake would benefit from a separate voltage rail for the E-cores, and maybe included the Ring on this secondary rail, too.]​

It’s not recommended to disable E-cores because you could end up with much higher voltage going to the P-Cores within your chosen power limits.

BuildZoid theorized that gaming workloads could be driving up P-Core frequencies (and therefore higher VID voltage requests) while staying under the PL1/PL2 power limits because they use far less of the E-Cores. In other words, voltages could remain elevated and still not trigger the power limiters. BuildZoid theorizes that this condition could be the reason why game developers are seeing unusually instability (and possible degradation) with gaming servers and gaming client CPUs.

IgorsLab sampling of CPU voltages shows that the best i7’s can run with 1.2xV while the worst need 1.4xV, which overlaps the i9’s that start at around 1.4V and go higher. Some i5’s overlap the i7’s, with the worst needing 1.4V. BuildZoid believes that some of these high voltage chips will continue to be a concern because he doubts that Intel will be able to bring their voltages down enough, ideally by 100mV.

[The implication is that CPUs that can already run with less than 1.4V might be fine if Intel can addresses the unusual voltage spikes.]​

However, users who are willing to sacrifice 100Mhz, or so, could lower voltages by up to 50mV, which might be enough to help safeguard the poorer silicon chips from degradation.

BuildZoid quote: "I'm not a fan of dumping 1.4V into the CPU at 80C."

[Coming from BuildZoid, the king of high voltage overclocking, that's saying something.]​

BuildZoid: Cinebench R15, R20 and R23 should all be used to test for stability (and not just R23) when making changes to voltage/frequency due to the different types of workload they produce. Of course, these are not stress test applications, but they are useful for catching instability if you have lowered the voltage too aggressively.

My interpretation of BuildZoid's conclusions:
Take whatever approach you want to (or can) in order to keep Vcore below 1.4V under all scenarios, and ideally 1.35V or lower. This effort could include one or more of the following approaches:

1. Install 0x125 and/or the new August microcode
2. Manually undervolt via CPU Lite Load
3. Manually undervolt via Adaptive Offsets
4. Manually lowering E-Core Ratios
5. Manually lowering P-Core Ratios
6. Eliminate Auto-boosting of Single/Twin Cores (EDIT: unless you absolutely have to have the best single thread performance possible, I think the risk of accelerated degradation is too high. BuildZoid has a separate video related to this issue, which I have now linked below.)

Note: Your ability to do the above will depend on the quality of your silicon. As stated earlier, i7's and even i5's have significant variation in VID tables. However, CiTay posted a voltage chart somewhere on this forum that illustrates how much additional voltage you typically need for each 100Mhz step in core frequency. Basically, as you get closer and closer to the maximum potential of the silicon, the increase becomes exponentially greater. The upside of this behavior, especially given how far Intel has pushed the Raptor Lake CPUs, is that a relatively small reduction in max frequency could have a very significant affect on lowering max voltages.

Optional:
Lower PL1/PL2 (Very useful for all-core workloads. See CiTay’s new guide found here.)
Lower TjMax temp from the default 100C.
Enable CEP (All but ultra-novices will likely find this option too intrusive to be beneficial.)

 

[crimson8814f802d4]

Thanks for the write up.
14700KF:
I use LiteLoad mode 7 (alongside power limits 175w on both PLs ans core current at 307a) and found that CEP cut my Cinebench score in half. As such, I’ve disabled CEP - CiTay suggested I do this.

Intel’s new bios profile (latest bios) originally put PLs at 253w, CC at 307A and LiteLoad at mode 15!

My core current never goes beyond 1.38v with my adjusted settings.

How can I lower LiteLoad without CEP jumping in?
How can I shave off more voltage?

Thanks!!

 

[FlyingScot]

I don't think even Buildzoid really understands what upsets CEP so much. I would like to think that we enable this feature being that Intel does recommend it. But it's an odd animal indeed. We've all seen what it can do to scores, but not everyone has the same problem. Given my very limited understanding of CEP, I would have to say that you might have to increase Vcore (raise Lite Load) to restore your scores. But you might not want that. I guess one could argue that if you are stable without it then you stick with that setup. And maybe if you run into future instability, you turn it on and be prepared to increase the voltage until it stops kicking you in the butt. So, to your question: How can I lower Lite Load without CEP jumping in? The answer might be that your VCore is already too low according to CEP.

You know, we can't even rule out that Intel might tweak CEP to become less intrusive with future microcode updates. I'm going to keep and open mind as to the usefulness of CEP. I'm not prepared to throw in the towel just yet. But I have absolutely no advice I can offer to others. I'll just have to test it thoroughly when I get the opportunity.

 

[FlyingScot]

How can I shave off more voltage?

Any one of my above suggestions could do it. If you stress test and find that you cannot lower it with your given core frequencies then you would have to sacrifice one or more of those frequencies. If we think in simple terms of the Voltage/Frequency VID table in each CPU, there really are only two basic inputs, voltage and frequency.

 

[crimson8814f802d4]

Thanks! As I mentioned, I don’t go above 1.38 Vcore with my current settings and even in OCCT 1 hour CPU stress test my CPU doesn’t hit 80 degrees.
I limit power to 175w and have disabled CEP.

No stability issues whatsoever, but I’d rather not kill perf with whatever CEP is doing.

Worth mentioning that I also disable Enhanced Turbo.

I wonder if CEP is essential for keeping the CPU safe though

 

[FlyingScot]

Worth mentioning that I also disable Enhanced Turbo

Enhanced Turbo is that MCE thing that ASUS first invented. That was a great "gift" for chips like the 3rd gen, but not so much for today's power hungry monsters.

You seem happy with your system. But, I hear you. 1.38v is a little on the higher side - if that's after Vdroop. What's your steady Vcore once Cin R23 gets rolling? I wonder if there is any room to lower Load Line Calibration to increase your Vdroop?

 

[crimson8814f802d4]

Enhanced Turbo is that MCE thing that ASUS first invented. That was a great "gift" for chips like the 3rd gen, but not so much for today's power hungry monsters.

You seem happy with your system. But, I hear you. 1.38v is a little on the higher side - if that's after Vdroop. What's your steady Vcore once Cin R23 gets rolling? I wonder if there is any room to lower Load Line Calibration to increase your Vdroop?

Cinebench goes to about 1.35v. The highest ever reading is 1.38 that I’ve observed. Idle tends to sit between 0.6 and 0.7v.

I wonder if I lock my two boost cores to x55 (rather than x56) if that would help?

Could I lower the LiteLoad to go for lower voltage?

 

[FlyingScot]

Cinebench goes to about 1.35v. The highest ever reading is 1.38 that I’ve observed. Idle tends to sit between 0.6 and 0.7v.

I wonder if I lock my two boost cores to x55 (rather than x56) if that would help?

Could I lower the LiteLoad to go for lower voltage?

Yes, I think you would do well to lock all cores to the same ratio (at the very least until Intel can reliably solve these mysterious 50ns voltage spikes, etc.). You should not have to touch the Lite Load unless you want to as a phase II. The CPU should automatically use less peak voltage because the VID request is linked directly with the frequency (but modified by other factors such as temperature). Speaking of temperature, this is what I have in my notes for the BIOS TVB Voltage Optimizations setting:

"Do not disable TVB Voltage Optimizations as this can lead to higher than necessary Vcore at lower temperatures. When enabled, the processor takes into account the temperature when requesting the core voltage based on how far the cores are from TJMAX (default = 100c). The further away (i.e the lower the temperature) the larger the negative offset. In other words, the lower your temperature, the lower the applied voltage. Without this optimization, the voltage requested is what would be needed to keep the CPU stable up to and including 100C. However, I read somewhere that Intel TVB is automatically turned off if you set fixed voltage."

So, given the above statement, you can also reduce peak voltages by improving your cooling (and of course in full-load scenarios by lowering PL1/PL2). EDIT: Although, I think I read somewhere that you can only expect a 2mV credit for every degree C you shave off. So you'll have to work hard for it!

 

[crimson8814f802d4]

Yes, I think you would do well to lock all cores to the same ratio (at the very least until Intel can reliably solve these mysterious 50ns voltage spikes, etc.). You should not have to touch the Lite Load unless you want to as a phase II. The CPU should automatically use less peak voltage because the VID request is linked directly with the frequency (but modified by other factors such as temperature). Speaking of temperature, this is what I have in my notes for the BIOS TVB Voltage Optimizations setting:

"Do not disable TVB Voltage Optimizations as this can lead to higher than necessary Vcore at lower temperatures. When enabled, the processor takes into account the temperature when requesting the core voltage based on how far the cores are from TJMAX (default = 100c). The further away (i.e the lower the temperature) the larger the negative offset. In other words, the lower your temperature, the lower the applied voltage. Without this optimization, the voltage requested is what would be needed to keep the CPU stable up to and including 100C. However, I read somewhere that Intel TVB is automatically turned off if you set fixed voltage."

So, given the above statement, you can also reduce peak voltages by improving your cooling (and of course in full-load scenarios by lowering PL1/PL2).

When I got to set the per-p core ratio limit from 56 (on core 4 and 5 - boost cores) to 55 MSI Bios tries to switch to “MSI unlimited settings”. Pretty strange. I have to manually put it back to Intel defaults.

I’ve enabled TVB Voltage Optimisations. It was set to Auto before.

 

[FlyingScot]

When I got to set the per-p core ratio limit from 56 (on core 4 and 5 - boost cores) MSI Bios tries to switch to “MSI unlimited settings”. Pretty strange.

I’ve enabled TVB Voltage Optimisations. It was set to Auto before.

TVB = Auto should = enabled. But it doesn't hurt to set it manually.

In regards to setting core ratios manually, you'll definitely need help from someone with a Z790 BIOS because there can be generational changes in behavior. I don't have one (yet) so I can't be of any help. Perhaps you can help by being a pioneer on this one!

 

[crimson8814f802d4]

TVB = Auto should = enabled. But it doesn't hurt to set it manually.

In regards to setting core ratios manually, you'll definitely need help from someone with a Z790 BIOS because there can be generational changes in behavior. I don't have one (yet) so I can't be of any help. Perhaps you can help by being a pioneer on this one!

I’ve set them manually. It’s just each time I input a value other than the default (56 in this case) it tried to impose the “water cooler” profile (as in unlocked power levels 4000w+!!)

I think it’s a bug though - probably assumed I was going to over clock the cores individually.

The fix appears to be, set the lower value and then manually re-enable the intel default profile. All of my core current / PL values haven’t changed, so we’re all good.

Oddly, with the aforementioned changed settings I’m now getting 88 degrees tops in cinebench r23. No changes to points score though.

I’m going to revert as I don’t like those temps.

 

[FlyingScot]

Oddly, with the aforementioned changed settings I’m not getting 88 degrees tops in cinebench r23.

Something else must have changed. Sounds buggy.

 

[FlyingScot]

Deleted post! Keeping it businesslike...

 

[crimson8814f802d4]

Went back to my original settings (auto clock settings, PL1 and PL2 at 175w, core current 307a, enhanced turbo off, LiteLoad mode 7) and I’m back to 1.346v / 79 degrees in Cinebench r23.

I may actually reduce the pl to 170 as HWINFO64 reports 189w on CPU package power (above my imposed 175w limit).

System is stable as it is, so I’m going to wait for the August bios before changing anything else.

At the end of the day, my system is currently stable; if it suddenly develops problems I’ll go back to Cyberpower PC for intel’s 3 year warranty.

 

[citay]

I am currently testing with a 12700KF on a PRO Z790-A WIFI with 2x 16 GB DDR5-6000, something i'm building for someone else (he really wanted an i7, and i said, ok, let's get a 12th gen one then, so he doesn't have to worry about CPU degradation).

Could I lower the LiteLoad to go for lower voltage?

You can go as low as possible without losing stability. There are two approaches. Normally i always recommend the first one, because it's less complex, and it does the job of lowering the power draw with one easy setting, if you test that it's fully stable. Usually after having set the power limits to the individual cooling situation.

So, first method: Just lower the mode for CPU Lite Load. Your only limit as far as lowering it should be the point where the CPU becomes unstable because the voltage becomes too low. Once you see instability in any tool (OCCT, y-cruncher, Prime95, whatever) or let alone in Cinebench (then it's quite unstable), you back off and test the next higher mode (so for example Mode 3 -> 4). Also, if the board shows an "F1/F2 Overclock Failed" screen during POST, or otherwise has trouble booting, the mode was too low. Once you find the lowest stable mode (say if Mode 4 is stable), i usually recommend to set one mode higher though (Mode 5 in this example), to have stability headroom. Done.

Note: If you see "IA CEP" intervening massively once you lower CPU Lite Load by more than 3 steps or so, meaning, when CPU performance like the Cinebench score massively goes down, then IA CEP will have to be disabled if you want to lower the voltages further. There doesn't seem to be any other way around it. But, since we are trying to lower the power draw, and since exposure to high voltages seems to be one of the key factors in this CPU degradation issue, i don't see why we should keep CPU Lite Load in a range where IA CEP stays happy. Yes, Intel recommended IA CEP to stay Enabled, but they were just listing every safety measure they could, while still investigating the cause for the degradation. Now they know, "exposure to elevated voltages" is the main culprit, so if we have a way to lower voltages further and keep the full performance (or sometimes even improve it, because it has more of the power budget to play with before the power limits kick in), then we should use it.

For a 14th gen i7 (by the far the most extreme i7 ever, way worse than the 13700K in regards of the power draw), and of course also the i9 models, you may not be able to lower the voltage by huge amounts. Because for their high frequencies, they need a high voltage, simple as that. Of course there are still better and worse CPU samples for every model, but on average, i found that the higher-up models can sometimes be a bit more resistant to shaving off voltage. There is no need to be hesitant about trying low modes though, just test for stability and you will soon see what it can do. Probably the only thing i would suggest is sticking to Mode 2 as the lowest to try, because Mode 1 only works with really good CPU samples.


Now, as some people have pointed out before, this method of lowering CPU Lite Load "Normal" mode is not perfect. The main two criticisms are:

1) It picks its own value combination for AC Loadline and DC Loadline. AC Loadline is what actually influences the voltage, it's a voltage added by the BIOS to make up for electrical properties of the CPU socket and so on. The background is not so important to understand, the main thing is, the higher this value is, the more voltage is added. This not only takes into account the electrical properties on the board, it also can make really bad CPU samples (when you lost the silicon lottery) run stable when set appropriately by default, or it can make CPUs be unstable from factory, if set too low by default. Lastly, if set too high by default, it will make the CPUs draw too much power and run too hot. The board makers have been going through various possibilities; Gigabyte at one time set the AC/DC Loadline way too high (equivalent to a high "CPU Lite Load" default mode on MSI), and even on MSI i recently saw one BIOS set Mode 22 which is among the highest possible, quite crazy. So they might have been trying to solve instability by adding more voltage, even though Intel now found that excessive voltage is what is causing problems.

For DC Loadline, "CPU Lite Load" on Normal (just changing the mode) will pick some value which might not result in the correct power draw readings anymore. Apart from that detail, the wrong value is not of much consequence. But if the cooling is such that power limits had to be set to prevent thermal throttling (which for a 14th gen i7 would almost universally be the case), then incorrect power draw readings will mess with the power limit throttling, it might set in too early or too late. That's where CPU Lite Load "Advanced" comes in.

In CPU Lite Load Advanced, you can select values for AC and DC Loadline seperately, without having some preset combination which can have the wrong DC Loadline. So now you can set the DC Loadline so it results in the correct power draw numbers. Doing that involves using HWinfo Sensors, creating full CPU load, then looking at the CPU's VID requests (in the "current values" column), which is the voltage the CPU asks for from the board, and comparing it to the current VCore value (note that i don't mean current as in Amperes, i mean current as in, actual live values, not Min/Max/Avg). If those are near-identical, the correct DC Loadline value has been found. The correct DC Loadline value depends on the LLC mode, another setting which influences the voltage (and not to be confused with CPU Lite Load / CLL, completely different). A table is here, but it's only a rough one, because each board model is built differently and would also need a bit different values, presumably.

2) It has been said that CPU Lite Load sometimes cannot be lowered as much when you don't also tweak the LLC mode. That's because, when keeping LLC mode on Auto, it results in a big VDroop (VCore reduction under high CPU load, to prevent an overshoot when the high CPU load suddenly stops and the voltage regulator has trouble reacting fast enough, causing a voltage overshoot if VDroop was not applied). So with CPU Lite Load, when lowering the mode there, the VDroop also has to be taken into account, so let's say we have to stay at CPU Lite Load "Normal" Mode 5, because otherwise, together with the big VDroop, the voltage under load wouldn't be enough for stability. But when we then look at low- to mid-load scenarios, VCore might be higher than necessary for stability, without the VDroop it might've been stable on Mode 2 or so. So what is then suggested, also set the LLC to something more aggressive (starting from LLC Mode 8 and going up towards Mode 1, it gets more and more aggressive in preventing VDroop, but Mode 1 is far too aggressive, usually you'd never go beyond Mode 4 or so).

What happens with tweaked LLC mode instead of Auto? Now the VDroop is less, so under full load, the voltage drops less. And now in turn, CPU Lite Load (either Normal or Advanced) can be lowered further than with LLC on Auto, because since the VDroop is less (prevented more), a lower CPU Lite Load settings may result in the same voltage than before (with a higher CPU Lite Load mode and LLC on Auto). With the added benefit that, supposedly, low- to mid-load scenarios of CPU workload now also have lower voltages than before.

However, this last step doesn't quite pan out for me so far in my full CPU load testing with this 12700KF on the PRO Z790-A WIFI. Because even if i set AC Loadline to 10, or even 1, the lowest possible value, and select a mild LLC setting like Mode 7, where the VDroop is still relatively big, i can't get anywhere near the low voltages and power draw i can reach just with a lowered CPU Lite Load and keeping LLC on Auto. I'm talking 20, 30 Watts difference here under full load (measured at the wall). In other words, lowered CPU Lite Load with LLC on Auto results in the lowest possible VCore and power draw under full load, which is arguable one of the most important scenarios to keep the power draw low for (to avoid cooling problems and in general).

Sorry if this is a bit messy and long, i wrote this while i'm pacing back and forth between rooms to tweak things and check the results on this guy's future PC...

I may actually reduce the pl to 170 as HWINFO64 reports 189w on CPU package power (above my imposed 175w limit).

Just a usual short-term overshoot, if you reset the values (clock symbol) and then look at the average, it should stay pretty much on 175W. Don't go by the reported power values, as mentioned, they can also be a bit wrong, it's better to just go by CPU temperatures, which will inform if you have to lower the power limits further or if you're good.

 

[FlyingScot]

Note: If you see "IA CEP" intervening massively once you lower CPU Lite Load by more than 3 steps or so, meaning, when CPU performance like the Cinebench score massively goes down, then IA CEP will have to be disabled if you want to lower the voltages further. There doesn't seem to be any other way around it. But, since we are trying to lower the power draw, and since exposure to high voltages seems to be one of the key factors in this CPU degradation issue, i don't see why we should keep CPU Lite Load in a range where IA CEP stays happy. Yes, Intel recommended IA CEP to stay Enabled, but they were just listing every safety measure they could, while still investigating the cause for the degradation. Now they know, "exposure to elevated voltages" is the main culprit, so if we have a way to lower voltages further and keep the full performance (or sometimes even improve it!), then we should use it.

That's a good counterpoint in the on-going CEP debate, at least the one in my feeble brain. I wonder, though, how much lower in reality you can go below the CEP "happy stage" and still remain stable under all conditions. That's a bit of an unknown. You can think your PC is stable and then up pops a WHEA error out of nowhere after a month of use.

 

[FlyingScot]

I am currently testing with a 12700KF

Terrific news! You can help me dial mine in, if and when I pull it out of the box. You would help old Flyingscot, won't you?

 

[crimson8814f802d4]

I am currently testing with a 12700KF on a PRO Z790-A WIFI with 2x 16 GB DDR5-6000, something i'm building for someone else (he really wanted an i7, and i said, ok, let's get a 12th gen one then, so he doesn't have to worry about CPU degradation).



You can go as low as possible without losing stability. There are two approaches. Normally i always recommend the first one, because it's less complex, and it does the job of lowering the power draw with one easy setting, if you test that it's fully stable. Usually after having set the power limits to the individual cooling situation.

So, first method: Just lower the mode for CPU Lite Load. Your only limit as far as lowering it should be the point where the CPU becomes unstable because the voltage becomes too low. Once you see instability in any tool (OCCT, y-cruncher, Prime95, whatever) or let alone in Cinebench (then it's quite unstable), you back off and test the next higher mode (so for example Mode 3 -> 4). Also, if the board shows an "F1/F2 Overclock Failed" screen during POST, or otherwise has trouble booting, the mode was too low. Once you find the lowest stable mode (say if Mode 4 is stable), i usually recommend to set one mode higher though (Mode 5 in this example), to have stability headroom. Done.

Note: If you see "IA CEP" intervening massively once you lower CPU Lite Load by more than 3 steps or so, meaning, when CPU performance like the Cinebench score massively goes down, then IA CEP will have to be disabled if you want to lower the voltages further. There doesn't seem to be any other way around it. But, since we are trying to lower the power draw, and since exposure to high voltages seems to be one of the key factors in this CPU degradation issue, i don't see why we should keep CPU Lite Load in a range where IA CEP stays happy. Yes, Intel recommended IA CEP to stay Enabled, but they were just listing every safety measure they could, while still investigating the cause for the degradation. Now they know, "exposure to elevated voltages" is the main culprit, so if we have a way to lower voltages further and keep the full performance (or sometimes even improve it, because it has more of the power budget to play with before the power limits kick in), then we should use it.

For a 14th gen i7 (by the far the most extreme i7 ever, way worse than the 13700K in regards of the power draw), and of course also the i9 models, you may not be able to lower the voltage by huge amounts. Because for their high frequencies, they need a high voltage, simple as that. Of course there are still better and worse CPU samples for every model, but on average, i found that the higher-up models can sometimes be a bit more resistant to shaving off voltage. There is no need to be hesitant about trying low modes though, just test for stability and you will soon see what it can do. Probably the only thing i would suggest is sticking to Mode 2 as the lowest to try, because Mode 1 only works with really good CPU samples.


Now, as some people have pointed out before, this method of lowering CPU Lite Load "Normal" mode is not perfect. The main two criticisms are:

1) It picks its own value combination for AC Loadline and DC Loadline. AC Loadline is what actually influences the voltage, it's a voltage added by the BIOS to make up for electrical properties of the CPU socket and so on. The background is not so important to understand, the main thing is, the higher this value is, the more voltage is added. This not only takes into account the electrical properties on the board, it also can make really bad CPU samples (when you lost the silicon lottery) run stable when set appropriately by default, or it can make CPUs be unstable from factory, if set too low by default. Lastly, if set too high by default, it will make the CPUs draw too much power and run too hot. The board makers have been going through various possibilities; Gigabyte at one time set the AC/DC Loadline way too high (equivalent to a high "CPU Lite Load" default mode on MSI), and even on MSI i recently saw one BIOS set Mode 22 which is among the highest possible, quite crazy. So they might have been trying to solve instability by adding more voltage, even though Intel now found that excessive voltage is what is causing problems.

For DC Loadline, "CPU Lite Load" on Normal (just changing the mode) will pick some value which might not result in the correct power draw readings anymore. Except that detail, the wrong value is not of much consequence. But if the cooling is such that power limits had to be set to prevent thermal throttling (which for a 14th gen i7 would almost universally be the case), then incorrect power draw readings will mess with the power limit throttling, it might set in too early or too late. That's where CPU Lite Load "Advanced" comes in.

In CPU Lite Load Advanced, you can select values for AC and DC Loadline seperately, without having some preset combination which can have the wrong DC Loadline. So now you can set the DC Loadline so it results in the correct power draw numbers. Doing that involves using HWinfo Sensors, creating full CPU load, then looking at the CPU's VID requests (in the "current values" column), which is the voltage the CPU asks for from the board, and comparing it to the current VCore value (note that i don't mean current as in Amperes, i mean current as in, actual live values, not Min/Max/Avg). If those are near-identical, the correct DC Loadline value has been found. The correct DC Loadline value depends on the LLC mode, another setting which influences the voltage (and not to be confused with CPU Lite Load / CLL, completely different). A table is here, but it's only a rough one, because each board model is built differently and would also need a bit different values, presumably.

2) It has been said that CPU Lite Load sometimes cannot be lowered as much when you don't also tweak the LLC mode. That's because, when keeping LLC mode on Auto, it results in a big VDroop (VCore reduction under high CPU load, to prevent an overshoot when the high CPU load suddenly stops and the voltage regulator has trouble reacting fast enough, causing a voltage overshoot if VDroop was not applied). So with CPU Lite Load, when lowering the mode there, the VDroop also has to be taken into account, so let's say we have to stay at CPU Lite Load "Normal" Mode 5, because otherwise, together with the big VDroop, the voltage under load wouldn't be enough for stability. But when we then look at low- to mid-load scenarios, VCore might be higher than necessary for stability, without the VDroop it might've been stable on Mode 2 or so. So what is then suggested, also set the LLC to something more aggressive (starting from LLC Mode 8 and going up towards Mode 1, it gets more and more aggressive in preventing VDroop, but Mode 1 is far too aggressive, usually you'd never go beyond Mode 4 or so).

What happens with tweaked LLC mode instead of Auto? Now the VDroop is less, so under full load, the voltage drops less. And now in turn, CPU Lite Load (either Normal or Advanced) can be lowered further than with LLC on Auto, because since the VDroop is less (prevented more), a lower CPU Lite Load settings may result in the same voltage than before (with a higher CPU Lite Load mode and LLC on Auto). With the added benefit that, supposedly, low- to mid-load scenarios of CPU workload now also have lower voltages than before.

However, this last step doesn't quite pan out for me so far in my full CPU load testing with this 12700KF on the PRO Z790-A WIFI. Because even if i set AC Loadline to 10, or even 1, the lowest possible value, and select a mild LLC setting like Mode 7, where the VDroop is still relatively big, i can't get anywhere near the low voltages and power draw i can reach just with a lowered CPU Lite Load and keeping LLC on Auto. I'm talking 20, 30 Watts difference here under full load. In other words, lowered CPU Lite Load with LLC on Auto results in the lowest possible VCore and power draw under full load, which is arguable one of the most important scenarios to keep the power draw low for (to avoid cooling problems and in general).

I have yet to find the perfect setting for this system. Whenever i think something is fully stable, sometimes i reboot, and the BIOS throws the "F1/F1 Overclock failed" screen, so it's back to square one. Although that screen could also have to do with the RAM-related voltages, i have to try to isolate it further, maybe starting from default settings everywhere again.

Sorry if this is a bit messy and long, i wrote this while i'm pacing back and forth between rooms to tweak things and check the results on this guy's future PC...



Just a usual short-term overshoot, if you reset the values (clock symbol) and then look at the average, it should stay pretty much on 175W. Don't go by the reported power values, as mentioned, they can also be a bit wrong, it's better to just go by CPU temperatures, which will inform if you have to lower the power limits further or if you're good.

Thank you so much for the detailed write up, mate. Really informative. I've been checking the VIDs against the VCORE with LiteLoad mode 7 and they seem to stay remarkable close. Currently, I am quite happy with the conservative settings I've imposed in the bios. I probably will stick to your first method above.

My current settings are:

• PL1 -170w; PL2 - 170w
• Core current at 307A
• IA CEP disabled
• LiteLoad Mode 7
• Enhanced Turbo disabled

I have now tested this extensively with Cinebench R23, R15, OCCT, and Prime95. Zero crashes and my temps are staying below 80 degrees and VCORE doesn't go above 1.35v now.

As I am just using this system for gaming, I am pleased with the results.

I actually have never had any stability issues with this CPU, but I have always reduced the PL values below Intel's recommended 253w. It's only recently that I've been playing around with other values in the bios (largely due to your support BTW).

Hopefully I can keep this system stable for a years to come. If I had known the i7 14th gen was such an untamed horse, I probably would have gone with AMD. But then I remember hearing about AMD and burning motherboards / boot times that go into the minutes around the time I was buying this PC...

 

[citay]

I wonder, though, how much lower in reality you can go below the CEP "happy stage" and still remain stable under all conditions. That's a bit of an unknown. You can think your PC is stable and then up pops a WHEA error out of nowhere after a month of use.

The thing is, CEP immediately puts an axe to the performance if you even slightly leave its preferred window. This is some window that Intel determined, and we all know by now that not everything Intel determines is necessarily good. Of course, better to not go to the edge of stability, this is something i always mention. And certain stress-testing is a must, throw all kinds of tests, workloads and scenarios at the CPU to see if it's stable. But to stop at some early point simply to keep IA CEP happy, i don't know if there is so much use to that.

Terrific news! You can help me dial mine in, if and when I pull it out of the box. You would help old Flyingscot, won't you?

I'll always help whenever i can, however, two CPUs of the exact same model are never identical, maybe you've seen this video. So if i tell you all my settings, they might be a decent starting point, but you never know what your CPU can do until you individually test it.

Thank you so much for the detailed write up, mate. Really informative. I've been checking the VIDs against the VCORE with LiteLoad mode 7 and they seem to stay remarkable close. Currently, I am quite happy with the conservative settings I've imposed in the bios. I probably will stick to your first method above.

Yes, i will keep recommending the conventional method as well, just trying to lower CPU Lite Load "Normal" mode a bit while staying stable. And before that, setting power limits so the cooling is protected.

I have now tested this extensively with Cinebench R23, R15, OCCT, and Prime95. Zero crashes and my temps are staying below 80 degrees and VCORE doesn't go above 1.35v now.

Yeah, this 12700KF is actually below 1.3V under load (had it stress-test-stable below 1.2V but sometimes getting boot problems, still investigating). But it's just a 12th gen, your 14th gen is a different beast, can't be compared. It all depends on the voltage without modifications, if it's a good improvement downwards from there. 170W limits are ok if this is used mainly for gaming, that doesn't cause the highest CPU load anyway.

 

[wroger196157202e2]

I am currently testing with a 12700KF on a PRO Z790-A WIFI with 2x 16 GB DDR5-6000, something i'm building for someone else (he really wanted an i7, and i said, ok, let's get a 12th gen one then, so he doesn't have to worry about CPU degradation).



You can go as low as possible without losing stability. There are two approaches. Normally i always recommend the first one, because it's less complex, and it does the job of lowering the power draw with one easy setting, if you test that it's fully stable. Usually after having set the power limits to the individual cooling situation.

So, first method: Just lower the mode for CPU Lite Load. Your only limit as far as lowering it should be the point where the CPU becomes unstable because the voltage becomes too low. Once you see instability in any tool (OCCT, y-cruncher, Prime95, whatever) or let alone in Cinebench (then it's quite unstable), you back off and test the next higher mode (so for example Mode 3 -> 4). Also, if the board shows an "F1/F2 Overclock Failed" screen during POST, or otherwise has trouble booting, the mode was too low. Once you find the lowest stable mode (say if Mode 4 is stable), i usually recommend to set one mode higher though (Mode 5 in this example), to have stability headroom. Done.

Note: If you see "IA CEP" intervening massively once you lower CPU Lite Load by more than 3 steps or so, meaning, when CPU performance like the Cinebench score massively goes down, then IA CEP will have to be disabled if you want to lower the voltages further. There doesn't seem to be any other way around it. But, since we are trying to lower the power draw, and since exposure to high voltages seems to be one of the key factors in this CPU degradation issue, i don't see why we should keep CPU Lite Load in a range where IA CEP stays happy. Yes, Intel recommended IA CEP to stay Enabled, but they were just listing every safety measure they could, while still investigating the cause for the degradation. Now they know, "exposure to elevated voltages" is the main culprit, so if we have a way to lower voltages further and keep the full performance (or sometimes even improve it, because it has more of the power budget to play with before the power limits kick in), then we should use it.

For a 14th gen i7 (by the far the most extreme i7 ever, way worse than the 13700K in regards of the power draw), and of course also the i9 models, you may not be able to lower the voltage by huge amounts. Because for their high frequencies, they need a high voltage, simple as that. Of course there are still better and worse CPU samples for every model, but on average, i found that the higher-up models can sometimes be a bit more resistant to shaving off voltage. There is no need to be hesitant about trying low modes though, just test for stability and you will soon see what it can do. Probably the only thing i would suggest is sticking to Mode 2 as the lowest to try, because Mode 1 only works with really good CPU samples.


Now, as some people have pointed out before, this method of lowering CPU Lite Load "Normal" mode is not perfect. The main two criticisms are:

1) It picks its own value combination for AC Loadline and DC Loadline. AC Loadline is what actually influences the voltage, it's a voltage added by the BIOS to make up for electrical properties of the CPU socket and so on. The background is not so important to understand, the main thing is, the higher this value is, the more voltage is added. This not only takes into account the electrical properties on the board, it also can make really bad CPU samples (when you lost the silicon lottery) run stable when set appropriately by default, or it can make CPUs be unstable from factory, if set too low by default. Lastly, if set too high by default, it will make the CPUs draw too much power and run too hot. The board makers have been going through various possibilities; Gigabyte at one time set the AC/DC Loadline way too high (equivalent to a high "CPU Lite Load" default mode on MSI), and even on MSI i recently saw one BIOS set Mode 22 which is among the highest possible, quite crazy. So they might have been trying to solve instability by adding more voltage, even though Intel now found that excessive voltage is what is causing problems.

For DC Loadline, "CPU Lite Load" on Normal (just changing the mode) will pick some value which might not result in the correct power draw readings anymore. Except that detail, the wrong value is not of much consequence. But if the cooling is such that power limits had to be set to prevent thermal throttling (which for a 14th gen i7 would almost universally be the case), then incorrect power draw readings will mess with the power limit throttling, it might set in too early or too late. That's where CPU Lite Load "Advanced" comes in.

In CPU Lite Load Advanced, you can select values for AC and DC Loadline seperately, without having some preset combination which can have the wrong DC Loadline. So now you can set the DC Loadline so it results in the correct power draw numbers. Doing that involves using HWinfo Sensors, creating full CPU load, then looking at the CPU's VID requests (in the "current values" column), which is the voltage the CPU asks for from the board, and comparing it to the current VCore value (note that i don't mean current as in Amperes, i mean current as in, actual live values, not Min/Max/Avg). If those are near-identical, the correct DC Loadline value has been found. The correct DC Loadline value depends on the LLC mode, another setting which influences the voltage (and not to be confused with CPU Lite Load / CLL, completely different). A table is here, but it's only a rough one, because each board model is built differently and would also need a bit different values, presumably.

2) It has been said that CPU Lite Load sometimes cannot be lowered as much when you don't also tweak the LLC mode. That's because, when keeping LLC mode on Auto, it results in a big VDroop (VCore reduction under high CPU load, to prevent an overshoot when the high CPU load suddenly stops and the voltage regulator has trouble reacting fast enough, causing a voltage overshoot if VDroop was not applied). So with CPU Lite Load, when lowering the mode there, the VDroop also has to be taken into account, so let's say we have to stay at CPU Lite Load "Normal" Mode 5, because otherwise, together with the big VDroop, the voltage under load wouldn't be enough for stability. But when we then look at low- to mid-load scenarios, VCore might be higher than necessary for stability, without the VDroop it might've been stable on Mode 2 or so. So what is then suggested, also set the LLC to something more aggressive (starting from LLC Mode 8 and going up towards Mode 1, it gets more and more aggressive in preventing VDroop, but Mode 1 is far too aggressive, usually you'd never go beyond Mode 4 or so).

What happens with tweaked LLC mode instead of Auto? Now the VDroop is less, so under full load, the voltage drops less. And now in turn, CPU Lite Load (either Normal or Advanced) can be lowered further than with LLC on Auto, because since the VDroop is less (prevented more), a lower CPU Lite Load settings may result in the same voltage than before (with a higher CPU Lite Load mode and LLC on Auto). With the added benefit that, supposedly, low- to mid-load scenarios of CPU workload now also have lower voltages than before.

However, this last step doesn't quite pan out for me so far in my full CPU load testing with this 12700KF on the PRO Z790-A WIFI. Because even if i set AC Loadline to 10, or even 1, the lowest possible value, and select a mild LLC setting like Mode 7, where the VDroop is still relatively big, i can't get anywhere near the low voltages and power draw i can reach just with a lowered CPU Lite Load and keeping LLC on Auto. I'm talking 20, 30 Watts difference here under full load (measured at the wall). In other words, lowered CPU Lite Load with LLC on Auto results in the lowest possible VCore and power draw under full load, which is arguable one of the most important scenarios to keep the power draw low for (to avoid cooling problems and in general).

I have yet to find the perfect setting for this system. Whenever i think something is fully stable, sometimes i reboot, and the BIOS throws the "F1/F1 Overclock failed" screen, so it's back to square one. Although that screen could also have to do with the RAM-related voltages, i have to try to isolate it further, maybe starting from default settings everywhere again.

Sorry if this is a bit messy and long, i wrote this while i'm pacing back and forth between rooms to tweak things and check the results on this guy's future PC...



Just a usual short-term overshoot, if you reset the values (clock symbol) and then look at the average, it should stay pretty much on 175W. Don't go by the reported power values, as mentioned, they can also be a bit wrong, it's better to just go by CPU temperatures, which will inform if you have to lower the power limits further or if you're good.

Citay....that F1/F1 Overclock failed screen is due to buggy XMP on this board(same as mine). That's why I was asking for help in the other thread. Same advice applies....turn XMP on and manually lower the speed from 6000 to 5600. It's stable after that.