Explained: How the new BIOS versions are causing higher temperatures

citay

Pro
SERGEANT
Joined
Oct 12, 2016
Messages
24,836
Note: This thread mostly contains the investigation and explanation for the higher power draw on the latest BIOS versions.
It doesn't go so much into detail about the solution, which can be found in my Guide: How to set good power limits in the BIOS and reduce the CPU power draw.


A lot of people have reported higher CPU temperatures after updating to the latest BIOS version for their 600-/700-series Intel board. Especially after updating to the version containing the 0x129 microcode revision (hotfix for the voltage spike problem with the true 13th/14th gen CPUs), or newer versions.

Now, right from the start, the logical explanation for a higher power draw (when applying the same workload, and if the power limits don't mask anything) would be a higher voltage. But what exactly is causing the higher voltage, which is leading to higher power draw, leading to more heat, leading to higher temperatures?

The hot candidate is "CPU Lite Load", which influences the the CPU voltage via the so-called AC and DC loadlines. The important one is the AC loadline, a voltage added by the BIOS to make up for electrical properties of the CPU socket and such. 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 of the board and the CPU socket, it also can make really bad CPU samples run stable (when set appropriately by default), or it can make CPUs be unstable from factory (if set too low by default). Finally, if set too high by default, it will make the CPUs draw too much power and run too hot.

But by what mechanism is there now more voltage added for the CPU? Does MSI choose a higher default mode for CPU Lite Load, and why? I wanted to find out, and since the stable BIOS with the 0x129 microcode just came out for my board model (at the time of writing this), i put it to the test. Now, i am using an i5-13500, so, not a true "Raptor Lake" CPU like a 13600K or 14600K for example (which are definitely affected by the microcode bugs), but an "Alder Lake" 12th-gen-based one. So, it being a 12th gen in disguise, it actually neither needs (nor uses) the newer microcode, it uses Alder Lake microcode instead. But still, i wanted to see if my CPU's voltage and thus power draw also go up.

So, on the old/previous BIOS version, i had CPU Lite Load optimized to Mode 4 for my specific CPU.
This means, Mode 4 is stable for that CPU, with a bit of stability headroom (Mode 3 was verified stable, then i raised it by one step).

First, taking some baseline numbers. The default for "CPU Lite Load" - with my CPU and my board, on the older BIOS version 7D32v1H - was Mode 12:

CPU Lite Load 1H0 Mode 12.png

Click to enlarge

I did some other optimizations there, like enabling all power-saving mechanisms plus Intel Speed Shift Technology. Most of that only lowers idle power draw though.

Then, updating to 7D32v1J:

2024-08-25 18.32.14.jpg


After the update, here's the revised cooler selection screen, which is really the power limit selection screen:

MSI_SnapShot_01 Intel Def.png


I chose the middle option, even though i already knew that - with my cooler and CPU - i would not even reach the middle option's limits. So i also could've chosen the bottom option with the maxed out limits, wouldn't matter in my case. My cooler can easily deal with my CPU's heat, so i could optimize the fan curves for low noise output. But for most people, choosing the middle option "MSI Performance" is a good starting point, from which they can lower the power limits if necessary. Because "MSI Performance" includes the highest power limits that make sense to allow.

MSI_SnapShot_03 MSI Performance.png


Note: The BIOS first has the values for the first option "Intel Default Settings" loaded. So after the middle option is selected, the menu under OC will still show the "Intel Default Settings" values, until you press F10 to save and exit, then the "MSI Performance Settings" are applied. But we know those three options are not that well-fitting for most people anyway, because everyone combines a different CPU with a different cooler. I just chose the middle option because it happens to have the maximum values i would allow for any CPU (even an i9). If there is any thermal throttling with those limits, they have to be optimized to the individual cooling capabilities, which i explain in my guide.

Now, after the update, we're on the new BIOS (the one with the 0x129 microcode). Let's check what the new default settings are. Remember, CPU Lite Load on Auto, in the old BIOS version, resulted in Mode 12. This was still quite high for my CPU, considering it was fully stable at Mode 4. So there's eight steps worth of additional voltage added to VCore, in order to make all CPUs of varying quality work.

Now on the new BIOS:

CPU Lite Load 1J0 Mode 18 defaults.png


Blimey! The new default is Mode 18! I wonder what that will do to the voltages, the power draw, the heat and the temperatures? Nothing good, i can already tell you.
Of course, some other settings were also reset. I enabled them all again manually, but kept CPU Lite Load on Mode 18 for testing.

CPU Lite Load 1J0 Mode 18.png


Now, about the testing, for Cinebench R15, i used Cinebench R15 15.0.37 with Extreme Edition mod, just to explain the oddly low scores for that.

For power draw testing, i mostly relied on an energy meter that's plugged in at the wall socket (actually, at the UPS), for the power cable going into the PSU. This energy meter / power draw measurement device is very exact and, unlike the sensors in the system, cannot be wrong. Additionally i took some measurements from the "CPU Package Power" sensor via HWinfo, which is the CPU-only power draw.

Here is the full comparison:

CPU Lite Load results.png


What can we see from this? All the scores stay basically the same, no matter which mode is active for CPU Lite Load. On some boards, for it to be like this, one would have to disable the "IA CEP Support" setting like i describe in my Guide: How to set good power limits in the BIOS and reduce the CPU power draw. On my board however, this setting is not available (as shown on the screenshots), and with my CPU and board combination, IA CEP clearly doesn't intervene, otherwise the scores would be cut in half with CPU Lite Load Mode 4. But they all stay almost identical within the margin of error.

So, the performance stays the same, but what about the power draw? On the old BIOS, using the default Mode 12 is already quite inefficient. Power draw can be a few dozen Watts higher than it would need to be for this CPU, due to higher-than-necessary Vcore that's applied by Mode 12. So optimizing this setting down to what the CPU actually needs for full stability (in my case, eight steps down to Mode 4) pays off nicely. Everything about how the CPU is running improves, and the scores stay the same. If my CPU was actually hitting a power/temperature limit, then the scores would even improve with Mode 4, because compared to Mode 12, the "power/temperature budget" simply lasts longer, and the CPU can clock higher within those limits.

But now look what happens on the new BIOS, MSI have a new default of Mode 18. This is a catastrophy, now my CPU is not just running eight steps above what it would need for full stability, it's running 14 steps above it! We're seeing 30-50W higher power draw (CPU only, for the whole PC it's up to 90W more) than necessary, and that's just on my lowly i5-13500. On an i7 or i9, the difference would be tremendous, because there are more cores and higher frequencies. And of course, the scores stay the same, the stability stays the same (there is no "more stable than stable"), but everything else has worsened considerably!

So this explains how the temperatures can be so much higher on the latest BIOS versions: The mode for CPU Lite Load has been raised considerably by default. Because it looks like what MSI is doing now is, they're adding a huge safety headroom for the default CPU voltage, most likely in an attempt to stabilize certain CPUs that have already degraded and have a bit of instability.

Their rationale might be, now that there's a voltage limit in place to take care of the voltage spikes, they can happily raise the default voltage (via a higher default CPU Lite Load mode) to stabilize shaky/unstable CPUs, basically the victims of the voltage spike bugs in the microcode. And that actually works for those CPUs that suffered degradation. But for everyone else with a stable CPU, this makes everything a lot worse!

So it has become even more important to try and lower the voltage, otherwise a stable CPU will have needlessly high power draw in all load states, effectively lowering the power and temperature budget and ultimately costing performance. This becomes evident due to instantly improved performance as soon as you undervolt (provided the CPU is hitting a power/temperature limit, which most 14th gen i7/i9 will do unless your cooling is out of this world).

Once you go by my guide, then any higher temperatures can be completely taken care of, because in step 1) you set safe power limits for your cooling, and in step 2) the voltage will be lowered to what your CPU sample actually requires (plus a bit of headroom). This is literally all that is required to bring down the temperatures, either to the level of the older BIOS version when those things were already optimized, or to a better level than ever before if they weren't.

Note: If you undervolted with an offset before (instead of lowering CPU Lite Load), or a combination of the two methods, then the offset undervolt will now happen from a higher baseline voltage. So the best thing in that case is to take note of the previous mode for CPU Lite Load, and apply it again on the new BIOS. The default mode in the newest BIOS version is crazy high! I don't know what they're thinking. Well, i have an idea, but i don't think they're doing anything good by this. For the vast majority of users, the CPU will run worse than before. Going by my guide linked at the very top, this can luckily be reversed.

To round this off, let's look at the "calculation efficiency" of the system in Cinebench R23 with different settings (higher is better):
Old default, CPU Lite Load 12: 119 points per Watt.
New default, CPU Lite Load 18: 100 points per Watt.
Optimized CPU Lite Load Mode 4: 145 points per Watt!

Mode 12 wasn't very efficient to begin with, and the new Mode 18 is just horribly inefficient.
Mode 4, which is still fully stable with my CPU and achieves the same performance, has much higher efficiency.


Lastly, on the far right, i did an additional test, checking the benefits of setting CPU Lite Load to Advanced (using the same AC loadline setting that Mode 4 results in), but optimizing the DC loadline setting so the VID matches the Vcore under full load. 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 value. 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. If those are near-identical, the correct DC Loadline value has been found.

VID.png


I have done this, and the result for my CPU on my board was AC loadline 30, DC loadline 117, which can also be read out in HWinfo later:

ACDCloadline.png



The concerns about CPU Lite Load "Normal" (that it won't always show the correct CPU Package Power anymore because the DC loadline is usually not properly adjusted to where it would need to be) are somewhat put into perspective. We have a mere 6W difference from the reported CPU power draw to the "actual" CPU power draw, under the highest load any normal program can create (CB 23 is fully multithreaded AVX load, but Prime95 uses dirty tricks, it's only used for stability testing). This is not gonna make or break out power limits, if we have had to set some.

And even with the reported power draw being slightly off like this on CPU Lite Load "Normal", this doesn't affect us much, we can just go by the maximum CPU temperature to inform us if our power limits are properly dialed in, or if we still need to adjust them according to our cooling. Plus, explaining CPU Lite Load "Advanced" makes it more complicated, which means less people will do it. So i think CPU Lite Load "Normal" is a good compromise.

By the way, this is what resistance/impedance in mΩ (milliOhm) the different CPU Lite Load settings correspond to, valid for both old and new BIOS versions:
CPU Lite Load Normal, Mode 4: AC loadline 0.3 mΩ, DC loadline 0.3 mΩ (this is what i lowered the mode to, verifying that it's stable)
CPU Lite Load Normal, Mode 12: AC loadline 1.1 mΩ, DC loadline 1.1 mΩ (this is the default on the older BIOS versions)
CPU Lite Load Normal, Mode 18: AC loadline 1.7 mΩ, DC loadline 1.7 mΩ (this is the way too high default on the latest BIOS version)
CPU Lite Load Advanced, AC 30 / DC 117: AC loadline 0.3 mΩ, DC loadline 1.17 mΩ (so this way you can set them both directly).
Note: It's possible that some other board/CPU combinations have somewhat different values for a certain mode. They can be read out in HWinfo, as shown above.


Conclusion:
The explanation for the higher temperatures is very simple: Needlessly raised default mode for "CPU Lite Load", causing higher voltage.

Never has it been more important to optimize each Intel CPU in each system individually, according to the cooling and according to what voltage it's running stable with. On the default settings of the latest BIOS versions, the voltage / power draw / heat / temperatures (one influences the next) are higher than ever! With any CPU that is running into power/temperature limits (so, either power limits that you have set to protect your cooling, or failing to do that, the thermal throttling that can happen), the performance will decrease as a result of the new BIOS defaults!

Luckily, with the help of my guide, all those parameters can be improved again: Voltage down, power draw down, heat down, temperatures down, performance identical or up!
This has no downsides other than investing some time for finding good values and testing that it stays stable. Your CPU and your cooling will be very thankful for that effort.
 

Attachments

  • MSI_SnapShot_26.png
    MSI_SnapShot_26.png
    204.5 KB · Views: 9,844
Last edited:
I do not know, but I find that setting VCore to a fixed value in BIOS appears to keep voltages in check, with the caveat that no, I cannot see the transients because I do not have an O-scope. With C-states, Speedstep and Windows Power plan, idle voltages are in the low 0.6 range, load voltages are never over 1.27. I am using the profile for a KS (PL1 = 320, PL2 = 320, IccMax = 400).

So, limiting the VCore also limits the Icc, or not?
It's funny that you two are discussing this because I recently started wondering if a fixed voltage with adaptive offset (assuming that it works the way I think I've seen it done) would be a possible way to implement the equivalent of IA VR Voltage Limit.
 
Last edited:
So, limiting the VCore also limits the Icc, or not?

Like i said, to get a general idea, just look at the power draw, a variable of the equasion P = I * V. You don't really have any unknowns if you know the value for two of the variables.

In general, the last word on the BIOS updates is far from having been spoken. I think we're gonna see a couple more updates and microcode revisions dealing with this issue. By the end of which, hopefully nobody has to resort to running a fixed voltage.
 
Like i said, to get a general idea, just look at the power draw, a variable of the equasion P = I * V. You don't really have any unknowns if you know the value for two of the variables.

In general, the last word on the BIOS updates is far from having been spoken. I think we're gonna see a couple more updates and microcode revisions dealing with this issue. By the end of which, hopefully nobody has to resort to running a fixed voltage.

Having done the math on my rig, there is a discrepancy with that (I will do some more checking on that tonight). Now, that is probably due to inaccuracies in the software and/or the sensors.

Anyway, would you agree that having the option to set a voltage limit limit would be worthwhile? Seeing as how the high voltage is the problem, of course.

I believe we are all seeking solutions to this fiasco. Shame that end users need to fix Intels screw-ups.
 
Anyway, would you agree that having the option to set a voltage limit limit would be worthwhile? Seeing as how the high voltage is the problem, of course.

Yep, mentioned it here (last paragraph). Any such powerful tool that can help would be a welcome addition.
 
I want to thank citay for the useful and informative tutorials to help address Intel's CPU issues. Some of the information was beyond my level of understanding but I used this thread to reset my BIOS settings to provide a safe environment for my CPU while maintaining it's performance level.
 
a possible way to implement the equivalent of IA VR Voltage Limit
voltage requested by CPU
= raw_VID + offset + (AC_LL * predicted_current) + (some minor stuff)

where

offset and AC_LL: we can set ourselves
predicted_current: always less or equal to iccmax
raw_VID: fixed, depends on CPU freq

Thus one can have a poor man's "IA VR Voltage Limit" quite easily.

For example, my 14700k has these settings: P-core limit 52x, lite load mode 6, iccmax 249, zero offset. All these together give a limit on what voltage CPU can request:

(raw_VID at 52x) + (0.30 mOhm * 249 ampere)
= (raw_VID at 52x) + 0.0747v

If raw_VID at 52x is 1.24v, then this gives 1.3147v as limit.

Note that LLC is applied afterward.
 
Last edited:
voltage requested by CPU
= raw_VID + offset + (AC_DL * predicted_current) + (some minor stuff)

where

offset and AC_DL: we can set ourselves
predicted_current: always less or equal to iccmax
raw_VID: fixed, depends on CPU freq

Thus one can have a poor man's "IA VR Voltage Limit" quite easily.

For example, my 14700k has these settings: P-core limit 52x, lite load mode 6, iccmax 249, zero offset. All these together give a limit on what voltage CPU can request:

(raw_VID at 52x) + (0.30 mOhm * 249 ampere)
= (raw_VID at 52x) + 0.0747v

If raw_VID at 52x is 1.24v, then this gives 1.3147v as limit.

Note that LLC is applied afterward.
Unfortunately, the one variable we cannot control is predicted_current and that’s what we think leads to the 1.6V spikes. Setting manual Ratios is a workaround.
 
Last edited:
Unfortunately, the one variable we cannot control is predicted_current and that’s what we think leads to the 1.6V spikes.
According to this post (and the related discussions) from overclock.net, we can control the max of predicted_current via iccmax.

IccMAX is the predicted current limit, not actual current limit.
IccMAX controlls the exactly same predicted current that make ACLL to boost voltage.
 
I’ll see if I can find it. It could be one of a few he did recently.
If I remember correctly (it was maybe the opposite of what I said earlier), IccMax had an effect down up to approximately 225/250A, but then stopped having any more effect all the way down to 150A. up to 500A. Presumably, VID requests from 307A down to 240ish were still too high. Maybe from 400 to 307 some benefit could be had, but most people are running the 307A already. Either way, I don’t think BuildZoid saw the IccMax as a reliable way to stop the high spikes. Some other variable/modifier is at work. If I stumble across the video where he talks about it, I will post the link. But it could be anywhere across 10 hours of his usual ramblings. And I’m getting tired of listening to this particular series of videos.
 
Last edited:
All I can say is, based on my personal experience, that setting manual VCore appears to limit the voltage.

I am not suggesting anyone use my settings, just try to find out what works for you.

Perhaps there is something abnormal with my motherboard, BIOS, CPU, RAM, PSU, etc. that allows me to do this while others cannot, I do not know.
 
All I can say is, based on my personal experience, that setting manual VCore appears to limit the voltage.

I am not suggesting anyone use my settings, just try to find out what works for you.

Perhaps there is something abnormal with my motherboard, BIOS, CPU, RAM, PSU, etc. that allows me to do this while others cannot, I do not know.
I can’t think of a reason why a fixed voltage wouldn’t have a positive impact on the voltage spikes. It would have to act in a similar way to the IA VR Voltage Limit, i,e, preventing the CPU from making requests that are higher than it. But as CiTay points out, it wouldn’t be my first choice if we were trying to come up with a universal way of helping everyone cool down their Raptor Lakes. Although, I would certainly be willing to experiment with it.
 
Last edited:
I actually found it. You may find the whole video worth listening to, so you have the complete context. However, it’s at the 33min mark where he discusses the strange relationship between VID and IccMax.
Thanks !

1. regarding whether the inequality predicted_current <= iccmax holds:

In BuildZoid's video dated 2024/08/13, at around 33min mark, he did talk about the relationship between VID and iccmax. He started with a load of ~150A actual current and ~225A predicted_current. Then he varied iccmax, and checked how iccmax affected voltage requested by cpu = raw_vid+offset+(ACLL*predicted_current). He found that if iccmax is set above ~225A, then iccmax has no effect on vid. This is exactly the behavior we expect: when iccmax is set higher than the calculated predicted current, then no throttling is necessary. We expect that setting iccmax lower than ~255A would lower vid in this particular load.

Think of it this way: if we have a load that cpu requests 1.2v, then setting IA VR Voltage Limit above 1.2v would have no effect. But setting IA VR Voltage Limit below 1.2v would throttle the CPU and lower vid. BuildZoid is just looking at the current limit instead of voltage limit.

(I think BuildZoid was testing the hypothesis predicted_current=iccmax, and he found out predicted_current is not the same as iccmax.)


In BuildZoid's video dated 2024/09/01, at around 01:22:38, he briefly mentioned that one can lower iccmax and avoid cpu requesting high voltage. So he seems to believe iccmax controls predicted_current, and is useful in stopping high spikes.


Then at around 01:35:03 of the same video, he talked about i5 with stock iccmax of 200amp pulling 250amp. My interpretation is that here he is talking about motherboard vendors failing to honor intel's recommendation of 200A by setting higher default iccmax in BIOS. In fact, he went on to re-affirm that lowering iccmax lowers the vid request.


Recall that intel's document states "IccMax is the maximum current the processor can draw, typically seen running a virus application (stress applications specifically designed to push the SoC to maximum power)."

I highly doubt predicted_current used in vid request can be higher than iccmax. If BuildZoid has ever demonstrated that predicted_current can be higher than iccmax, that would be very interesting and counter-intuitive.


By the way, recently Bonestar post a pic at overclock.net with IOUT > iccmax, but the consensus there seems to be that the IOUT readout is incorrect. You might be interested in this picture provided in one of the replies, which gives interesting insight into iccmax and IA CEP.


2. regarding effectiveness of this poor man's "IA VR Voltage Limit"

I think it is well-suited for people who use AC_LL undervolt method recommended by citay.

For example, if we set AC_LL=10, iccmax=400A, then the 2nd term is

(AC_LL*predicted_current) <= 0.04v

which leaves a lot of room for raw_vid+offset.

For people who want to keep IA CEP enabled, AC_LL is much higher than 10. Thus the 2nd term is larger and this formula does not give very effective voltage limit.


In fact, in BuildZoid's video dated 2024/08/13, at around 36m mark, he says something like "So basically, if you set your AC load line really, really low, even if you disable v-droop on the VRM, you don't get those giant voltage spikes that you get with the sort of typical Intel default behavior."

==========
EDIT:

At overclock, people are discussing iccmax, predicted current, and peak current, etc. [post]

Anyway, a more extreme method is to set
- AC_LL=1,
- LLC=sane level,
- test stability, change offsets if necessary

With this method, the 2nd term, AC_LL*predicted_current, is essentially eliminated (less than 0.01v). Thus vid request depends mostly on raw_vid+offset, which one can put limit on easily.
 
Last edited:
Thanks !

1. regarding whether the inequality predicted_current <= iccmax holds:

In BuildZoid's video dated 2024/08/13, at around 33min mark, he did talk about the relationship between VID and iccmax. He started with a load of ~150A actual current and ~225A predicted_current. Then he varied iccmax, and checked how iccmax affected voltage requested by cpu = raw_vid+offset+(ACLL*predicted_current). He found that if iccmax is set above ~225A, then iccmax has no effect on vid. This is exactly the behavior we expect: when iccmax is set higher than the calculated predicted current, then no throttling is necessary. We expect that setting iccmax lower than ~255A would lower vid in this particular load.

Think of it this way: if we have a load that cpu requests 1.2v, then setting IA VR Voltage Limit above 1.2v would have no effect. But setting IA VR Voltage Limit below 1.2v would throttle the CPU and lower vid. BuildZoid is just looking at the current limit instead of voltage limit.

(I think BuildZoid was testing the hypothesis predicted_current=iccmax, and he found out predicted_current is not the same as iccmax.)


In BuildZoid's video dated 2024/09/01, at around 01:22:38, he briefly mentioned that one can lower iccmax and avoid cpu requesting high voltage. So he seems to believe iccmax controls predicted_current, and is useful in stopping high spikes.


Then at around 01:35:03 of the same video, he talked about i5 with stock iccmax of 200amp pulling 250amp. My interpretation is that here he is talking about motherboard vendors failing to honor intel's recommendation of 200A by setting higher default iccmax in BIOS. In fact, he went on to re-affirm that lowering iccmax lowers the vid request.


Recall that intel's document states "IccMax is the maximum current the processor can draw, typically seen running a virus application (stress applications specifically designed to push the SoC to maximum power)."

I highly doubt predicted_current used in vid request can be higher than iccmax. If BuildZoid has ever demonstrated that predicted_current can be higher than iccmax, that would be very interesting and counter-intuitive.


By the way, recently Bonestar post a pic at overclock.net with IOUT > iccmax, but the consensus there seems to be that the IOUT readout is incorrect. You might be interested in this picture provided in one of the replies, which gives interesting insight into iccmax and IA CEP.


2. regarding effectiveness of this poor man's "IA VR Voltage Limit"

I think it is well-suited for people who use AC_LL undervolt method recommended by citay.

For example, if we set AC_LL=10, iccmax=400A, then the 2nd term is

(AC_LL*predicted_current) <= 0.04v

which leaves a lot of room for raw_vid+offset.

For people who want to keep IA CEP enabled, AC_LL is much higher than 10. Thus the 2nd term is larger and this formula does not give very effective voltage limit.


In fact, in BuildZoid's video dated 2024/08/13, at around 36m mark, he says something like "So basically, if you set your AC load line really, really low, even if you disable v-droop on the VRM, you don't get those giant voltage spikes that you get with the sort of typical Intel default behavior."

==========
EDIT:

At overclock, people are discussing iccmax, predicted current, and peak current, etc. [post]

Anyway, a more extreme method is to set
- AC_LL=1,
- LLC=sane level,
- test stability, change offsets if necessary

With this method, the 2nd term, AC_LL*predicted_current, is essentially eliminated (less than 0.01v). Thus vid request depends mostly on raw_vid+offset, which one can put limit on easily.
Thanks. I will carefully review the information you have assembled and linked to. My VID research is currently a side interest, so this information comes at a good time.
 
I tried the suggested settings for CPU Lite Load Normal and although it lowered the package power and temperatures - great, it also lowered the OCCT benchmark for multithread & Intel XTU numbers by more than 40%. Here are the results, the only change between the runs is CPU Lite Load:

CPU Lite Load Normal, Mode 10 (same as default values for Avdanced): AC loadline 0.5 mΩ, DC loadline 1.1 mΩ
OCCT Benchmark: Single SSE 108.88 / Multiple SSE 747.89 / Single AVX 194.46 / Multiple AVX 1132.03

CPU Lite Load Normal, Mode 12: AC loadline 0.7 mΩ, DC loadline 1.1 mΩ
OCCT Benchmark: Single SSE 110.73 / Multiple SSE 1164.23 / Single AVX 194.98 / Multiple AVX 1790.54

CPU Lite Load Normal, Mode 16 (same as Normal:Auto, same as Intel Default): AC loadline 1.1 mΩ, DC loadline 1.1 mΩ
OCCT Benchmark: Single SSE 111.33 / Multiple SSE 1375.46 / Single AVX 201.1 / Multiple AVX 2085.57

As you can see the Multiple scores are almost half.

I'm on MSI MAG Z790 TOMAHAWK MAX WIFI (MS-7E25), running i7-14700K, power limit to 325W, current to 395A. I'm on August bios, A.70.

Any explanation for the results? @citay results suggest that perf should remain in the same ballpark.
 
Last edited:
I tried the suggested settings for CPU Lite Load Normal and although it lowered the package power and temperatures - great, it also lowered the OCCT benchmark for multithread & Intel XTU numbers by more than 40%. Here are the results, the only change between the runs is CPU Lite Load:

CPU Lite Load Normal, Mode 10 (same with default values for Avdanced): AC loadline 0.5 mΩ, DC loadline 1.1 mΩ
OCCT Benchmark: Single SSE 108.88 / Multiple SSE 747.89 / Single AVX 194.46 / Multiple AVX 1132.03

CPU Lite Load Normal, Mode 12: AC loadline 0.7 mΩ, DC loadline 1.1 mΩ
OCCT Benchmark: Single SSE 110.73 / Multiple SSE 1164.23 / Single AVX 194.98 / Multiple AVX 1790.54

CPU Lite Load Normal, Mode 16: AC loadline ? mΩ, DC loadline ? mΩ
OCCT Benchmark: Single SSE 111.33 / Multiple SSE 1375.46 / Single AVX 201.1 / Multiple AVX 2085.57

As you can see the Multiple AVX score is almost half.

I'm on MSI MAG Z790 TOMAHAWK MAX WIFI (MS-7E25), running i7-14700K, power limit to 325W, current 395A.

Any explanation for the results? @citay results suggest that perf should remain in the same ballpark.
Quick question. Is CEP enabled?
 
Last edited:
Yep, forgot to disable CEP. With it disabled I was able to go down to Mode 4 but it is not stable with OCCT Multiple AVX or Cinebench crashing and WHEA reporting CPU cache L0 errors.
Should I attempt to fix it further? Some sources say this is a RAM problem and I need to set the clock and voltage manually (I'm running XMP Profile 2 on Corsair Dominator DDR5-6200 @ 3200MHz now). Or just back up CPU LL to Mode 5 or 6.
 
Back
Top