Optimal settings Z790 Tomahawk Wifi Max i9 14900k

[MigraineFilms]

Hi all, as I was filling other threads with questions, here is a dedicated thread to my question.

Can you help me getting my i9 under control, it keeps thermally throttling even under very low power settings, and I dont fully grasp the undervolt settings.

My system:
i9 14900k
MAG Z790 Tomahawk Wifi max
Deepcool LT720 AIO
Contact Frame
DDR5 Trident Z 6400
Samsung 990 Pro
RTX 4080 Super
Corsair Shift RM1000x

So far I've done this:

Lowered PL1 and PL2 to 175 watts
Lite Load 9 (AUTO)
Voltage adaptive/offset -0.125 (screenshots still say -0.110, changed this, still throttles)

It can run for hours perfectly fine, benchmarks all doing great, cinebench R23 multiple runs, Prime95 30 minutes heat/stress test and then when I take a screenshot, HWiNFO says its thermally throttling without showing any high temperatures. HWiNFO also refreshes slowly, even on 1000ms, since the latest update.

Any advice is much appreciated

P.S: I do not experience any instability in the past 6 months I owned this computer, besides a sporadic lag in the audio, there are no crashes, app hangs, game failures, BSOD's or any of that kind.

Edit: All the information is a bit overwhelming and I'm feeling I'm getting lost in all the numbers a bit

 

[MigraineFilms]

Disabling C1E voids warranty?

No I was just making an example of settings they could remove.

This being said: when I set my power settings to MSI unlimited, a warning pop ups that using those settings may void warranty. Which is concerning, what if intel has a loophole to not replace the chips?

 

[Arctucas]

Which BIOS?

 

[MigraineFilms]

Which BIOS?

7E25vA81 BETA, Tomahawk Max Wifi X790

 

[MigraineFilms]

Spend 3 hours rendering chroma keyed footage, P-core 7 was 89 celsius at peak, 82 average, no thermal throttling whatsoever.
250 watt, 5,7ghz P core, 4,3 ghz E-core, worked like a charm!

 

[giggettoma15af02e3]

Hi everyone, sorry for the bad English.... I have the I9 14900K on the Z790 Tomawak WI FI MAX, 64GB RAM (2*32) at 6400 in XMP1 and an RTX 4090. The cooling is entrusted to an AIO 360 + 2 input fans and one output. Initially, since the CPU exceeded 90 degrees, I did undervolt (offset + adaptive) - 0.060 and the situation improved quite a bit, the system was stable and Cinebench24 finished rendering without problems (I don't give much weight to the score which was about 2180 in multicore anyway). I never managed to go beyond -0.060, I don't understand how you manage to go to -0.100 and have a stable system.
Anyway now with the new microcode, even Cinebench24 finishes rendering (score is now lower, about 2100), I started to find several errors in windows 11. Not BSDO, but inexplicable software errors detected by MSINFO32 (like Bad Bucket...).
Could it be due to a processor malfunction?

 

[citay]

I never managed to go beyond -0.060, I don't understand how you manage to go to -0.100 and have a stable system.

Each CPU is completely individual, this is what i also write in my guide (down where the video is linked). Each CPU has either more or less undervolting headroom, exactly as each CPU also has either more or less overclocking headroom (which is basically the same thing, in the opposite direction), according to its silicon quality. In other words, when you have two 14900K CPUs, or even a dozen CPUs, each one will be a bit better or a bit worse for undervolting, never the same. This wouldn't mean that your CPU is bad per se, it still passed validation to become a 14900K. It's just not comparable to the undervolting someone else has reached. Plus, all the best 14900K are binned to become a 14900KS anyway, so the chances to get a really good 14900K are not high.

As you can see in my guide, for undervolting i usually recommend the CPU Lite Load method (lowering the mode and stress-testing for stability). Because when you use the Offset/Adaptive method, the newer BIOS might apply a different (often higher) baseline voltage, so your previous offset values are not correct anymore. With CPU Lite Load, you don't have that problem, because that's the exact setting that they would change the default mode of in a different BIOS version. So you can just set the value again that you previously found to be stable, and it will do the job in any newer BIOS too.

It is possible for a 14900K to suffer from some CPU degradation, they were among the most affected models, but of course it depends on the use etc., and it's sometimes not that easy to find out. Usually, when you can still undervolt a bit (by whatever method) and it's not immediately becoming unstable everywhere, then the degradation shouldn't be bad. Because with really degraded CPUs, you have to actually add voltage, to gain some stability.

 

[MigraineFilms]

Because with really degraded CPUs, you have to actually add voltage, to gain some stability.

This is by far the best advice I discovered, and give to other people.

I would also like to advice to use the Lite Load method to undervolt. That said; I can only go as low as Lite Load mode 10 until it becomes unstable (a lot of people reach 7, 6, 5 or even 4). Manual undervolt makes me go lower, and shaves off more from the higher voltage loads when gaming.

Try the guide CiTay mentioned first, and when you get comfortable with undervolting (I was VERY anxious about it and had my fair share of crashes and mistakes, none of them damaging) you can try set the voltage offset manually.

For reference:
My i9 14900k, Tomahawk Max Wifi Z790 DDR5, 64 Gig DDR 6400, Deepcool LT 720 320 AIO, RTX 4080 Super, gets around 89 degrees celsius with undervolting, at a power limit of 253 watt.
Every CPU is different, every CPU has different results. I have currently set it to Power Limit 210 Watts, making it run 79 degrees celsius with 6% less performance. I've got a feeling that pumping 253 watt through the CPU everytime I edit ProRes 4444 footage is healthy for the CPU in the long term.

 

[giggettoma15af02e3]

Each CPU is completely individual, this is what i also write in my guide (down where the video is linked). Each CPU has either more or less undervolting headroom, exactly as each CPU also has either more or less overclocking headroom (which is basically the same thing, in the opposite direction), according to its silicon quality.

Congratulations for the guide that I read with great interest.
Currently I have the intel default settings (PL1 253W) and as I said "offset + adaptive" -60. Doing stress tests with AIDA64 I reach the max CPU temperature of 80°, while the average oscillates between 70° and 75° (depends on the ambient temperature). Now I want to try to set the CPU Lite Load as you suggest. However, I have some doubts, if I may:
1. I don't understand what the difference is between this setting and the Lite Load Calibration (which should lower the voltages as the load increases) and why the CPU LITE LOAD would be better than LLC;
2. is the CPU Lite Load compatible with undervolt? Or after setting it do I have to remove the undervolt? (to tell the truth I have the same question with reference to LLC);
3. with the temperatures that I already have by setting CPU LITE LOAD would I have improvements?
I hope I don't take too much advantage of your availability

 

[citay]

1. I don't understand what the difference is between this setting and the Lite Load Calibration (which should lower the voltages as the load increases) and why the CPU LITE LOAD would be better than LLC;

They are completely different. "CPU Lite Load" controls how much additional voltage the board adds to the voltage that the CPU requests. The higher the mode, the more is added. "Load Line Calibration" (LLC) controls the amount of Vdroop for the CPU voltage.

2. is the CPU Lite Load compatible with undervolt? Or after setting it do I have to remove the undervolt? (to tell the truth I have the same question with reference to LLC);

You could combine a lowered CPU Lite Load with another undervolting method. However, it does tend to become more complicated to test for stability, as you can introduce additional difficulty for mid-load scenarios for example. So for most people, lowering CPU Lite Load is not only the easiest, but also the perhaps most effective way to improve things, especially for full-load scenarios (where it matters most), while still giving a good benefit in mid-load scenarios. LLC does not need to be adjusted.

3. with the temperatures that I already have by setting CPU LITE LOAD would I have improvements?

The temperatures under full load depend mainly on the power draw under full load, so i would also compare the power draw when comparing different undervolting. And usually, you can lower the power draw very well by lowering the mode for this setting. It also depends a bit on things like your fan curves of course.

So, in summary, you should test the method from my guide. Reset everything from your previous undervolting method to defaults beforehand. It's better not to combine everything if you are not exactly sure what the result can be.

 

[giggettoma15af02e3]

They are completely different. "CPU Lite Load" controls how much additional voltage the board adds to the voltage that the CPU requests. The higher the mode, the more is added. "Load Line Calibration" (LLC) controls the amount of Vdroop for the CPU voltage.

So, if I understand correctly, the lower the CPU LITE LOAD value, the less voltage the card adds than normally required? Is this correct?

 

[FlyingScot]

So, if I understand correctly, the lower the CPU LITE LOAD value, the less voltage the card adds than normally required? Is this correct?

Hey, if you have the time and inclination, see the information that I linked in my signature. There’s some good bedtime reading in those links (to be read in sequence), especially if you like to read a good horror story for Halloween. It’s called the Intel Loadline mechanism.

 

[citay]

So, if I understand correctly, the lower the CPU LITE LOAD value, the less voltage the card adds than normally required? Is this correct?

There is always some extra voltage required, because of things like the contact resistance of the socket pins and so on, and of course the CPU's quality (an unknown), which all has to be compensated for. But the board (via the default BIOS setting) tends to add more voltage than is required, just to make things stable for most users and to have less support requests. Lately, they may even want to stabilize already degraded CPUs, it seems, by adding more voltage than ever. So then, if your CPU is not degraded and is of decent quality, the compensation in added voltage is vastly higher than would be required for stability. Therefore you can also improve things dramatically by lowering the CPU Lite Load mode to what your CPU is still fully stable with.

 

[giggettoma15af02e3]

Hi guys and thanks for the valuable advice. If they can be useful I will describe the tests I did...
1. Through HWINFO I saw that with the simple offset+ adaptive -60 I had as core VIDs maximum 1.455V, while the default value of CPU LITE LOAD was 18: the max temperature with AIDA64 stress test reached 80°.

2. I then decided to follow your advice and I set CPU LL to 9 bringing offst+ adaptive to -50: the temperature actually dropped quite a lot, but a simple CPUZ bench showed that the CPU result in multi core was almost halved, while in single core it remained unchanged.

3. I then disabled AI CEP as per CITAY guide, so CPU LITE LOAD 9, AI CEP DISABLED and ADAPTIVE+OFFSET -50: the maximum Core VIDs value dropped to 1.349V and the max temperature during the AIDA64 stree test dropped to 68°; now with CPUZ bench I have a better result (even if only slightly) than test 1., both in multi and single core, furthermore the system has not shown signs of instability even in gaming so far.

I attach the HWINFO screens of test 3. during the AIDA64 stress test and the CPUZ results achieved. IN HWINFO the RAM value is in red because I also launched the RAM stress test at the same time. What do you think of the results achieved?

However, I have a doubt: AI CEP protects the CPU from excessive voltage differences. so it should work like the differential that we have in our electricity meters, that is, it should intervene if an excessive voltage difference is created, which could be either in addition or in decrease. Now, in the case of excessive drops, there would be no hardware problems, but only software, that is, there would be system instability. But in the case of increasing peaks, the CPU would practically no longer have protection and there could be hardware problems.
It is a reasoning that I made by deduction and I am not sure of the correctness of the consideration that I made, what do you think?

 

[citay]

This is all pretty much as expected. Lowering CPU Lite Load is highly effective in improving how the CPU is running. Then however, if you don't disable IA CEP at first, you will have the performance cut in half. Disable it and it's back to normal, or even better.

But in the case of increasing peaks, the CPU would practically no longer have protection and there could be hardware problems.

Don't worry, the voltage spike protection/prevention is in the microcode that the BIOS loads into the CPU during the early boot stage. There is no need to rely on IA CEP for it, it will not prevent anything of importance when it's interfering here. We want to lower the voltage, which is the best thing you can possibly do for a CPU, as you can see for yourself. IA CEP only gets in the way. If you selected the MSI Performance cooler setting, it should set IccMax (CPU Current Limit) of 307A, this limit will act independently from IA CEP and should be enough to keep things in check on the currents side of things. The voltage we have nothing to worry about, we lower the baseline voltage and the spikes are capped by the microcode.

 

[MigraineFilms]

Just a little update:

PL 253
iccmax 307
AC40
DC45
LLC6
-0.100 undervolt
results in R23 39500 points, max temps of 89 degrees celsius

Same settings but PL230 results in R23 38800 points and 75 degrees celsius. Huge temp difference, neglegable performance hit

 

[FlyingScot]

Just a little update:

PL 253
iccmax 307
AC40
DC45
LLC6
-0.100 undervolt
results in R23 39500 points, max temps of 89 degrees celsius

Same settings but PL230 results in R23 38800 points and 75 degrees celsius. Huge temp difference, neglegable performance hit

After hearing the latest Arrow Lake rumors, which includes lots of BSOD issues on top of the rumored poor comparable gaming performance versus a 14900K, and even a 13900K, I think the new strategy for Raptor Lake owners should be “Better the devil you know.”

Personally, I think Intel needs to think hard about their naming conventions because things took a turn for the worse after Comet Lake (still my favorite Intel CPU). What I mean by this:

Rocket Lake = blew up on the launchpad
Raptor Lake = bit everyone on the butt
Arrow Lake = destined to break everyone’s heart

 

[MigraineFilms]

Arrow Lake = destined to break everyone’s heart

As rumors go, I think Arrow Lake will be a monster of a production chip, albeit STILL at 253 watt, with high temperatures. It will not be a proper gaming chip, but thats not really due to the frequency or efficiency, but the X3D method from AMD being way better in games, Intel needs its own version of that to compete again.

 

[FlyingScot]

As rumors go, I think Arrow Lake will be a monster of a production chip, albeit STILL at 253 watt, with high temperatures. It will not be a proper gaming chip, but thats not really due to the frequency or efficiency, but the X3D method from AMD being way better in games, Intel needs its own version of that to compete again.

That’s true. Rumors are that it could be even more efficient than Zen5 in some workloads. But if other rumors are true, I can’t believe that Intel are going to repeat the mistakes of RPL and launch yet another product with buggy microcode. That’s hard to believe. It’s like we now live in a world where no one tests anything. Tomorrow should be very interesting.

 

[giggettoma15af02e3]

results in R23 39500 points, max temps of 89 degrees celsius

This is strange...
We have a similar configuration, but with different settings...

With LLC9 and undervolt -0.050 I get 37095 points on r23 (I didn't close the background programs though), but the strange thing is that the maximum temperature that the system reaches is 70°...
In fact I should reach higher temperatures, since I have less offset than you and a higher LLC than yours.
My other settings are default (except AI CEP that I disabled, obviously).
Can I ask you the voltage values that HWINFO reads with your settings?

 

[MigraineFilms]

A higher LLC will cause more vdroop, but then it will become unstable quickly and give me high medium load voltages (right now, they will not exceed 1,36 at most)

Here is a screenshot during a R23 run, background apps still running, that causes about 1000 points difference
Do note that I have ICCMAX at 307 at the moment, even though that gives performance limit warnings, it doesnt cause any noticable performance loss whatsoever.