Intel Core i9-12900K Alder Lake Tested at Power Limits between 50 W and 241 W

[JustBenching]

Just FYI, on Intel CPUs it is possible to set the "DC Loadline" parameter to very low or very high values and massively change up or down reported package power and VID, without any difference in actual power consumption or voltages.

When DC Loadline has been correctly set, VID matches Vcore under load.

VCCCORE DC Specifications - 001 - ID:655258 | 12th Generation Intel® Core™ Processors Datasheet, Volume 1 of 2

edc.intel.com

Which would explain the problem with the review...? Interesting, i played with dc ac loadlines back when i had the apex but it seemed to make a difference on custom settings. When you are running default vid already matched vcore

 

[Solid State Brain]

An incorrect DC loadline setting could one reason, but other settings can affect CPU behavior as well.

For example, MSI likes to use fairly high AC loadline (which affects actual load voltages with Adaptive voltages), with the result that default voltages and therefore power consumption tend to be on the high side on motherboards from this manufacturer.

Furthermore, not all CPUs are equal and some may use lower voltages from the factory than others, even if they are supposed to have the same base specifications.

As I mentioned earlier, BIOS bugs or quirks may also occasionally lead to strange frequency behavior, in particular when setting the power limit to a level lower than base power (125W for the 12900K).

All of these factors and possibly more may lead to inconsistent results. In principle it would be just a matter of setting the power limit and comparing results with other CPUs, but in practice it is more complicated than this and one has to be careful checking out motherboards settings and understanding what they do, as well as monitoring CPU behavior throroughly. I don't expect busy hardware reviewers to do this properly. For instance, have you ever seen any of them mentioning AC/DC loadline values in these tests? There's no true "Intel default" for these, just a maximum specification and motherboards will generally use some value below that.

 

[JustBenching]

An incorrect DC loadline setting could one reason, but other settings can affect CPU behavior as well.

For example, MSI likes to use fairly high AC loadline (which affects actual load voltages with Adaptive voltages), with the result that default voltages and therefore power consumption tend to be on the high side on motherboards from this manufacturer.

Furthermore, not all CPUs are equal and some may use lower voltages from the factory than others, even if they are supposed to have the same base specifications.

As I mentioned earlier, BIOS bugs or quirks may also occasionally lead to strange frequency behavior, in particular when setting the power limit to a level lower than base power (125W for the 12900K).

All of these factors and possibly more may lead to inconsistent results. In principle it would be just a matter of setting the power limit and comparing results with other CPUs, but in practice it is more complicated than this and one has to be careful checking out motherboards settings and understanding what they do, as well as monitoring CPU behavior throroughly. I don't expect busy hardware reviewers to do this properly. For instance, have you ever seen any of them mentioning AC/DC loadline values in these tests? There's no true "Intel default" for these, just a maximum specification and motherboards will generally use some value below that.

Well I don't think the discrepancy is due to cpu variance, we are talking about a 70% difference which is nuts. Btw, since you have a 12700k, you can test it yourself. You will get way higher numbers at those wattages even when you are missing 4 ecores.

 

[Solid State Brain]

I don't normally use Windows but I have it on a different SSD. So I had to reboot into it.
My CPU is moderately undervolted below its built-in (i.e. "fused") voltage-frequency curve.

Note how in my case under load VID = Vcore and thus power measurements should be about correct, since I correctly configured the DC Loadline parameter in BIOS.

12700K 75W (16568 pts):




12700K 125W (21501 pts):

 

[JustBenching]

I don't normally use Windows but I have it on a different SSD. So I had to reboot into it.
My CPU is moderately undervolted below its built-in (i.e. "fused") voltage-frequency curve.

Note how in my case under load VID = Vcore and thus power measurements should be about correct, since I correctly configured the DC Loadline parameter in BIOS.

12700K 75W (16568 pts):

View attachment 252018


12700K 125W (21501 pts):

View attachment 252019

So, you are proving my point. There is something gone horribly wrong with the review....

 

[Solid State Brain]

I still maintain that a combination of CPU quality and BIOS settings (or early BIOS) may have caused poor results with lower power limits. Though, I just tried to intentionally realistically mess up default BIOS settings in my case and the worst I could get with my 12700K with CB R23 at 125W was about 18500 points.

A previous 11th gen Intel motherboard I owned also had performance issues when setting PL1 to the same value as PL2 (probably due to bugs), so something along these lines could have possibly occurred as well in this review when limiting power. I think I already mentioned this much earlier on in this thread.

 

[JustBenching]

I still maintain that a combination of CPU quality and BIOS settings (or early BIOS) may have caused poor results with lower power limits. Though, I just tried to intentionally realistically mess up default BIOS settings in my case and the worst I could get with my 12700K with CB R23 at 125W was about 18500 points.

A previous 11th gen Intel motherboard I owned also had performance issues when setting PL1 to the same value as PL2 (probably due to bugs), so something along these lines could have possibly occurred as well in this review when limiting power. I think I already mentioned this much earlier on in this thread.

Yeah, I know in 11th gen (had a 11600k for a while) setting PL1 = PL2 caused all kinds of weird stuff.

Anyways, thanks for willing to test - and validate my results. The doubting crowd has gone silent though, good good

 

[Solid State Brain]

For what it's worth, the issue might have been as simple as mistakenly setting the CPU to fixed voltage, preventing it from getting automatically downvolted when power-throttled. I tried configuring my i7-12700K CPU to fixed 1.35V and repeating the CB R23 125W test and I could only get barely above 10000 points, so that's clearly an easy way for affecting CPU power to a large extent.

If voltage was fixed, power consumption would decrease roughly linearly with clock speed, and scores would decrease accordingly (linearly) on benchmarks that are directly influenced by it. This seems very close to be the case for Cinebench R23 Multithread scores.

Factor (Relative Power)​	Power (W)​	CB23 Multi Score (Calculated)​	CB23 Multi Score (TPU)​
1.00​	125​	18138​	18138​
0.80​	100​	14510​	14819​
0.60​	75​	10883​	11325​
0.40​	50​	7255​	8872​

At lower power levels calculated results appear to diverge slightly, but otherwise they seem overall similar to those calculated by decreasing 125W scores by a fixed factor proportional to the relative power.

 

[kraiggers]

Very enlightening.
Honestly, the 12900K'd have been a much better product at 190W - losing 2% of performance for significantly more accessible cooling/VRM requirement's a great deal tbh.

Isn't PL1=125, PL2=190 equivalent to a i7 12700k? (Or maybe its 190, 190?)

More specifically, I recently got a HP z2 mini g9 maxed with the i9 12900k. I wish I had seen the information about thermals and performance beforehand — I foolishly selected the i9 when the i7 would have been a *much* better choice for the enclosure.

To that end, what's the best way to 'dial down' the i9 to give a thermal envelope more (or exactly) like the i7 in the mini g9? My understanding is the mini g9 is limited to a 125W base? I presume that would be PL1=125, PL2=?? 190? Less? Both to 125? Easier to just turn off Turbo?

Thanks for any thoughts...

 

[Solid State Brain]

Set PL1 to either 125W or the maximum power your system can sustain.

You can leave PL2 at 241W and set a lower temperature limit (e.g. 85 °C) if your motherboard allows it, otherwise you could set it to 190W like the 12700K. Alternatively, set a lower Tau (28s or even something in the order of 12-14s or less) so that power can still peak to the default limit for brief burst loads, but then decreased to the base before temperatures become an issue.

On my 12700K (overclocked to 5.2/3.9 GHz) I currently have PL1=125W, PL2=190W, IccMax=240A (so far like the default configuration), but also Tau=128s, TjMax=85°C. It is set so that the CPU thermally-throttles early, with 85 °C as a target point, while still allowing the base 125W PL1 to engage for prolonged rendering workloads (typically after 3-4 minutes; this never happens with games and regular usage).

 

[kraiggers]

Set PL1 to either 125W or the maximum power your system can sustain.

You can leave PL2 at 241W and set a lower temperature limit (e.g. 85 °C) if your motherboard allows it, otherwise you could set it to 190W like the 12700K. Alternatively, set a lower Tau (28s or even something in the order of 12-14s or less) so that power can still peak to the default limit for brief burst loads, but then decreased to the base before temperatures become an issue.

On my 12700K (overclocked to 5.2/3.9 GHz) I currently have PL1=125W, PL2=190W, IccMax=240A (so far like the default configuration), but also Tau=128s, TjMax=85°C.

Hey thanks. It is an OEM HP, so I don't think I have the granularity of control I would like. I hope I can use Intel Extreme Tuning Utility to adjust turbo boost power settings?

For all I know, the PL1 in this OEM machine might be LESS than 125w. Maybe 65w?

I guess I would set the short power max to 190w (like the i7) and the (long) power max to ?? I haven't found any specific information for this system as it has only been available at retail for a few months… And I don't think it really appeals to people who care about this level of operational detail. I mean, I got suckered into the i9 when it is clearly a bad choice for this box. Dumb.

 

[Solid State Brain]

Hey thanks. It is an OEM HP, so I don't think I have the granularity of control I would like. I hope I can use Intel Extreme Tuning Utility to adjust turbo boost power settings?

Probably. TPU Throttlestop might also work for configuring the power limits.

For all I know, the PL1 in this OEM machine might be LESS than 125w. Maybe 65w?

If you leave everything at default, on Windows you can check out in detail what are your settings with HWInfo. Here is a screenshot from the Internet:

I guess I would set the short power max to 190w (like the i7) and the (long) power max to ?? I haven't found any specific information for this system as it has only been available at retail for a few months… And I don't think it really appeals to people who care about this level of operational detail. I mean, I got suckered into the i9 when it is clearly a bad choice for this box. Dumb.

If your system was designed to thermally support the i7-12700K, it should be fine to leave PL1 (long duration limit) at 125W.


EDIT:

For all I know, the PL1 in this OEM machine might be LESS than 125w. Maybe 65w?

It looks like the TDP (PL1) is configured to 90W. PL2 might have been already configured accordingly to your system's cooling capabilities.

https://h20195.www2.hp.com/v2/GetDocument.aspx?docname=c08109685

 

[kraiggers]

It looks like the TDP (PL1) is configured to 90W. PL2 might have been already configured accordingly to your system's cooling capabilities.

https://h20195.www2.hp.com/v2/GetDocument.aspx?docname=c08109685

Thats fascinating. But I'm not sure I understand what that footnote means... PL1=90 PL2=125? And whatever it does mean, it seems the same for the i5, i7, and the i9?

But the reviews of this mini g9 all seem to indicate the i9 gets HOT and LOUD (and also throttles) in ways the i7 doesn't. Or at least doesn't as much. Which makes me think there must be some kind of difference in PL2 between the default i7 and i9 settings *in this actual machine*? But if so I can't find the detail info...

I guess one thing I've learned reading up on these processors is that disabling Turbo Boost entirely is an awfully blunt instrument in terms of just trying to control the thermals.

A more nuanced solution that *should* offer the best combination of thermal control (and therefore quiet running) with still optimal performance is to find the best combination of PL1, PL2, and Tau for this particular enclosure… Which HP has clearly *not* done considering the noise levels mentioned in every review with the i9 at stock settings.

 

[Solid State Brain]

Thats fascinating. But I'm not sure I understand what that footnote means... PL1=90 PL2=125? And whatever it does mean, it seems the same for the i5, i7, and the i9?

It likely means it supports 12th gen Intel processors with a nominal TDP of 125W, but they will be used with a scaled down TDP (PL1) of 90W, and this will be the same for the i5, i7, i9.

PL2 (peak power, short-term power) is not specified here. You can check it out with HWInfo as instructed above. It might be set to a relatively high value so that the CPU will stay at TjMax (normally 100 °C) for a period while power slowly decreases until PL1 is engaged.

"Throttling" is not necessarily a bad word. All modern processors throttle as part of their normal operation for a reason or another (power limits, current limits, thermal limits, etc). You can can set lower power (PL2) or temperature limits so that the CPU will throttle earlier (i.e. more) and will not heat up as much, causing less noise.

 

[kraiggers]

It likely means it supports 12th gen Intel processors with a nominal TDP of 125W, but they will be used with a scaled down TDP (PL1) of 90W, and this will be the same for the i5, i7, i9.

PL2 (peak power, short-term power) is not specified here. You can check it out with HWInfo as instructed above. It might be set to a relatively high value so that the CPU will stay at TjMax (normally 100 °C) for a relatively short period while power slowly decreases until PL1 is engaged.

"Throttling" is not necessarily a bad word. All modern processors throttle as part of their normal operation for a reason or another (power limits, current limits, thermal limits, etc).

Yeah, I have no qualms whatsoever about throttling — everything is a trade off. I don't want to hear the fan blowing hard ever. That's my trade-off priority.

So if PL1=90, then why does HP even put 125 in the table? That makes… no sense? And I'd wager that PL2 is higher (maybe a lot higher) for i9 than for i7, which is the only thing I can think of that would explain the differences observed.

If HWInfo shows me that PL2 is (say) 250 or whatever, then I can try setting it to 190 which would presumably match the i7, or even lower just to keep things cooler/quieter. Or really just bring it down to whatever I need it to be in practice.

Do OEM (eg HP) machines typically allow the use of Intel ETU? I saw this in a Throttlestop guide:

TPL is the Turbo Power Limit module, which is mostly useful for enabling Speed Shift on supported notebooks that don’t have it enabled in a BIOS update (i.e. the XPS 9550 and 9560). On some machines, some users have claimed to be able to set PL1 and PL2 limits through this module, though I personally have not been able to do so.

 

[Solid State Brain]

Yeah, I have no qualms whatsoever about throttling — everything is a trade off. I don't want to hear the fan blowing hard ever. That's my trade-off priority.

So if PL1=90, then why does HP even put 125 in the table? That makes… no sense? And I'd wager that PL2 is higher (maybe a lot higher) for i9 than for i7, which is the only thing I can think of that would explain the differences observed.

If HWInfo shows me that PL2 is (say) 250 or whatever, then I can try setting it to 190 which would presumably match the i7, or even lower just to keep things cooler/quieter. Or really just bring it down to whatever I need it to be in practice.

My guess is that they want to make it clear that it supports ordinary 125W desktop CPU parts. OEMs are allowed by Intel to use them at lower TDP if required (e.g. due to cooling limitations).

Yes, probably PL2 is higher than with the i7 to get better overall performance and better justify the price difference, although to get a comfortable noise level you will likely need to set it considerably lower than 190W.

Do OEM (eg HP) machines typically allow the use of Intel ETU? I saw this in a Throttlestop guide:

I'm not sure. It can be that only some functions are enabled.

 

[kraiggers]

My guess is that they want to make it clear that it supports ordinary 125W desktop CPU parts. OEMs are allowed by Intel to use them at lower TDP if required (e.g. due to cooling limitations).

Yes, probably PL2 is higher than with the i7 to get better overall performance and better justify the price difference, although to get a comfortable noise level you will likely need to set it considerably lower than 190W.

I was re-watching this review, and I noticed they show the Intel XTU briefly at 8:02. It has the PL1 at 90 — per the table you found earlier — and the PL2 for the i9 is at 241w. It looks like you can't fiddle with the voltage or fan curve, but the PL1 and PL2 appear editable…?

So, in theory, I could set that to 190w and presumably match the i7 thermal envelope in this machine. Or I could go even lower and really keep things cool, if necessary based on usage. (And I could lower the Tau value from 128 sec to something much less. But I have no sense of how that impacts heat.)

Which brings us back around to the original concept of the article — using the i9 at lower power thresholds is very doable, and can save a ton of heat for modest performance reduction. I mean, HP should never have put the i9 into this micro enclosure, and certainly not with PL2 at 241w.

Aside: The synthetic single core scores on Geekbench specific to this machine (not just the generic i9 and i7 chips) show an averaged 8% slower score for the i7. 1992 for the i9 and 1823 for the i7. Lowering PL2 to 190 (same as the i7) would bring the performance of the i9 down some, but how much? Does the lower base clock speed matter at all? I suppose it doesn't matter — I will set PL2 and Tau to whatever is necessary to keep things quiet. But it is pretty interesting stuff….

The price premium for i9 is a rip-off unless you have a seriously strong cooling system and/or don't mind a hurricane fan blower. Which very few OEM machines — and certainly not this z2 mini g9 — have.

 

[bug]

I was re-watching this review, and I noticed they show the Intel XTU briefly at 8:02. It has the PL1 at 90 — per the table you found earlier — and the PL2 for the i9 is at 241w. It looks like you can't fiddle with the voltage or fan curve, but the PL1 and PL2 appear editable…?

So, in theory, I could set that to 190w and presumably match the i7 thermal envelope in this machine. Or I could go even lower and really keep things cool, if necessary based on usage. (And I could lower the Tau value from 128 sec to something much less. But I have no sense of how that impacts heat.)

That is what I did to my 12600k, so absolutely possible.

I think I have it at 75/100W (but I'm not sure, it's been a while since I set it), with no visible performance penalty and a Freezer i34 that is rarely audible past the boot sequence.

 

[Solid State Brain]

So, in theory, I could set that to 190w and presumably match the i7 thermal envelope in this machine. Or I could go even lower and really keep things cool, if necessary based on usage. (And I could lower the Tau value from 128 sec to something much less. But I have no sense of how that impacts heat.)

Tau determines how much time the CPU tries to stay at PL2 before PL1 is engaged. It's a time constant, not an exact duration, so depending on operating conditions the effective time can be lower or higher than configured.

A strategy could be setting Tau to just a few seconds (as low as 2-4 seconds if needed) so that PL2 can get as high as it can for a brief period before fan speeds start to ramp up; this would be basically taking advantage of fan control inertia.

Lowering PL2 to 190 (same as the i7) would bring the performance of the i9 down some, but how much? Does the lower base clock speed matter at all? I suppose it doesn't matter — I will set PL2 and Tau to whatever is necessary to keep things quiet. But it is pretty interesting stuff….

CPU performance should decrease very much non-linearly with power, and mostly in multithreaded loads, so in practice you might not notice too much difference unless you're only doing rendering-type workloads. A while back I tried doing a test in this regard with my 12700K configured to 5.0 GHz, decreasing power limits. The 12900K will likely have a similar trend:

The base clock is the minimum guaranteed clock that the CPU is supposed to maintain at base power (usually 125W for 125W TDP CPUs) at 100 °C in a worst-case workload without AVX instructions and worst-case silicon quality. At 90W the base clock would be lower, but how much it's not clear.

 

[bug]

Tau determines how much time the CPU tries to stay at PL2 before PL1 is engaged. It's a time constant, not an exact duration, so depending on operating conditions the effective time can be lower or higher than configured.

It will stay lower than tau in PL2 if you don't have sufficient cooling, but it's not supposed to stay longer than tau under any circumstances.

A strategy could be setting Tau to just a few seconds (as low as 2-4 seconds if needed) so that PL2 can get as high as it can for a brief period before fan speeds start to ramp up; this would be basically taking advantage of fan control inertia.

That's a strategy, all right. A strategy to fry your CPU, but a strategy nonetheless.

Best explanation that I know of: https://www.anandtech.com/show/13544/why-intel-processors-draw-more-power-than-expected-tdp-turbo

 

[Solid State Brain]

It will stay lower than tau in PL2 if you don't have sufficient cooling, but it's not supposed to stay longer than tau under any circumstances.

This is wrong. Tau is not a fixed time, it's used to calculate the so-called Exponentially-Weighted Moving Average (EWMA) used by the CPU for determining when PL1 should be engaged. As an example, if you set Tau to 8 seconds, PL1 is 125W and PL2 is 150W, the CPU will maintain PL2 for about 14 seconds. This duration largely depends on the difference between current package power (which may or may not be fixed at PL2) and PL1.

That's a strategy, all right. A strategy to fry your CPU, but a strategy nonetheless.

Wrong. With the PL2 at 241W and Tau=128s as reported earlier by kraiggers, the CPU is already going as high as it can. TjMax (by default 100 °C) is the limiting factor here. CPU power will be progressively decreased (that is, throttled) to maintain the temperature fixed at the temperature limit, and as it does so, fan speed ramps up.

By decreasing Tau to just a few seconds and keeping PL2 the same, the CPU will engage PL1 much earlier, ideally before fan speeds start accelerating. This would yield a lower thermal and electrical load than the default configuration.

 

[kraiggers]

This is wrong. Tau is not a fixed time, it's used to calculate the so-called Exponentially-Weighted Moving Average (EWMA) used by the CPU for determining when PL1 should be engaged. As an example, if you set Tau to 8 seconds, PL1 is 125W and PL2 is 150W, the CPU will maintain PL2 for about 14 seconds. This duration largely depends on the difference between current package power (which may or may not be fixed at PL2) and PL1.

Wrong. With the PL2 at 241W and Tau=128s as reported earlier by kraiggers, the CPU is already going as high as it can. TjMax (by default 100 °C) is the limiting factor here. CPU power will be progressively decreased (that is, throttled) to maintain the temperature fixed at the temperature limit, and as it does so, fan speed ramps up.

By decreasing Tau to just a few seconds and keeping PL2 the same, the CPU will engage PL1 much earlier, ideally before fan speeds start accelerating. This would yield a lower thermal and electrical load than the default configuration.

Let me see if I can clarify how this applies to my situation...

According to what I've seen online, I *think* the default settings for this HP z2 mini g9 are PL1=90w PL2= 241w and Tau=128s for the i9k, and same except PL2=190w for i7k.

Reviews have ALL mentioned this machine getting notably hot and also therefore loud under load with the i9 — all the review units have the i9, apparently — and also throttling both power and thermally. This is with the OEM setup already starting with PL1=90w for both i7 and i9, even though PL1 doesn't matter as much under a full testing load. One would assume the cooling in this tiny chassis is insufficient for these chips. Which isn't very surprising — its tiny!

Assuming I am able to set PL2 and Tau with the XTU, the optimal settings for me — preferencing quiet operation over all-out performance — will depend on the workloads I'm doing (cad and adobe, mostly):

If my work is bursty but not so much sustained, I could in theory keep the PL2 somewhat higher (eg 190w like i7k) but control the fan by lowering the Tau and therefore keeping the temps from going too high by throttling down the cpu before things get too hot and the fan goes gonzo.

If my work is more sustained (renders or such line), I would probably need to lower PL2 more aggressively (eg, 125w or whatever low value) while keeping Tau higher to allow the best sustained performance while also holding down the temps to something the fans can handle at acceptable speeds (ie not 7000 rpm!) without getting too loud.

The sweet spot may in fact be some combination of both lowering PL2 more than 190w (like, say to 150w) and ALSO lowering Tau to some value less than 128s — maybe 30s or something. Who knows, I'm just wild-ass guessing until I have the system in hand to experiment. Start high and adjust incrementally; I may not need to do much at all (besides lowering PL2 to 190w) depending on my usage...

Is that a reasonable and accurate description of the situation?

Tangential question — is it worth keeping this order (i9k, 64gb ecc ram, 1 tb ssd, 2nd ssd slot open, rtx a2000) at $1495 incl tax? That seems like a freakishly good price, even for a thermally crippled machine? I LIKE the tiny footprint, I'm not gaming, and I'm not doing hardcore video or rendering, generally.

 

[Solid State Brain]

[...] Is that a reasonable and accurate description of the situation?

In general, yes. If can use a lower PL2, peak CPU core temperatures should be lower (or at least not reach the limit as quickly) and you should in turn be able to use a longer Tau without the cooling fan spinning rapidly too early. However, this depends also on how the fan is controlled on your system.

 

[kraiggers]

In general, yes. If can use a lower PL2, peak CPU core temperatures should be lower and you should in turn be able to use a longer Tau without the fans spinning rapidly too early. However, this depends also on how fans are controlled on your system.

E.g. is there actually a fan curve, or does the firmware just slam the fans to 100% once the temp crosses some threshold? I don't know. None of the reviews actually address this. And my understanding is fan control is not available in the bios nor through XTU. Awesome.

The easiest solution to achieve my goal would be if I could set a *max* fan speed in the bios and let the system dynamically adjust the performance. Ha ha! Why would anyone care about THAT?!

 

[Solid State Brain]

As far as I am aware of, the Intel-suggested fan control behavior is to ramp up fan speed linearly from 80 °C to 100 °C, perhaps it's what happens in your case.
If so, keeping temperatures below 80°C or not much above that should help.

Check out here: https://edc.intel.com/content/www/u...1-of-2/009/fan-speed-control-scheme-with-dts/