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

[Solid State Brain]

You wouldn't put too much import on that one anecdotal sample? (And, also, the frame rate in the games test was pretty similar, FWIW.)

No, especially since this is from LTT, which is more of an entertainment show compared to other techtubers.

Faster because i9k has more cores and larger cache? Does the lower base clock (3.2ghz vs 3.6ghz) mean the i9k suffers compared to i7k for single core performance, holding power draw constant?
The $100 delta is not a deal-breaker. I feel equally great about the price, wither its $1400 for the i9k system, or $1300 for the i7k system.

The i9-12900K should be faster due to slightly higher frequencies (5.2 vs 5.0 GHz on single core, 4.9 vs 4.7 GHz all-core), larger number of cores (mostly important in MT loads), and better binning (lower voltages at the same frequencies). Keeping power draw the same, it should be faster all-around than a typical i7-12700K.

The base clock is just a guaranteed minimum frequency that the CPU should be able to attain in all-core MT loads without AVX instructions at a specific "base" power (125W). It does not affect single-core performance. The more cores, the lower the base clock will be for the same base power.

What kind of BIOS-related issues might crop up?
Is it in any way meaningful that none of the 3rd party sales channels (eg B&H etc) sell an i9k config? They all top out at i7k...

BIOS-related issues could be for example aggressive voltages by the motherboard when it detects an i9, which could make it throttle earlier.

I don't know what to make of the default CPU selection for this HP. Perhaps vendors expect that most users won't need the best CPU available in a small form factor system.

 

[kraiggers]

Keeping power draw the same, it should be faster all-around than a typical i7-12700K.

So to be clear, there is no reason (other than cost) not to get the i9, as long as I’m willing to dial down the PL2 to ameliorate heat+noise, which appears to be possible thanks to XTU.

And the upside is marginally better performance — at constant power — than the i7k or i5k.

 

[Mussels]

You're missing that 90W is going to get you higher clock speeds out of a lesser amount of cores
A 6 core at 90W is going to run full performance, while an i9 is going to throttle extremely hard


If you already have the i9, the best you can do is improve the cooling somehow - you cant lower voltages on a locked BIOS, so you're stuck with it how it is

 

[kraiggers]

You're missing that 90W is going to get you higher clock speeds out of a lesser amount of cores
A 6 core at 90W is going to run full performance, while an i9 is going to throttle extremely hard

If you already have the i9, the best you can do is improve the cooling somehow - you cant lower voltages on a locked BIOS, so you're stuck with it how it is

Are you saying that an i5k is faster than an i7k which is faster than an i9k below some low power level threshold (eg 100w)? Especially for multi core loads? That's somewhat counter-intuitive, but not inconceivable, I guess? If that's true, then why does anyone ever buy the high-spec chips for SFF or laptops?

Let me see if I understand you with a hypothetical example on our throttled 90w PL1 machine (PL2 wouldn't matter for a load test) —
In theory, a 6 P-core i5k at 90w would probably run single core at the full 4.9ghz, and all six cores at maybe 3.2ghz.
An 8 P-core i9k could maybe run a single core at full 5.2ghz but all eight cores at maybe 2.4ghz

So in that type of scenario, the i9k might be marginally faster at single-core tasks, but potentially slower at multi-core tasks, depending on the exact number of cores utilized? The numbers are made up but is that the concept you are describing? I guess the question how much does more cores (in the i9k) compensate for a lowered average frequency when many (or all) of them are being utilized?

[SIZE=4]Solid State Brain[/SIZE] suggested the best option was getting an i9k and manually throttling PL2 to control max heat. That sounds to me, within a thermally and power throttled SFF, like preferencing single core performance over multi core. Do you disagree? If so, can you elaborate with a bit more specificity?​

 

[Mussels]

Are you saying that an i5k is faster than an i7k which is faster than an i9k below some low power level threshold (eg 100w)? Especially for multi core loads? That's somewhat counter-intuitive, but not inconceivable, I guess? If that's true, then why does anyone ever buy the high-spec chips for SFF or laptops?

Let me see if I understand you with a hypothetical example on our throttled 90w PL1 machine (PL2 wouldn't matter for a load test) —
In theory, a 6 P-core i5k at 90w would probably run single core at the full 4.9ghz, and all six cores at maybe 3.2ghz.
An 8 P-core i9k could maybe run a single core at full 5.2ghz but all eight cores at maybe 2.4ghz

So in that type of scenario, the i9k might be marginally faster at single-core tasks, but potentially slower at multi-core tasks, depending on the exact number of cores utilized? The numbers are made up but is that the concept you are describing? I guess the question how much does more cores (in the i9k) compensate for a lowered average frequency when many (or all) of them are being utilized?

[SIZE=4]Solid State Brain[/SIZE] suggested the best option was getting an i9k and manually throttling PL2 to control max heat. That sounds to me, within a thermally and power throttled SFF, like preferencing single core performance over multi core. Do you disagree? If so, can you elaborate with a bit more specificity?​

If you're splitting 100W between four cores, they get 25W each
between 8, they get 12.5W

It's basic math
There is no preferencing of ST over MT or anything like that - simply that the more cores are active the slower they will all be



This is the core of my entire problem with the current intel lineups (and the top AM5 CPUs) is that with such high power requirements, the moment you dont have the power available or the cooling, performance goes to shit.

 

[kraiggers]

Mussels said:
This is the core of my entire problem with the current intel lineups (and the top AM5 CPUs) is that with such high power requirements, the moment you dont have the power available or the cooling, performance goes to shit.

If the math was that simple, then every one of these chips would bench the same single core performance at the same power level, wouldn't it?

According to the article intro, PL1 doesn't matter anymore? Then the closest thing I could find to this scenario is also in the original article. The PL1=PL2=190 version of the i9k is slightly to meaningfully faster than a i7 (at stock PL2=190) in almost every test that was run, across single- and multi-core apps. There were a *small* number where the i7 was minimally faster or a push.

Now, maybe a lower power threshold (eg PL1=PL2=90) for *both* chips changes that outcome? We don’t really know.

But the (admittedly very minimal) evidence I have found suggests the i9 would be as fast or faster than the i7 at the same power levels. Which is consistent with Solid State Brain‘s recommendation.

Unless I’m missing some tests or data that suggest otherwise?

Hey, I just want to get the best chip for my situation. Fast and quiet would be my preference! No review I've seen says, Hey guys, the i5 or i7 is FASTER than the i9 in your baller laptop or SFF. They sometimes say, hey the i5 or i7 is good enough and makes less heat or noise in a tough situation. But in this SFF, all three would be power throttled, maybe to the same number, or maybe not. I suppose there is a possibility that holding quiet operation as the constant, the i5 could comfortably run at say 150w (less cores) and the i7 could do 125w and the i9 at 100w. Those are just speculative numbers.

But, then what was even the point of this article (what happens when you power throttle the i9) if the better option is to just use an i7 or i5?

 

[Super Firm Tofu]

If you're splitting 100W between four cores, they get 25W each
between 8, they get 12.5W

It's basic math
There is no preferencing of ST over MT or anything like that - simply that the more cores are active the slower they will all be



This is the core of my entire problem with the current intel lineups (and the top AM5 CPUs) is that with such high power requirements, the moment you dont have the power available or the cooling, performance goes to shit.

That's not how it works. If that was the case, the 5950x would perform worse than the 5800x. They both have the exact same power budget. The 5950x is almost twice as fast as the 5800x in multi-thread workloads using the same power. Yes, maximum clock speed under all core load is lower, but it still does more work overall.

Also, power limiting (to a point) doesn't reduce single, or low thread performance as those loads don't approach the maximum power.

In the example above, it's better to have more cores with less power (per core) rather than fewer cores with more power (per core).

The entire premise of this article is that power use and performance do not scale linearly.

 

[Mussels]

That's not how it works. If that was the case, the 5950x would perform worse than the 5800x. They both have the exact same power budget. The 5950x is almost twice as fast as the 5800x in multi-thread workloads using the same power. Yes, maximum clock speed under all core load is lower, but it still does more work overall.

Also, power limiting (to a point) doesn't reduce single, or low thread performance as those loads don't approach the maximum power.

In the example above, it's better to have more cores with less power (per core) rather than fewer cores with more power (per core).

The entire premise of this article is that power use and performance do not scale linearly.

It is entirely how it works


If you throttle the system wattage down like this entire post is talking about

Go put a 5800x and 5950x down to 65W and compare the clock speeds in all core loads or AVX workloads.

 

[InVasMani]

You can optimize frequency and VID across individual cores however for stairway scaling across additional cores of frequency which will improve wattage and in way that minimizes the issue you're overstating a bit however. Honestly it depends on the chips in question though for Alder Lake that won't work nearly as well as it will with Raptor Lake from a general standpoint though even in the case of Alder Lake it does work and you can inverse a bit how Intel positions them to operate for marketing purposes.

 

[Solid State Brain]

You're missing that 90W is going to get you higher clock speeds out of a lesser amount of cores
A 6 core at 90W is going to run full performance, while an i9 is going to throttle extremely hard

Clock speeds will indeed be lower in multithreaded workloads with the bigger chip, but MT performance will still be higher, since by running at lower frequencies the cores will run at more efficient levels.

For example, a typical i9-12900K at 125W scores about 24000 points and at 65W 17500 points with Cinebench R23. With my voltage-optimized i7-12700K I get ~22000 pts at 125W and ~16500 pts at 65W. I need to overclock my CPU and make it draw 230W or so to get about the same score the i9 gets at 125W.

AMD Ryzen 9 7950X vs. Intel Core i9-12900K at 125W and 65W | Club386

What happens when you dial-down power limits? You might be surprised.

www.club386.com

My own tests:

 

[kraiggers]

Clock speeds will indeed be lower in multithreaded workloads with the bigger chip, but MT performance will still be higher, since by running at lower frequencies the cores will run at more efficient levels.

I think I understand the non-linear nature of the performance curve. That's why the i9k, even throttled down to lower power budgets, is faster both at single core AND (somewhat surprisingly) at multi-core workloads. Though there aren't a ton of articles making that comparison, sadly.



There are two things I still don't really get that I think are relevant to this conversation...

First, does PL1 have anything to do with these performance results? Is that number meaningful anymore? From the introduction:

Unlike past generations of processors that were constrained by the Tau time value to hold maximum power draw, or PL2, Alder Lake processors now run at maximum power draw indefinitely if the load demands it and as long as the processor doesn't hit the thermal limit of 105°C. This is done without inventing a new system; Intel simply tweaked the PL1 and PL2 values and set them both to 241 W, which effectively means the processor can run at 241 W all the time as long as it doesn't overheat. The "125 W" limit now only exists on paper and in marketing documents.

So in all these tests, unless specified otherwise, are we to assume that PL1=PL2=241w for i9k, PL1=PL2=190 for i7k, and PL1=PL2=150 for i5k? So in the results for this article, the PL1=PL2=125 gets 86% performance, and the i7k gets 90% — but at PL1=PL2=190w — right?

And, how does the performance of the chips change when PL1 is different than PL2? Say, when PL1=90w (like in the oem HP)? Maybe that doesn't matter since all three are set at the same PL1=90?


Second — and I guess related to what I wrote above — what are there thermal differences between the three chips (i5k, i7k, i9k) when run at the same power levels? Or, another way, can the i7k run at a higher power level holding heat constant? (I'm not even sure whether that would be PL1, PL2, or both?)

Can we assume that thermal load is highly correlated to power consumption? I mean, obviously yes, but HOW tightly bound is that correlation…? Let me collate some data and see if there are any inferences to be drawn...

System power levels + chip temp:
i7k (stock, 190w)……...56w idle….87w single…..221w multi…..234w stress…..64C

i9k (stock, 241w)……...56w idle…..91w single…..297w multi…..350w stress…..92C
i9k (throttled, 190w)………...n/a…………...n/a…..241w multi…..278w stress…..?
i9k (throttled, 125w)…..55w idle…………...n/a…..194w multi…..204w stress…..57C
i9k (throttled, 100w)………...n/a………….. n/a…..172w multi…..177w stress…..?
i9k (throttled, 75w)……...…...n/a…………..n/a…..146w multi…..147w stress…..?
i9k (throttled, 65w)…….54w idle…………..n/a…………….n/a…..138w stress…..38C
i9k (throttled, 50w)…………..n/a…………...n/a…..120w multi…..138w stress…..?

Well with only three data points we can clearly determine that the power level to thermal load curve for i9k is not linear. Doubling from 65w -> 125w yields a 20C rise. Doubling again from 125w -> 241w yields a 35C rise.

We know from the full i9k review that the i7k scores 64C under load. (Again, presumably PL1=PL2=190w.) We don't know what temp the i9k gets at that power level, but we know from this club386 review that at 125w (I'm assuming PL1=PL2=125w?) the i9k is at 57C under load.

So I'm left with a few questions:

The throttled 190w i9k system is consuming more power than the stock 190w i7k system, even when the chip is set at the same 190w power level. That's interesting, and honestly, unexpected? Doesn't that undercut the whole idea of power limiting?

Here's the nut — we don't know what the temperatures are for the i9k at 190w or 100w. (Nor for the i7k.) I mean, 190w will be higher than 57C and closer to 92C than 57C. Presumably, higher than the 64C the i7k gets? And I assume the 75w and 100w temps will slot in between the 65w at 38C and the 125w at 57C?

I'm left with the gut feeling that while an i9k might be as fast or faster than an i7k at a constant power level, it will also (inexplicably?!) consume more system power and be hotter. I would like to hold the thermal limit constant, and then see what kind of performance the chips would give. If my inferences are vaguely right, then the i9k would be (slightly?) more throttled than the i7k, and that might erode the performance advantage we see at constant power levels?

And to bring it back around to the little HP SFF, the i7k stock is set to PL1=90 and PL2=190. The stock i9k is set to 90, 241. If I understand what PL1 does, both of those chips would be throttled down significantly under load to 90w. Which would generate-steady state temps well under 60C (I think?). But the i7k would be able to go longer at the PL2 because it is overall a cooler chip than i9k at constant power…?

Might that explain what we see in the LTT video re. i7k outperforming the i9k?


Clear as mud, am I right? I have to admit I'm frustrated by this choice, even though I know the performances differences are almost certainly less than 10% either way, either single or multi, theoretically holding heat/noise constant.

So do I stick with the i9k, throttle it down to (whatever TBD) for low noise, and get the benefit of better single/low thread work but maybe take a hit on higher load multi-thread usage? Or switch to the i7k, slower at single core but cooler, and therefore maybe a touch better performing multi-core at the low noise ceiling???

 

[Solid State Brain]

First, does PL1 have anything to do with these performance results? Is that number meaningful anymore? From the introduction:

{quote}

So in all these tests, unless specified otherwise, are we to assume that PL1=PL2=241w for i9k, PL1=PL2=190 for i7k, and PL1=PL2=150 for i5k? So in the results for this article, the PL1=PL2=125 gets 86% performance, and the i7k gets 90% — but at PL1=PL2=190w — right?

PL1 will affect results when it's used. Unfortunately there has been a lot of confusion in this regard, and the Intel marketing department has been pushing that PL1=PL2=241W is/should be the default for the i9-12900K. But some motherboards may use by default specification values with a 125W PL1 and 56s Tau, while others will use basically no limits. In some cases, reviewers deliberately set power limits to the Intel specification/recommendation, in others they will try to follow marketing suggestions of PL1=PL2.


As for power testing, unfortunately, there are problems with the TechPowerUp results which were recognized earlier in this thread also by other users, but have never been addressed. This is better understood with a diagram, see below.

The i9 from the TPU review had scores decreasing linearly with power. Furthermore, efficiency (score/watt) barely increased with decreasing power limit.
The Club386 results on the other hand, although they provide only a few data points in total, roughly follow the expected power-performance curve.

And, how does the performance of the chips change when PL1 is different than PL2? Say, when PL1=90w (like in the oem HP)? Maybe that doesn't matter since all three are set at the same PL1=90?

Difficult to predict performance when PL2 and PL1 are both used and CPU power can be temperature-limited. The CPU will run for a while at PL2 (or the highest it can up to PL2, depending on operating conditions), then continue processing at PL1. Most online reviews try instead to run the CPU at a constant power level.

Second — and I guess related to what I wrote above — what are there thermal differences between the three chips (i5k, i7k, i9k) when run at the same power levels? Or, another way, can the i7k run at a higher power level holding heat constant? (I'm not even sure whether that would be PL1, PL2, or both?)

Assuming the same cooling efficiency and architecture, the chip with the larger amount of cores for the same amount of power should run at lower temperatures, since heat density will be lower (distributed over a larger surface area).

The throttled 190w i9k system is consuming more power than the stock 190w i7k system, even when the chip is set at the same 190w power level. That's interesting, and honestly, unexpected? Doesn't that undercut the whole idea of power limiting?

The stock i7-12700K with default PL2=190W doesn't actually reach 190W (package power) during rendering benchmarks, but more like 165-175W depending on voltage settings and silicon quality. This is probably why the i9-12900K limited to 190W has a higher system consumption.

Also, most stock i9-12900K CPUs probably won't actually reach 241W during a Cinebench R23 run either, but stop at a somewhat lower level.

I'm left with the gut feeling that while an i9k might be as fast or faster than an i7k at a constant power level, it will also (inexplicably?!) consume more system power and be hotter. I would like to hold the thermal limit constant, and then see what kind of performance the chips would give. If my inferences are vaguely right, then the i9k would be (slightly?) more throttled than the i7k, and that might erode the performance advantage we see at constant power levels?

I don't see good reasons why a somewhat larger CPU (just 4 more E-cores) of the same architecture should be hotter at the same power draw, if all other variables (cooler, motherboard) are kept the same.

From my own testing, throttling (of any kind) apparently has a very slight effect on CPU efficiency in the order 1% or so compared to the unthrottled CPU running at the same power level, it's basically within error margins.

The only way throttling can seriously affect CPU efficiency and performance is if temperatures cannot be decreased anymore by lowering voltages and frequencies. At that point clock modulation is introduced and the CPU runs at a 25% duty cycle. However, I haven't managed to see that happening on my own system, not even after disabling the CPU fan and letting it run a rendering test continuously while getting severely temperature-throttled. I guess it will require frequency to first drop to the minimum level.

And to bring it back around to the little HP SFF, the i7k stock is set to PL1=90 and PL2=190. The stock i9k is set to 90, 241. If I understand what PL1 does, both of those chips would be throttled down significantly under load to 90w. Which would generate-steady state temps well under 60C (I think?). But the i7k would be able to go longer at the PL2 because it is overall a cooler chip than i9k at constant power…?

The i7k isn't and shouldn't be an inherently cooler chip at the same power.

Might that explain what we see in the LTT video re. i7k outperforming the i9k?

I doubt it.

 

[kraiggers]

Assuming the same cooling efficiency and architecture, the chip with the larger amount of cores for the same amount of power should run at lower temperatures, since heat density will be lower (distributed over a larger surface area).

I don't see good reasons why a somewhat larger CPU (just 4 more E-cores) of the same architecture should be hotter at the same power draw, if all other variables (cooler, motherboard) are kept the same.

The i7k isn't and shouldn't be an inherently cooler chip at the same power.

First, thanks for your time and consideration! I really appreciate your thoughts on this topic.

My limited understanding is that 'binning' slots processors into performance buckets. In other words, a chip that can perform at higher frequencies and power (and also hotter) reliably is binned to a higher bucket than lesser chips. That suggests to me that the corollary would also be true — chips that can run 'higher' would also run cooler or more efficiently at lower power / frequency levels? I don't know if this is correct, but it certainly 'feels' consistent with your third statement above.

Anyway, notwithstanding the above, the only data points I have between the i9k and i7k strongly suggest the i9k would be hotter at the same 190w power level. From my sad little table:

System power levels + chip temp:
i7k (stock, 190w)……...56w idle….87w single…..221w multi…..234w stress…..64C

i9k (stock, 241w)……...56w idle…..91w single…..297w multi…..350w stress…..92C
i9k (throttled, 190w)………...n/a…………...n/a…..241w multi…..278w stress…..?
i9k (throttled, 125w)…..55w idle…………...n/a…..194w multi…..204w stress…..57C

Are you saying this data is sufficiently 'rough' as to be unreliable or non-indicative?


And back to my fundamental question:

So do I stick with the i9k, throttle it down to (whatever TBD) for low noise, and get the benefit of better single/low thread work but maybe take a hit on higher load multi-thread usage? Or switch to the i7k, slower at single core but cooler, and therefore maybe a touch better performing multi-core at the low noise ceiling???

It sounds like you would say my premise is flawed. You're saying, pardon my paraphrasing, that the i9k, throttled down to (whatever TBD) for low noise, would be faster single/low thread work and still should be as fast or faster at higher load multi-thread usage. Whereas the i7k would be slower at single core, and no better or slightly worse performing multi-core, at BOTH the same low power AND at a low(er) thermal & noise ceiling?

And therefore sticking with the i9k is the correct choice even when power limited PL1=90w PL2=TBD1 specifically to reduce heat and noise, compared to the i7k (which would also be power limited to PL1=90 PL2=TBD2, where TBD1 may or may not match TBD2 — but they should match, or even TBD1 might be higher than TBD2 because i9k should run cooler at constant power, data point above notwithstanding)?

 

[Solid State Brain]

My limited understanding is that 'binning' slots processors into performance buckets. In other words, a chip that can perform at higher frequencies and power (and also hotter) reliably is binned to a higher bucket than lesser chips. That suggests to me that the corollary would also be true — chips that can run 'higher' would also run cooler or more efficiently at lower power / frequency levels? I don't know if this is correct, but it certainly 'feels' consistent with your third statement above.

Anyway, notwithstanding the above, the only data points I have between the i9k and i7k strongly suggest the i9k would be hotter at the same 190w power level. From my sad little table:

Are you saying this data is sufficiently 'rough' as to be unreliable or non-indicative?

Typical i9-12900K chips need voltages in the 1.25-1.35V range for 5.2 GHz, while typical i7-12700K need about the same voltages for 5.0 GHz. So, for the same frequencies and number of active cores the i9 will run at lower voltages and thus cooler.

Note that for 'need' I mean as reported by the internally programmed voltage-frequency curve in the CPU (which depends on silicon quality). Motherboards or end-users may apply more or less voltage than that. Some motherboards (Notably ASUS ROG ones) directly show these CPU-fused voltages in BIOS, but there are methods for retrieving them on those which don't.

It sounds like you would say my premise is flawed. You're saying, pardon my paraphrasing, that the i9k, throttled down to (whatever TBD) for low noise, would be faster single/low thread work and still should be as fast or faster at higher load multi-thread usage. Whereas the i7k would be slower at single core, and no better or slightly worse performing multi-core, at BOTH the same low power AND at a low(er) thermal & noise ceiling?

At the same power the i9 should be faster in single/low threads, and it should be also faster in MT usage than the i7.

Additionally, given the same thread usage and no power throttling, the i9 should run at the same temperatures or somewhat cooler (due to generally better silicon quality). In all-core MT loads at the same power the i9 should run cooler (due to lower power density across cores).

The above should be true unless:

• Your specific i9 has IHS planarity issues ("Contact Frames" have been devised to correct this);
• Your BIOS has issues with the i9;
• The i7 would have had good silicon quality, while the i9 has bad silicon quality.

And therefore sticking with the i9k is the correct choice even when power limited PL1=90w PL2=TBD1 specifically to reduce heat and noise, compared to the i7k (which would also be power limited to PL1=90 PL2=TBD2, where TBD1 may or may not match TBD2 — but they should match, or even TBD1 might be higher than TBD2 because i9k should run cooler at constant power, data point above notwithstanding)?

If you are so concerned that the i9 will possibly have higher temperatures and be noisier that you need to ask the same question repeatedly in many different ways, please get the i7 so you can achieve peace of mind.

I know I would personally get the i9 and power-limit it to reasonable levels, unless money was the issue.

 

[Mussels]

If the same wattage is spread over more cores, core temperatures will be lower assuming the cooling can keep up with the total wattage
This is why there are no simple answers because it's multiple variables together

Ryzen has some easy examples of this, where 140W on a 5800x (8 cores, 1 CCX) runs hot while the 2 and 3 CCX variants (5900x and 5950x) run a lot colder

The key is that as the power is spread over more cores the core temps go down due to the greater surface area, but each core also runs at lower clock speeds which both increases their efficiency and reduces their performance (increasing the temperature benefits more)

Low threaded content boosts the CPU's up high, which can result in those cores still spiking up rather high which is why we have these days of single threaded loads being hotter than multi-threaded

 

[kraiggers]

If you are so concerned that the i9 will possibly have higher temperatures and be noisier that you need to ask the same question repeatedly in many different ways, please get the i7 so you can achieve peace of mind.

I know I would personally get the i9 and power-limit it to reasonable levels, unless money was the issue.

So, just to circle back on this…

I got the i9k workstation. It was in fact pretty dang loud under any load. There is apparently a **very** steep ramp from no-fan to jet-f’ing-engine.

I was able to utilize XTU, to set custom PL1 and PL2 and tau. But honestly, these didn’t make nearly as much difference as I had hoped. The fan still spun up and got loud; the adjustments merely allowed me to shorten the duration of the high fan noise. And even for that they had to be *really* aggressive. Like, 90, 90, 5 sec. Or turning off boost altogether.

So, I returned the i9k config, and I’m awaiting the i7k config I ordered. I don’t think I’ll notice any big difference, and it was $130 cheaper. (The money wasn’t really the point, but I’ll happily save it.)

A fan control utility would perhaps be a better solution for noise control?

 

[Solid State Brain]

If not even PL1=PL2=90W solved issues with fan speeds, then it's a specific BIOS/firmware-related problem with fan tuning with that system, possibly occurring only when an i9-12900K is installed. This goes beyond ordinary CPU power-related temperature problems that could be solved with PL/Tau tuning.

Fan speed "overshoot" (speeds increasing more than necessary under load, then quickly decreasing again under stabilized conditions) shouldn't be occurring, especially if power is limited to the TDP, i.e. what the system is supposed to be able to sustain indefinitely.

Software fan control could be a solution, but software fan control might be locked on your system. Also I don't know of many general applications that can be used for this.
Perhaps this one: https://getfancontrol.com/

 

[kraiggers]

If not even PL1=PL2=90W solved issues with fan speeds, then it's a specific BIOS/firmware-related problem with fan tuning with that system, possibly occurring only when an i9-12900K is installed. This goes beyond ordinary CPU power-related temperature problems that could be solved with PL/Tau tuning.

Fan speed "overshoot" (speeds increasing more than necessary under load, then quickly decreasing again under stabilized conditions) shouldn't be occurring, especially if power is limited to the TDP, i.e. what the system is supposed to be able to sustain indefinitely.

Yeah, I wonder if this enclosure is just a particularly bad fit for the i9? Or maybe the bios is poorly tuned for it? Or maybe the i7 will be just as loud, and switching to the lower processor was dumb. [Shrug] Frustratingly difficult to say.

I’m probably much more noise averse — in my home office — than most people would be in an open office work-type setting.

That fan control looks pretty dang compelling. Worth a try:

 

[Solid State Brain]

Yeah, I wonder if this enclosure is just a particularly bad fit for the i9? Or maybe the bios is poorly tuned for it? Or maybe the i7 will be just as loud, and switching to the lower processor was dumb. [Shrug] Frustratingly difficult to say.

I don't see any valid reason for the i7 (also limited to PL1=PL2=90W) to intrinsically behave better, except if a different fan behavior was programmed in the BIOS/firmware for this processor. The CPUs are physically the same; the i7 just has a disabled E-core cluster and slightly lower frequencies (but generally the same voltages).

 

[kraiggers]

I don't see any valid reason for the i7 (also limited to PL1=PL2=90W) to intrinsically behave better, except if a different fan behavior was programmed in the BIOS/firmware for this processor. The CPUs are physically the same; the i7 just has a disabled E-core cluster and slightly lower frequencies (but generally the same voltages).

That makes logical sense to me.

My thought is, if they’re both going to be louder than I’d like, and I’m going to (aggressively?) throttle them both one way or another, I might as well save the money?

I hope fan control can help me achieve my desired quiet outcome...