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

[Solid State Brain]

Just setting PL2 in the BIOS doesn’t help much since the CPU and motherboard will still use high voltages for no reason. Best method is setting PL2 at 241w and reducing the voltage via V/F and you’ll get same performance at around 170w CPU package in Cinebench. 27k multi points at 170w is efficient as hell.

Motherboards sometimes actually already use lower CPU voltages than they should, to the point of instability during certain loads at high power levels. My Gigabyte Aorus B560M was one such cases. At normal or even moderately intense loads it was stable, but it quickly failed with intense benchmarks like Prime95 or Linpack, giving calculation errors, all at stock/default settings. The motherboard is fine; default settings were bad.

So, while you may find your voltage too high for typical loads, care must be taken to make sure that by lowering it stability is still maintained at all loads and also temperatures, at least if you value that.

 

[pexxie]

Amazing. W1zzard is becoming the new Bill Nye.

Yeah, dropping those watts like that; it probably makes sense to disable cores, especially for max thread performance. I would keep an eye on the frequencies, and would want them all to be reaching max mhz. Although that's a thread perf optimization effort, not so much a for-science effort.

 

[Garrus]

I should point out that not all motherboards perform properly with custom PL settings. I've had some boards cause the CPU to wildy swing up and down between frequencies instead of settling on lower frequencies. After building several computers for clients where I just set the PL limit to the cooler limit and told them similar things to this article, I ran in to a problem where the PL limit setting didn't work right with a motherboard and it put egg on my face, as a system builder. I still find it best to just manually tune a lower voltage and frequency for all cores instead of relying on the PL setting. 120W at 4.5ghz with an i7-12700k for example was possible with a custom voltage setting. Solved the client's problem.

If you are building an i9 or an i7 system, it is best to just buy a very expensive cooler instead. Noctua U12A for example. AMD has the advantage here in that you can use a Noctua U12 non A (such as Redux) with every CPU and it is just so much easier to build systems with AMD that my clients prefer it.

 

[AusWolf]

Reviewers tend to test the product as is, i.e. recommended/ out of the box settings for the product. If Intel have changed the way the PL1 and 2 works, then it makes sense that it is tested as is, i.e. running at PL2 indefinitely as long as the cooler can maintain the temps below the threshold. And out of the box, the i9 12900K is factually a very power hungry product. So I don't think there is anything wrong with the testing. Only power users will go in and tweak the settings to make it more power efficient, and how many % of people buying an Intel Alder Lake CPU will be going into the BIOS to tweak it? In OEM machines, that option to tweak power limit may not even be available. So I don't think the general consensus from review sites paints an unrealistic picture.

Intel made it clear several times that this is the way the product is intended to be used.

Fixed that for you.

Intended: maybe, stock: no. At least not for pre-12th gen.

I've dealt with a few 400 and 500 series motherboards, and they all default to Intel indicated TDP for PL limits, unless core enhancement / optimization is turned on. MSi has an option to toggle power limits to fit your cooling (air = Intel stock, high end and water = raised PL). My Asus TUF B560M-Plus treats my 11700 as a proper 65 W CPU locked to 65/228 W PL1/2 with a 28 s Tau. With "Asus optimizer" enabled, my PLs are 200/250 W. That is not stock. It's Asus's spec for this particular motherboard and CPU. Testing with such features enabled and calling it "default" is misleading.

Maybe it's changed for 12th gen, I don't know.

Anyway, great review.

 

[KaitouX]

Intended: maybe, stock: no. At least not for pre-12th gen.

I've dealt with a few 400 and 500 series motherboards, and they all default to Intel indicated TDP for PL limits, unless core enhancement / optimization is turned on. MSi has an option to toggle power limits to fit your cooling (air = Intel stock, high end and water = raised PL). My Asus TUF B560M-Plus treats my 11700 as a proper 65 W CPU locked to 65/228 W PL1/2 with a 28 s Tau. With "Asus optimizer" enabled, my PLs are 200/250 W. That is not stock. It's Asus's spec for this particular motherboard and CPU. Testing with such features enabled and calling it "default" is misleading.

Maybe it's changed for 12th gen, I don't know.

Anyway, great review.

Many boards for the last few generations have MCE on by default, which throw power limits out of the window. I believe GN and HUB both already did videos on MCE being on by default.

You basically described the 5800 non-X for OEMs. That chip is one of the most efficient Desktop CPUs out there.

I believe activating the Eco mode achieves about the same result, unfortunately it was ignored by most reviews. Even if it wasn't the center of the review, letting people know that you have the option would have been nice.

 

[TheoneandonlyMrK]

Intended: maybe, stock: no. At least not for pre-12th gen.

I've dealt with a few 400 and 500 series motherboards, and they all default to Intel indicated TDP for PL limits, unless core enhancement / optimization is turned on. MSi has an option to toggle power limits to fit your cooling (air = Intel stock, high end and water = raised PL). My Asus TUF B560M-Plus treats my 11700 as a proper 65 W CPU locked to 65/228 W PL1/2 with a 28 s Tau. With "Asus optimizer" enabled, my PLs are 200/250 W. That is not stock. It's Asus's spec for this particular motherboard and CPU. Testing with such features enabled and calling it "default" is misleading.

Maybe it's changed for 12th gen, I don't know.

Anyway, great review.

We're not on about pre 12th gen, your deluded and rambling pal.
( Takes a look at thread Title) sigh I'll leave it but damn I'm tempted because , well just stop.

 

[W1zzard]

Intended: maybe, stock: no. At least not for pre-12th gen.

Before 12th gen the power limits were something like 125 W/250 W 56 seconds .. this has changed with Alder Lake to be PL1=PL2. This is the new default for Alder Lake, everything else is not spec, but Intel keeps adding "if you want to configure it differently you may do so"

 

[Borc]

Unfortunately something is completely broken in this test, the 100W, 65W etc. scores are way too low. Much lower than any other scaling test.

 

[AusWolf]

We're not on about pre 12th gen, your deluded and rambling pal.
( Takes a look at thread Title) sigh I'll leave it but damn I'm tempted because , well just stop.

Instead of calling me delusional for not having any experience with a 12th gen Intel system, how about letting me know where I'm wrong? Something like this:

Before 12th gen the power limits were something like 125 W/250 W 56 seconds .. this has changed with Alder Lake to be PL1=PL2. This is the new default for Alder Lake, everything else is not spec, but Intel keeps adding "if you want to configure it differently you may do so"

Thanks for clarifying this instead of attacking me for coming from a 10/11th gen point of view. This is the attitude we need more of on the forum.

 

[Mussels]

@W1zzard, any plans to do head-to-head comparisons with Zen3? P-cores only at 88W/141W to match 5800X?

That'd be interesting to do a "watt to watt" comparison 12900k vs 5950x

Poor w1zz, the time it takes to do these things is immense but ooooh brain candy

 

[Solid State Brain]

Unfortunately something is completely broken in this test, the 100W, 65W etc. scores are way too low. Much lower than any other scaling test.

Scaling with power does seem strange in some cases, for example Cinebench R23 Multi. Possibly it might have to do with the E-cores, or perhaps there's an issue similar to what I observed on my i9-11900 when setting PL1=PL2 at lower power levels.

This might be clearer by graphing the results like this:

EDIT: I made a short test with my 11900 and it scales similarly (it maxed out at 146W during the test), so it could just be that the score scaling feels unusual compared to the time scaling with fixed workloads (e.g. Blender, etc).

 

[Mussels]

You basically described the 5800 non-X for OEMs. That chip is one of the most efficient Desktop CPUs out there.

AMD pretty much did just that.

Yeah, that's my point. If you look at the early zen 3 leaks, we had a 5700x mentioned that never appeared - i think both companies are fighting to win irrelevant benchmarks at this point

 

[bug]

I'd be curious to know whether there's a setting for tau as well in the UEFI.

 

[Ed_1]

I'd be curious to know whether there's a setting for tau as well in the UEFI.

In the bios there 3 settings you can change, short time max power (PL2), long time base power (PL1) and TAU. so you have two wattages and one time value.
Stock PL1=PL2 so Tau doesn't come into play.

 

[Borc]

Scaling with power does seem strange in some cases, for example Cinebench R23 Multi. Possibly it might have to do with the E-cores, or perhaps there's an issue similar to what I observed on my i9-11900 when setting PL1=PL2 at lower power levels.

This might be clearer by graphing the results like this:

EDIT: I made a short test with my 11900 and it scales similarly (it maxed out at 146W during the test), so it could just be that the score scaling feels unusual compared to the time scaling with fixed workloads (e.g. Blender, etc).

It needs to be checked what is wrong, I mean all other get 22500-23000 points in Cinebench R23 at 125W and he barely gets 19000 points. The 190W+241W scores seems to be fine.

 

[LTUGamer]

 

[bug]

View attachment 228301

Mind you, there's nothing wrong with self overclocking and making use of thermal headroom as long as the silicon can take it.

 

[Mussels]

View attachment 228301

When theres no competition, they tune them for efficiency and give us OC'ing headroom

When the competition runs hot, both teams run the chips balls to the wall and leave efficiency behind

 

[InVasMani]

600W balls to the wall the ways it's meant to be played by Uncle Huang leather jacket not included. That is a pretty good assessment they slap down the efficiency and cruise control when competition is far in the rear view mirror, but closer than it appears.

 

[GURU7OF9]

Great and very interesting review .
Well done wizzard !

 

[AusWolf]

When theres no competition, they tune them for efficiency and give us OC'ing headroom

When the competition runs hot, both teams run the chips balls to the wall and leave efficiency behind

"Hey boss, this new chip we're working on is still 0.1% behind AMD."
"Hmm... let's see what happens if we run the power output of 5 atomic plants through it."

 

[Solid State Brain]

When theres no competition, they tune them for efficiency and give us OC'ing headroom

When the competition runs hot, both teams run the chips balls to the wall and leave efficiency behind

This seems correct. On one hand, I find positive that users can now easily decide whether to use CPUs in factory-overclocked mode, or use them more like they were used to be produced in the past just by changing a few options (often just the power limits). On the other hand, I think a reasonably "efficient mode" should have been the default, perhaps through some sort of officially-defined simple PL preset system.

 

[wolf]

600W balls to the wall the ways it's meant to be played by Uncle Huang leather jacket not included.

What are you on about?

 

[londiste]

Scaling with power does seem strange in some cases, for example Cinebench R23 Multi. Possibly it might have to do with the E-cores, or perhaps there's an issue similar to what I observed on my i9-11900 when setting PL1=PL2 at lower power levels.

This might be clearer by graphing the results like this:

EDIT: I made a short test with my 11900 and it scales similarly (it maxed out at 146W during the test), so it could just be that the score scaling feels unusual compared to the time scaling with fixed workloads (e.g. Blender, etc).

I bet if you put frequency on the same graph it turns out to be just because the frequencies it reaches require more and more power.
At 241W P-cores run at 4.9GHz, at 190W it is probably 4.7GHz

 

[Solid State Brain]

I bet if you put frequency on the same graph it turns out to be just because the frequencies it reaches require more and more power.
At 241W P-cores run at 4.9GHz, at 190W it is probably 4.7GHz

Performance in multithread benchmarks like Cinebench should increase almost perfectly linearly with frequency, but the presence of slower E-cores on Alder Lake makes this relationship less clear.

I tried repeating a test at fixed all-core frequencies from 4.6 GHz down to the base clock of 2.5 GHz with my i9-11900. Plotting the results against CPU frequency makes them easier to understand. I just varied all-core boost frequency, not the PL, so perhaps this has an effect as well, but it is clear that power requirements increase more or less exponentially with frequency, so squeezing a few hundreds more MHz would have a very large effect on them. This should a general principle valid for all CPUs.

EDIT: by the way, the effect of temperature on power consumption can alter the results as pointed out earlier by Xebec, so this would have to be taken into account when doing CPU power testing. Hotter CPUs will throttle at lower frequencies for the same power limit when hot, or conversely consume more power if the frequency is fixed.