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

[W1zzard]

We test Intel's Core i9-12900K at various TDP levels all the way down to 50 W to determine how much efficiency is really in the new Alder Lake core, and how these power limits affect performance. Competing with the efficiency of AMD's Zen 3 Ryzen lineup is just two settings changes away.

Show full review

 

[Vecix6]

Interesting info... But difficult to compare against Zen 3 without manual power adjust...

Mmm... Probably a typo at page 9, 4th paragraph:

Had they used the classic PL1=241, PL2=125 setting, then Core i9-12900K would end up significantly slower in reviews.

It would be PL1=125, PL2=241, no?

Sorry if there is another easier way to summit this kind of mistakes.

 

[AusWolf]

Awesome test, thank you!

It shows exactly my point when I straight away dismiss claims saying "Intel is inefficient and runs hot". With custom PL settings, it's as efficient as the user wants it do be. Dropping PL values to reasonable levels gives the user a barely noticeable difference in observable performance, but a huge gain in efficiency and heat output. Reading reviewers that only test at maxed out power limits gives a one-sided and unrealistic picture.

 

[Wirko]

The effect of setting the lowest power limit of 50W is actually similar to what you get when you enable E-cores only.

Does the CPU become more inclined to put load on E-cores when it's squeezed by a low power limit?

 

[Metroid]

Incredible tests, the 12900k 100w mode is pretty good however in all tests 5900x was better overall with the same 100w. These tests showed that AMD 5900x at 100w is better than Intel 12900k at 100w.

 

[Imsochobo]

If they did 190 or 215 W or so, it would be so much better looking.
Just stupidity, makes the 12700K look so much more attractive.

 

[Minus Infinity]

I'd probably run something like a PL1 = 175W, PL2 = 225W split and you be probably at 98% of PL1=PL2=241W and a tad faster than the 190/190 split.

Any way great test and I'm really looking forward to Rocket Lake which will see significant improvements in power efficiency, IPC uplifts and up to 2x the Gracemont cores. Zen 4 and Zen 4c (Bergamo) though will provide much stiffer competition than Zen 3 or Zen 3+.

 

[AlwaysHope]

Isn't the i9-12900k stock settings with its IMC only officially support DDR5-4800? Curiously, why did the test setup go with DDR5-6000? This was also the case with i9-12900k review.
I know this is a review about power consumption, but just pointing that out as I think the benchmark results will be influenced by this.

 

[Selaya]

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.

 

[Camm]

BRB, setting P states to 125w and moving on with my life.

 

[Fouquin]

Yikes Intel... Stretching the TDP nearly 50% for a 5% gain. Giving me Presler flashbacks.

 

[Crackong]

Nice article as usual.

Some might even say Intel switched to PL1=PL2=241 W only to beat AMD's Zen 3 Ryzen processors in Cinebench.

That pretty much matches the reality.

 

[Rowsol]

It's interesting seeing the efficiency chart show a sweet spot of 75-100w. Surprised to see 50w rated worse.

 

[lightofhonor]

Crazy to see it never matches 5950X efficiency, even at 50-75W.

 

[KaitouX]

Awesome test, thank you!

It shows exactly my point when I straight away dismiss claims saying "Intel is inefficient and runs hot". With custom PL settings, it's as efficient as the user wants it do be. Dropping PL values to reasonable levels gives the user a barely noticeable difference in observable performance, but a huge gain in efficiency and heat output. Reading reviewers that only test at maxed out power limits gives a one-sided and unrealistic picture.

But you can also drop the PL with Ryzen CPUs, which would make at least the "Intel is inefficient" part continue to be true when compared to current AMD offerings. Testing at maxed out power limits isn't unrealistic when it's the default behavior in many cases, by using your argument, you could argue that the 5800X reviews were unrealistic as you could activate the Eco mode and show a much cooler and efficient CPU, but that isn't the default performance and neither is what most buyers are going to do with it. While I would love to see more reviews caring more about efficiency tweaks over overclocking, most people don't care about it.


Based on ComputerBase testing of the Eco mode on the 5900X and 5800X, they can achieve ~25%+ lower power consumption(150W at eco, 200 at stock, both measured at the wall) with ~8% performance loss, while the 5950X with the eco mode performed the same as a stock 5900X, losing 20% performance compared to stock, making it only slightly more efficient than the 5900X in Eco mode. The 5600X barely changes anything, only lowers the power consumption by 10W while performing basically the same as stock.
Would be interesting to see similar tests with other CPUs, maybe the i7 would benefit more from it, just like the 5900X and 5800X benefit more from the eco mode?

 

[watzupken]

I think this review concludes that Intel threw power efficiency out of the window in their quest to take back their single core performance crown. The Golden Cove is actually a very good step up when it comes to IPC improvements, but because they are pushed so hard in terms of clock speed, they are burning through too much power to get there. Therefore, if anyone wants to get an Alder Lake, I think it makes sense to avoid i9 12900 series. The bulk of the processing power comes from the P-Cores, so I don't see why I want to spend so much for an extra 4 E-Cores, and a chip that guzzles power for a couple of hundred Mhz more.

Awesome test, thank you!

It shows exactly my point when I straight away dismiss claims saying "Intel is inefficient and runs hot". With custom PL settings, it's as efficient as the user wants it do be. Dropping PL values to reasonable levels gives the user a barely noticeable difference in observable performance, but a huge gain in efficiency and heat output. Reading reviewers that only test at maxed out power limits gives a one-sided and unrealistic picture.

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.

 

[Selaya]

no, the power efficiency is quite irrelevant for 1T, they threw out power efficiency for their (failed) attempt on nT perf ... (also why E-cores exist but yea)

 

[chrcoluk]

125w the sweet spot then.

above that the efficiency collapses.

 

[Selaya]

125w the sweet spot then.

above that the efficiency collapses.

no, that's definitely not true. Intel's efficiency curve doesn't plummet until like 150-180W.

 

[Mussels]

This reminds me so much of what AMD did with the 5800x - gave it a PPT of 142W, and let it burn at 90C to win cinebench runs.
Cap it to ~120W and suddenly you're 30C lower for 1-2% less performance


125W seems to be the best middle ground to me as well, maybe 150W (which was not tested) would give the best of everything (And in that case, using very similar wattages to the high end ryzens eliminating the complaints about higher cooling needs)

 

[Solid State Brain]

Thanks for this test, very illuminating. To be honest, I thought the 12900K would have had even better overall processing efficiency at lower power limits, but the result was not as good as I imagined it would be.

To reach or beat the 5950X in this regard, Alder Lake will likely also need 16 cores / 32 threads clocked at relatively low speeds (in multithread) and probably good binning.

 

[InVasMani]

I'd like to see if 60w PL1/PL2 would improve the efficiency results over the 75W PL1/PL2. It might be a better sweet spot a nudge higher due to some vdroop though I'm not sure speculation really. Is the 50W results a case of PL1/PL2 falling below nominal voltage!?

no, that's definitely not true. Intel's efficiency curve doesn't plummet until like 150-180W.

Far as I can tell it's about 120W to 180W PL2 where it really nose dives hard on efficiency in exchange for minor performance for maximum turbo power while 60W to 120W is ideal sweet spot range for the PL1 base power. Basically about 1/4 between min/max of the nominal which is 120W however with two processors and a base and turbo PL it's a little complex. That's my take away at least based on what I see of the results to glee from. It would be a bit easier to get a stronger idea certain with a little more testing to see if it holds true more or less.

This pretty much falls closely in line with what I felt all along with big LITTLE when Intel first announced they'd be taking such a chip approach and people were like whoa no little oh hell no I want more big cores that consume 8000W for 2% performance gains.

 

[Yraggul666]

Thank you so much for this test!
In my ignorance i had no idea that as a user you were given control of PL1 and PL2;
but in my defense who expected Intel to ever do that?
Ahhh! I LOVE THE COMPETITION!!!!!!

 

[Solid State Brain]

I'd like to see if 60w PL1/PL2 would improve the efficiency results over the 75W PL1/PL2. It might be a better sweet spot a nudge higher due to some vdroop though I'm not sure speculation really. Is the 50W results a case of PL1/PL2 falling below nominal voltage!?

I think efficiency can improve as long as the frequency can be decreased, but by default the CPU won't normally try to decrease frequency below the "base clock" to obey the PL1 setting. The base clock should be configurable in BIOS, however.

By the way: on Rocket Lake strange throttling behavior can occur if PL1=PL2, and both are set to a low level which would make frequency decrease below the base clock. At least this is what I observed on my i9-11900 (non-k).

 

[W1zzard]

It would be PL1=125, PL2=241, no?

Sorry if there is another easier way to summit this kind of mistakes.

Fixed, this is the easiest way and perfectly fine. You may also contact me through any other means

Reading reviewers that only test at maxed out power limits gives a one-sided and unrealistic picture.

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

Does the CPU become more inclined to put load on E-cores when it's squeezed by a low power limit?

Doubt it, but will test to verify

Isn't the i9-12900k stock settings with its IMC only officially support DDR5-4800? Curiously, why did the test setup go with DDR5-6000? This was also the case with i9-12900k review.

The official memory spec is always too low. Intel supports DDR4 at 3200, which nobody who's serious runs at, so I try to pick a reasonable config that people like us run (DDR4-3600 CL16 1T). In hindsight I have to admit DDR5-6000 is a bit high, but it was the kit GSKILL offered when not much was known about DDR5 choices, options and supply.

I think this review concludes that Intel threw power efficiency out of the window in their quest to take back their single core performance crown.

Multi-core, not single-core. Check SuperPi or MP3, no difference between power limits there

that as a user you were given control of PL1 and PL2;

That's the most important dial nowadays, especially on non-K models. Your cooler can't handle the heat? Just dial down the power limit a bit--no need to manually change the clocks. It will boost as high as it can, always 100% stable, and once it hits the power limit it will reduce clocks to stay within that limit. Thanks to Throttlestop this works on every system, including laptops and OEM boxes