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

[Solid State Brain]

For what it's worth, on my i7-12700K I observed that depending on BIOS settings the CPU can favor Ring or E-core frequency rather than P-core frequency when altering the power limit, and therefore efficiency might not necessarily scale with power as expected.

 

[Melvis]

No it doesn't, it's just that the numbers on the review are plain wrong.

Says who? and im no going by your word (looks at your system specs) as that just be plain biased......

We already know the 59050X vs 12900K at stock clocks is a miss match in who wins what, till you put on PBO and the 5950X runs away in heavy 100% CPU benchmarks over the 12900K and using alot less power thats just fact. So please enlighten us?

 

[JustBenching]

Says who? and im no going by your word (looks at your system specs) as that just be plain biased......

We already know the 59050X vs 12900K at stock clocks is a miss match in who wins what, till you put on PBO and the 5950X runs away in heavy 100% CPU benchmarks over the 12900K and using alot less power thats just fact. So please enlighten us?

It's just plain biased to put a 125w power limit in my 12900k and test it? LOL.

I'm telling you, the 12900 at DEFAULT settings with a 125w power limit does around 23500 to 24200 points in CBR23, and around 18500-19000 at 75 watts. I don't know exactly what the reviewer did, but it's obvious wrong.

Do you want me to open my 3700x computer and post with that one? Will that make me less biased?

 

[Mussels]

It's just plain biased to put a 125w power limit in my 12900k and test it? LOL.

I'm telling you, the 12900 at DEFAULT settings with a 125w power limit does around 23500 to 24200 points in CBR23, and around 18500-19000 at 75 watts. I don't know exactly what the reviewer did, but it's obvious wrong.

Do you want me to open my 3700x computer and post with that one? Will that make me less biased?

For all we know, you've got a flawed test setup. As the one making the claims you have to actually prove that - TPU reviewers do so by publishing dozens of reviews and a lot of repeat testing, so flaws get noticed (changing CPU and motherboard for reviews is an easy example)

Have you confirmed your theory works with other motherboards? Other bios revisions?
It wouldnt be the first time power limits have been ignored or falsely reported by motherboard vendors to win benchmarks, on AMD or intel

 

[JustBenching]

For all we know, you've got a flawed test setup. As the one making the claims you have to actually prove that - TPU reviewers do so by publishing dozens of reviews and a lot of repeat testing, so flaws get noticed (changing CPU and motherboard for reviews is an easy example)

Have you confirmed your theory works with other motherboards? Other bios revisions?
It wouldnt be the first time power limits have been ignored or falsely reported by motherboard vendors to win benchmarks, on AMD or intel

Yeap, tested 3 different 12900k on 4 different mobos (3 apexes and a unify x). Power limit hasn't been ignored, hwinfo reported 75.06w peak consumption for the 75w power limit test.

Besides testing, it's just common sense. The 12600k cannot tie the 12900k at the same power limit in a MT test...

 

[Mussels]

Yeap, tested 3 different 12900k on 4 different mobos (3 apexes and a unify x). Power limit hasn't been ignored, hwinfo reported 75.06w peak consumption for the 75w power limit test.

Besides testing, it's just common sense. The 12600k cannot tie the 12900k at the same power limit in a MT test...

Common sense has nothing to do with how computers behave. My TV can make my GPU stutter and my ex-housemates PC would bluescreen if you opened his bar-fridge while his xbox controller was connected (power surge issue)

It's absolutely possible for wattage to get thrown to the wrong CPU cores and screw with results, especially since this review was when the platform was new

What matters here is repeatable data, making sure all tests are done the same, and if there is anomaly narrowing it down.
You wont get anywhere being hostile - you WILL get somewhere if you list exactly what you tested, the results you got, how to reproduce it and ask why the reviews was different.
(To be fair, I often come off as angry as a ferret in someones undies when i'm trying to solve problems, myself)


Is this something one of the BIOS updates could have fixed, or even the updates to windows 11's scheduler for these new CPU's?

 

[JustBenching]

Common sense has nothing to do with how computers behave. My TV can make my GPU stutter and my ex-housemates PC would bluescreen if you opened his bar-fridge while his xbox controller was connected (power surge issue)

It's absolutely possible for wattage to get thrown to the wrong CPU cores and screw with results, especially since this review was when the platform was new

What matters here is repeatable data, making sure all tests are done the same, and if there is anomaly narrowing it down.
You wont get anywhere being hostile - you WILL get somewhere if you list exactly what you tested, the results you got, how to reproduce it and ask why the reviews was different.
(To be fair, I often come off as angry as a ferret in someones undies when i'm trying to solve problems, myself)


Is this something one of the BIOS updates could have fixed, or even the updates to windows 11's scheduler for these new CPU's?

The bios on one of the mobos I tested is the default the mobo ships with. So i dont know where thr problem lies exactly, but there is definitely something wrong with the results.

Im not angry, angry are the people accusing of bias cause i have the cpu. As if i could know there was a problem if i didn't have the cpu...

Angry is also the person spamming laugh emoticons on my posts cause they don't want to admit im right

 

[TheoneandonlyMrK]

The bios on one of the mobos I tested is the default the mobo ships with. So i dont know where thr problem lies exactly, but there is definitely something wrong with the results.

Im not angry, angry are the people accusing of bias cause i have the cpu. As if i could know there was a problem if i didn't have the cpu...

Angry is also the person spamming laugh emoticons on my posts cause they don't want to admit im right

Happy is the man not baited by the fool.

Your proof is still missing, so I'll likely stay out the convo, as I was , bait me back in and I'll be pm ING you my exact opinion.

 

[JustBenching]

What proof do you want? I can post some screenshots but how would you know im running default settings? Honestly, what proof does the review have? Its just some bars made on paint...

 

[InVasMani]

TPU gets sent test review samples by big corporations to put some bars made on paint...just how many have you been sent personally and how do you choose to display benchmark results for comparison analysis that is so much better in terms of being clear and concise to look at for easy interpretation!!?

 

[JustBenching]

TPU gets sent test review samples by big corporations to put some bars made on paint...just how many have you been sent personally and how do you choose to display benchmark results for comparison analysis that is so much better in terms of being clear and concise to look at for easy interpretation!!?

I think you missed the point, and you are intentionally strawmaning.

Never said they make stuff up, what im saying is, bars are not proof. I don't require proof from TPU cause i trust them, so i dont need proof. But me trusting them doesnt mean their reviews contain proof, cause they don't. Its just a bunch of graph bars. Thats just a fact

 

[TheoneandonlyMrK]

I think you missed the point, and you are intentionally strawmaning.

Never said they make stuff up, what im saying is, bars are not proof. I don't require proof from TPU cause i trust them, so i dont need proof. But me trusting them doesnt mean their reviews contain proof, cause they don't. Its just a bunch of graph bars. Thats just a fact

Those bar graphs are compiled from real test data, and as a test engineer, This is the Way.

If you're dumb ass doesn't see that FACT who the f£#@ could possibly help you.

 

[InVasMani]

You said the tests are wrong and numbers are wrong. If you trusted their information is correct you wouldn't be on this forum arguing to the contrary that it's wrong. Show some sort of evidence around why it's incorrect. What more proof do you want than the test results they gathered!!?

 

[JustBenching]

You said the tests are wrong and numbers are wrong. If you trusted their information is correct you wouldn't be on this forum arguing to the contrary that it's wrong. Show some sort of evidence around why it's incorrect. What more proof do you want than the test results they gathered!!?

The tests are wrong. Im not saying he made them up, im saying either he or the motherboard he used applies something that shouldn't. I can link you a screenshot from my 12900 running cbr23, but how would you know im running stock settings?

Those bar graphs are compiled from real test data, and as a test engineer, This is the Way.

If you're dumb ass doesn't see that FACT who the f£#@ could possibly help you.

And yet you dont have any proof that the graphs are compiled from real data. You just trust the reviewer. If you cant understand what proof actually is then please, make your "threat" a reality and ignore me. Dont want to talk to a person that doesnt know what proof is. Especially one that calls himself a test engineer

 

[TheoneandonlyMrK]

The tests are wrong. Im not saying he made them up, im saying either he or the motherboard he used applies something that shouldn't. I can link you a screenshot from my 12900 running cbr23, but how would you know im running stock settings?


And yet you dont have any proof that the graphs are compiled from real data. You just trust the reviewer. If you cant understand what proof actually is then please, make your "threat" a reality and ignore me. Dont want to talk to a person that doesnt know what proof is. Especially one that calls himself a test engineer

See bold. Confused. Com.

Argue with yourself from now on , you might make sense to you.

 

[JustBenching]

See bold. Confused. Com.

Argue with yourself from now on , you might make sense to you.

You are confused indeed. Just because I don't think he made them up doesn't mean he provided any evidence for them....do you not understand the difference?

 

[InVasMani]

You need to do more than simply profess that TPU is wrong. You aren't winning anyone over on that argument simply stating that to be the case. Show example cases of where it doesn't align with what TPU was testing for in the first place by other credible reviewers and maybe then there can be further discussion over the matter on why the results differ and there can be more determination on which is right or wrong for which reason.

 

[JustBenching]

You need to do more than simply profess that TPU is wrong. You aren't winning anyone over on that argument simply stating that to be the case. Show example cases of where it doesn't align with what TPU was testing for in the first place by other credible reviewers and maybe then there can be further discussion over the matter on why the results differ and there can be more determination on which is right or wrong for which reason.

I did that already. Igorslab has blender results from a 12900k @ 125watts, it handily beats the 12600k by a large margin, yet TPU has the 2 very close.

Anyways, I don't need to win anyone over, I'm just telling you what is. If you don't want to accept reality, from an actual owner of the actual CPU we are talking about , then so be it, you don't have to believe me. You can believe that the 12900k loses in MT to a 12600k at same wattage...I don't mind

Here are some CBR23 runs, at stock 75w and tuned 75w

 

[HenrySomeone]

It has been shown several times already, including on this site, that when properly configured, 12900k is a monster performer at lower power limits, this for example is at 35W:

 

[JustBenching]

It has been shown several times already, including on this site, that when properly configured, 12900k is a monster performer at lower power limits, this for example is at 35W:
View attachment 251955

I can get over 15k at 34 watts actualy. And I'm talking about package power, not ia cores power.

 

[HenrySomeone]

I can get over 15k at 34 watts actualy. And I'm talking about package power, not ia cores power.

Seems entirely possible, since the picture I posted is back from November, therefore with extra time for bios updates or just more in-depth tuning, I can totally believe what you're saying (but I'm pretty sure a number of team red associates will take offense to that, hehe )

 

[JustBenching]

Seems entirely possible, since the picture I posted is back from November, therefore with extra time for bios updates or just more in-depth tuning, I can totally believe what you're saying (but I'm pretty sure a number of team red associates will take offense to that, hehe )

Still, here is a screenshot for the non believers I can do better, but then system starts losing stability at idle or light loads. The minimum Vcore drops to less than 0.6. To actually get better STABLE results ill need to start tuning the VF curve for lower loads, but honestly I don't see the point, it's already the most efficient CPU on planet Earth

 

[Solid State Brain]

What are your load VID and Vcore? They have to roughly match for power measurements to be correct and power limits to work as intended. In the above screenshot only core VIDs are visible.

 

[JustBenching]

What are your load VID and Vcore? They have to roughly match for power measurements to be correct and power limits to work as intended. In the above screenshot only core VIDs are visible.

What? Vid is the number requested by the processor, Vcore is the actual voltage supplied

 

[Solid State Brain]

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

14. Load Line (DC) should be measured by the VRTT tool and programmed accordingly via the BIOS Load Line override setup options. DC Load Line BIOS programming directly affects power measurements (DC).