Intel Core i9-14900K Raptor Lake Tested at Power Limits Down to 35 W

[W1zzard]

There's no denying that Intel's Core i9-14900K is a power hog. This article examines the power, performance and thermals effects of limiting its power consumption, with very interesting results. Additionally, we investigate undervolting to determine if efficiency can be optimized even further.

Show full review

 

[Denver]

Good test to remember how impressively efficient the 7800x3D is. It would be interesting to see how well the X3D performs under an even lower TDP.

 

[phanbuey]

This is a great insight, thank you for this review.

 

[izy]

Not bad with the undervolt , they better release an i9-14900 non K , it would be a better buying option

 

[AnotherReader]

Thanks for this interesting test showing what a saner 14900K would have been like. Testing the 7950X with lower power limits would be interesting too.

 

[Imouto]

This makes my head hurt.

Energy Efficiency
Just looking at power draw in watts won't paint the whole picture for any given processor. It's not only important how much power is consumed, but also how quickly a task is completed—taking both into account results in "efficiency." Since a faster processor will complete a given workload quicker, the total amount of energy used might end up less than on a low-powered processor, which might draw less power, but will take longer to finish the test. In this section, we divide the performance achieved by the power usage, to get a Cinebench points per watt single-threaded and multi-threaded result. For gaming (with GeForce RTX 4090) we divide the average FPS of all our games by the power consumed, for a frames per watt rating. All these tests are based on our CPU-only power measurements, not whole system.

When I thought the English language couldn't be more stupid it goes and allows you to freely exchange "power" and "energy". Like... seriously... something like "power divided by power equals uptime".

Ugh.

 

[HD64G]

3% gaming performance increase going from 125 to 253W! Desperation in all its glory...

 

[Tom Yum]

This makes my head hurt.


When I thought the English language couldn't be more stupid it goes and allows you to freely exchange "power" and "energy". Like... seriously... something like "power divided by power equals uptime".

Ugh.

Power (Watts) is the energy consumed (Joules) per second. By measuring the total Watts consumed for a task, and the time the task took, then you can calculate the energy consumed for the task as the time variable (seconds) cancel out.

Some benchmarks though give performance as a rate of work (Cinebench for example), so the time unit is baked into the measurement and therefore a 'Joules per Cinebench pt' can't be calculated (though you could maybe calculate a 'Joules per run' measure of energy efficiency'. So it makes sense that TP use 'Cinebench pts per watt' as a measure of energy efficiency, as both have a hidden 'per second' variable.

There are a few errors in there (frames per watt should be frames per Joule, FPS per watt would also make sense), but the use of words power and energy seem fine to me.

 

[FreezingPC]

Give your CPU 70w~ less and you keep 95%~ of its gaming power, 95w UV sounds like a deal.

 

[theouto]

Man, it feels like they just factory OC these, the difference between 125w and stock is almost non existent when compared to the power draw.

They could have very good marketing on efficiency if they were a bit more sensible with their tuning, but I guess performance is performance

 

[Selaya]

no like-for-like comparison w/ 13900k(s) at the same power limits? mehh

 

[Imouto]

Power (Watts) is the energy consumed (Joules) per second. By measuring the total Watts consumed for a task, and the time the task took, then you can calculate the energy consumed for the task as the time variable (seconds) cancel out.

Some benchmarks though give performance as a rate of work (Cinebench for example), so the time unit is baked into the measurement and therefore a 'Joules per Cinebench pt' can't be calculated (though you could maybe calculate a 'Joules per run' measure of energy efficiency'. So it makes sense that TP use 'Cinebench pts per watt' as a measure of energy efficiency, as both have a hidden 'per second' variable.

Well, thing is you don't use "total watts consumed" in Spanish because it makes no sense. It would be something like "average power draw multiplied by time equals energy used". I

I'm pretty sure the formula for CB's score is just (1 / t) * whatever number they felt comfortable with = points. The points are there just for easier e-peen measurement since it's a fixed work.

It's also kinda disappointing that having Blender data for both power and performance and choosing such a pointless benchmark as CB. It would be so easy to calculate in actual energy used instead of a made up metric to please the average reader.

There are a few errors in there (frames per watt should be frames per Joule, FPS per watt would also make sense), but the use of words power and energy seem fine to me.

Those instances are what I'm referring to but you can't really point them out as errors since in English you can exchange power and energy as you please. Maybe it's not as "technical" and you'd expect, but correct nonetheless.

Also, as I said, power is never consumed in Spanish. Power is used, draw, etc. Consumed? That's for energy and sounds so wrong when paired with power.

3% gaming performance increase going from 125 to 253W! Desperation in all its glory...

The real comedy gold is in the multithreaded benchmarks. Both AMD and Intel are so desperate to claim the "king of the hill" title they run efficiency into the ground.

 

[W1zzard]

This makes my head hurt.


When I thought the English language couldn't be more stupid it goes and allows you to freely exchange "power" and "energy". Like... seriously... something like "power divided by power equals uptime".

Ugh.

these terms are not interchangeable. Good explanations from the other posters

i did simplify the naming of some averages slightly to make it a bit more approachable for a general audience, the underlying physics is still correct imo

Joules

We used joules in the previous cpu bench and nobody understood it, so i changed it to a more familiar way of saying the same thing by exposing watts instead of joules

 

[Dr. Dro]

Shame that you didn't include the 13900KS in the data, W1zz. I wanted to see if the advanced binning that it enjoys actually helps on the efficiency/undervolting side. The 14900K is clearly a high-volume production part and perhaps that is its greatest achievement, the first "commonly produced" CPU that can hit 6 GHz clocks, even though at stock, it's not as behaved as the KS - understandably so, that's where the special binning shines. But it's not particularly cherry-picked like the 13900KS is, and it seems most 14900K chips that made to the hands of enthusiasts are not better processors on average.

I very much doubt that a 14900KS will ever come out at this point, given they were unwilling to make the 14900K's final clocks any higher than they are on the 13900KS and knowing that there are zero physical changes and no further revision made to the processor.

 

[FrostWolf]

This convinces me more as I read it that when I have my system rebuilt with the 13700k (in progress, replacing the 13600k I had) that I should disable multicore enhancement, drop the PL1/PL2 and undervolt as well. I can’t see my way to any CPU going over 200w tops, and looking at this, even that seems high.

 

[Dr. Dro]

This convinces me more as I read it that when I have my system rebuilt with the 13700k (in progress, replacing the 13600k I had) that I should disable multicore enhancement, drop the PL1/PL2 and undervolt as well. I can’t see my way to any CPU going over 200w tops, and looking at this, even that seems high.

Why not a 14700K? Seems like this 8P+12E SKU is actually the one to get, with the pricing being what it is and all. The 14900K... well, it's cheaper than a 13900KS and is basically one, so I guess you can't hit it too much other than for being as low effort as Intel could have it.

 

[Imouto]

these terms are not interchangeable. Good explanations from the other posters

They either are or you are using them incorrectly. I've checked the definition of "power" to make sure and to my utter disdain this is one of the definitions:

"Electrical or mechanical energy, especially as used to assist or replace human energy."

i did simplify the naming of some averages slightly to make it a bit more approachable for a general audience, the underlying physics is still correct imo

No.

"It's not only important how much power is consumed, but also how quickly a task is completed—taking both into account results in "efficiency."

Taking both into account you get "energy".

We used joules in the previous cpu bench and nobody understood it, so i changed it to a more familiar way of saying the same thing by exposing watts instead of joules

I'm so sorry, honestly.

 

[Outback Bronze]

Good to see a review like this where these CPU's can bring light onto power consumption when manipulated correctly.

They don't have to be the melting point of diamonds that everybody perceives.

Great review W1z. Thanks!

 

[AnotherReader]

these terms are not interchangeable. Good explanations from the other posters

i did simplify the naming of some averages slightly to make it a bit more approachable for a general audience, the underlying physics is still correct imo


We used joules in the previous cpu bench and nobody understood it, so i changed it to a more familiar way of saying the same thing by exposing watts instead of joules

I think Task energy, as seen in some of your previous reviews, is better for longer running multi-threaded workloads.

 

[THU31]

Interesting thing about reported power vs. measured power. I have an ASUS Z790 board now, but when I had a Gigabyte Z690 board, it reported about 15 W less when fully loaded with 6 P-cores (E-cores disabled), but temps were the same as on my current mobo. Idle power was reported as half of what I see now, low load was also lower.
I replaced the board because I had an unsolvable USB audio issue, I should've stuck with ASUS in the first place. I was using their boards exclusively for 15 years and they never let me down. I chose Gigabyte because of the X3D SoC voltage fiasco, but I guess that was a mistake. Maybe I was just unlucky.

I'm also curious about the whole "SVID behavior" thing. What is actually considered default? Auto, worst-case or failsafe? You can get drastically different results with all these options.

Anyway, in my opinion 125 W should be the default turbo power limit of all these Intel K CPUs. Who cares that you win Cinebench if you get trashed in every single review? Even with that power limit, efficiency is behind Zen 4, but at least it's respectable.
You should still be able to just increase the power limit to whatever you want and the CPU should boost as high as it can, but the default behavior should be limited to 125 W.

 

[JustBenching]

As I was saying for the past few years, the T versions of Intel CPUS are untouchable in terms of efficiency. You can clearly see now the 14900k at 35w easily topping the charts.

 

[Wirko]

As I was saying for the past few years, the T versions of Intel CPUS are untouchable in terms of efficiency. You can clearly see now the 14900k at 35w easily topping the charts.

Sure but these charts don't include other chips, neither Intel nor AMD, runnung at 35W.

 

[JustBenching]

Sure but these charts don't include other chips, neither Intel nor AMD, runnung at 35W.

AMD chips can't run that low due to the IO die, performance plummets under 50w. So - it doesn't matter.

Other Intel chips don't make a difference, I've tested a 12900k and a 13900k, they both get similar performance at 35w. Both undervolted, my CBR23 score was 15.200 for the 12900k and 15.900 for the 13900k.

 

[oiawshboiwsahnoidaws]

Nice work! Interesting to see how efficient things can be done at lower power/clocks and would be nice to see how well amd does with lower power limits/undervolting
Only found one another comparison between 13900k vs 7950x and amd seems more efficient at lower power but guess could be that their 13900k was a dud or crap board settings on intel

 

[Icon Charlie]

This makes my head hurt.


When I thought the English language couldn't be more stupid it goes and allows you to freely exchange "power" and "energy". Like... seriously... something like "power divided by power equals uptime".

Ugh.

When my brain begins to hurt like this, I go to my 17 year old home made, 80 Proof, Kalhua. Double the fire power of the store brought brand and made with real dark brown sugar instead of sugar.

Double shot of my stuff on ice and milk and the pain goes away. Yup, it goes away with a smile.

Oh and thank you W1zzard for the article. A nice read overall.