Intel Core i9-12900K

[defaultluser]

Good article as usual.




Intel PR marketing Leaks:

8XX CPU-Z score
Best XXX CPU
Total dominance


Reality:
Win some, Lose some
Double the power consumption
Double the heat
Double the platform cost
Windows 11

And the best part of this whole thing is: if the platform cost is reduced by buying a DDR4 mother board (and not waiting months for sold-out kit,) the 5% Adler Lake performance advantage swings back in AMD's favor!

Zen 3+ is going to crush this thing into the ground!

 

[Valantar]

Seems like a pretty decent effort from Intel! Definitely major performance improvements across the board. Ineresting to see how most applications seem to handle the E cores fine, but some seem to stumble completely with them active. Wonder if this is down to the scheduler or the application.

Power consumption is still worrying though, and the inability to cool the CPU properly at all with a U14s - which is not a small cooler! - is pretty shocking. This is a top-of-the-line CPU, sure, but it shouldn't require an AIO still.

@W1zzard two questions:
- Why are your graphs consistently ranked with the best result at the bottom? This feels very counterintuitive and weird. If nothing else I would question how good a choice this is in terms of readability/accessibility (whether for those with sight impairments, dyslexia or others), as the convention of 'best on top' is pretty universally accepted and breaking conventions like that can make reading much more difficult.
- Is your motherboard actually respecting Intel's stock power limits including Tau? Power draw numbers for stock and unlimited are near identical, which would seem to indicate that either Tau is infinite or something else fishy is going on. Shouldn't the limited version be stepping down to 125W for a steady-state power draw?

 

[oxrufiioxo]

Shouldn't the limited version be stepping down to 125W for a steady-state power draw?

I think intel ditched that approach with Alderlake likely to win in some MC benchmarks.

 

[W1zzard]

as the convention of 'best on top'

Not sure if 'convention', and it's how we've done things since forever. Happy to change it if there's sufficient demand

Is your motherboard actually respecting Intel's stock power limits including Tau?

Of course. Intel stock limits for the 12900K is PL1=PL2=241 W. There is no stepping down and no steady state 125 W.

The SKU table mentions "Processor Base Power" "125 W", how does that work when PL1=PL2=241W by default?

Intel’s processor specifications and programming guidelines allow for setting PL1 within a range of values including the base power level and PL2 level.

My translation is: "The default is PL1=PL2=241, but you can change the value, manually, to any other number if you want". Which is factually 100% correct of course.

What is more important here is what they didnt say: "but the default really is 125W", "wizz you got it all it all wrong", "why we abolished 125 W" "is there even 125 W besides the specs table" "is 125 W bs"

 

[Pilgrim]

This looks depressing tbh. I don't know how you guys could give the 12900K a "Recommended" award. Recommended for what exactly? And over which processor? The only decent processor in this lineup is the 12600K.

 

[Darmok N Jalad]

It looks to me like like they're fighting for the highest power consumption crown. 10nm, excuse me, I meant to say 7nm, with a new architecture and still horrendous power figures, how is this possible ?

Edit : I just remembered that this is supposed to have 8 "efficiency cores", holy crap this is beyond laughable. What if this was an "actual" 16 core CPU ? What would it use ? 400W ?



No, it's not, DDR4 vs DDR5.

And it's not about it being unfair, having just one platform on DDR5 isn't enough to infer how good these CPUs actually are. Any CPU with faster memory will also perform better, nothing new here.

Yeah, the e-cores do nothing for idle savings, probably due to the complex power delivery system needed to supply a CPU that is allowed to consume 241W for as long as it can. This is the concern I kept bringing up—complexity adds to cost. Not only you pay more when you purchase, but operating the system isn’t any cheaper than anything else—it is probably even more expensive to operate since you’re consuming more power and fighting more heat. What value do the e-cores bring beyond eeking out some multithreaded wins? Sounds like they end up as a net-negative, since they can accidentally receive threads meant for the P-cores. It’s a dubious design that requires a well-behaved scheduler.

Honest question, but how many gamers do any of these top CPUs really help out? I was under the impression we were past “CPU-limted” gaming performance some time ago. What does this offer over something more mid-grade in terms of real-world value? I don’t game anymore, so I’m curious how much value this really brings beyond being considered the fastest.

 

[Lord_Soth]

Physics PhD, may I presume?

You don't need a an physics PhD to understand that intel has pulled this processor to the clocks he needs to rival AMD, whatever power consumption this required.
The thing that seems no one has noticed in the review is that any to PL1, PL2 or E cores setting has almost no impact on consumption, which remains very high if not increase.
The 125w in the specs is a total scam...

 

[The red spirit]

If I hadn't knew that this was supposed to be a whole new architecture, from results alone I would have thought that Intel just raised clocks again and increased PLs. Performance improvement is really underwhelming and it couldn't decisively beat Ryzens. Energy efficiency is in toilet, old ass i5 10400F is winning there. And impossible to cool with any normal means. Embarrassing FX 9590 was possible to cool with puny 120mm AIO or Hyper 212, but this can't be cooled with D15, 280mm AIOs. This is yet another garbage release from Intel, very dissapointing.

 

[FedericoUY]

THere shouldn't be required to turn off E cores to get the best performance out of the cpu. I hope that gets fixed soon...

 

[cadaveca]

Not sure if 'convention', and it's how we've done things since forever. Happy to change it if there's sufficient demand


Of course. Intel stock limits for the 12900K is PL1=PL2=241 W. There is no stepping down and no steady state 125 W.




My translation is: "The default is PL1=PL2=241, but you can change the value, manually, to any other number if you want". Which is factually 100% correct of course.

What is more important here is what they didnt say: "but the default really is 125W", "wizz you got it all it all wrong", "why we abolished 125 W" "is there even 125 W besides the specs table" "is 125 W bs"

How much of this is up to teh board maker....?

 

[W1zzard]

How much of this is up to teh board maker....?

They may set any value, but they Intel default is PL1=PL2=241 W for 12900K. Many boards actually set PL1=PL2=maximum=4095 W by default. Have for years, it's that whole ASUS MultiCore Enhancement debate again. I always test my CPUs at stock power limits, and provide an additional data point for all power limits removed

 

[luches]

Looking at all the charts, what are the odds of Zen3+ completely closing the gap again and taking back the crown ? They did said average 15% uplift in games.

 

[W1zzard]

Added CPU-Z screenshot for 5.0 GHz OC, OC text, OC results for power and temps. Now benching OC 5.0 performance, will have results in around 4 hours

 

[Flanker]

Hi Flanker. You mentioned on pctuning.cz, that you managed 28 593 points with your R9 5950X in Cinebench R23. Can you classify your setup and If you had PBO enabled? W1zzard managed only 25813 points in this review. As I checked many reviews have even lower scores than his.

Unfortunately that's another person with the same username

 

[Vya Domus]

Yeah, the e-cores do nothing for idle savings, probably due to the complex power delivery system needed to supply a CPU that is allowed to consume 241W for as long as it can. This is the concern I kept bringing up—complexity adds to cost. Not only you pay more when you purchase, but operating the system isn’t any cheaper than anything else—it is probably even more expensive to operate since you’re consuming more power and fighting more heat. What value do the e-cores bring beyond eeking out some multithreaded wins? Sounds like they end up as a net-negative, since they can accidentally receive threads meant for the P-cores. It’s a dubious design that requires a well-behaved scheduler.

To me power consumption isn't usually a problem, but when you have a CPU that outputs as much heat as a mid range GPU it's starting to become kind of insane. Intel's E-cores would make sense in a laptop but now we know they're completely worthless because they still use a ton of power anyway.

Honest question, but how many gamers do any of these top CPUs really help out? I was under the impression we were past “CPU-limted” gaming performance some time ago. What does this offer over something more mid-grade in terms of real-world value? I don’t game anymore, so I’m curious how much value this really brings beyond being considered the fastest.

You're right, game these days are basically never CPU limited unless you specifically look for that. But there are many who would still not shut up about how getting 400 FPS instead of 350 in CS:GO or something like that makes huge difference, so here we are.

 

[ValenOne]

AMD may need to release XT Zen 3 variants i.e. make overclock official. 3D cache Zen 3+ would be overkill.

 

[Lightofhonor]

With a SFF PC it's pretty tough to run a 5950X, but seems like it would be impossible without undervolting/limiting the new 12900K.

 

[Valantar]

Not sure if 'convention', and it's how we've done things since forever. Happy to change it if there's sufficient demand

Yeah, I think I've noticed it before as well, I guess this is just the first time in a while I've been looking at this many charts at once. As for questioning whether "best on top" is a convention ... universally used phrases such as "who is on top", "top of the line", "topping the charts" etc. should be plentiful evidence for that being the dominant convention. The only place I'm used to seeing "number one at the bottom" is in listicle-type writing where the point is to make readers go through the entire list and not just look at number one and then leave. At least that line of reasoning doesn't apply here.

Another thing: you're not really consistent about it. At least in the 12600K review's power consumption section you have the lower numbers (i.e. the better ones) on top, while in the energy efficiency part (on the same page) you have the higher (worse) numbers on top. The same is true for the temperature graph in this review - lowest/best on top. And IIRC I've seen similar inconsistency previously. I could understand a hard-line "the higher value will always be on top, regardless if it's good or bad"

Of course. Intel stock limits for the 12900K is PL1=PL2=241 W. There is no stepping down and no steady state 125 W.

So they removed TDP and replaced it with two more informative specifications just to immediately render one of them irrelevant. Great move!

My translation is: "The default is PL1=PL2=241, but you can change the value, manually, to any other number if you want". Which is factually 100% correct of course.

That sounds to me like they're talking about motherboard manufacturers - "programming guidelines" is not something that end users are privy to to my knowledge. Which I guess just means that, as you say, MCE all over again.

What is more important here is what they didnt say: "but the default really is 125W", "wizz you got it all it all wrong", "why we abolished 125 W" "is there even 125 W besides the specs table" "is 125 W bs"

So effectively the only thing that's changed is that PL2 is now listed in the spec table. I guess that's ... "progress"?

 

[robert3892]

Using DDR5 6000 memory in your Alder Lake system versus DDR4 3600 skews the results in my opinion. Why didn't you use higher frequency DDR4?

 

[ShurikN]

Good thing intel released the Alder Lake line-up close to winter, all joking aside, is 12900K viable in an ITX setup? Can I dare to think air cooled and ITX???

Dave2D tried. Says it cant be done.

 

[maxfly]

Well done! Honestly appreciate your answering all our questions W1zzard.
Looking forward to seeing how the ddr4 boards fare. The added 12900k heat means nothing to me personally. My loop is configured to handle anything. But the combination of ddr5 and blown up mb expense is frustrating to say the least. It was expected that ddr5 pricing would be stupid but mb mannies need a good kick in the rear.
AMD will continue to be my go to until both come down in price significantly.
Even if i were upgrading my own rig im not into early release memory. So by the time something acceptable is released at a reasonable price, it'll be time to evaluate AMDs next swing at the fences. Not a bad thing

 

[cadaveca]

Using DDR5 6000 memory in your Alder Lake system versus DDR4 3600 skews the results in my opinion. Why didn't you use higher frequency DDR4?

given how memory stability is, i'd say the equivalents are basically there. I mean, i get what you're saying, but then shouldn't we be seeing like DDR5-6600-6800?

Dave2D tried. Says it cant be done.

In that little blue box? LOL.

 

[Pilgrim]

Using DDR5 6000 memory in your Alder Lake system versus DDR4 3600 skews the results in my opinion. Why didn't you use higher frequency DDR4?

It won't matter much since Zen 3 infinity fabric can't clock much higher than 1800Mhz (Some lucky chips will go up to 2000Mhz). They need 1:1 infinity fabric to DRAM frequency to get the best performance. 3600MT/s is already the sweet spot

 

[ShurikN]

In that little blue box? LOL.

Not really sure what you're getting at with this comment.

 

[Valantar]

It won't matter much since Zen 3 infinity fabric can't clock much higher than 1800Mhz (Some lucky chips will go up to 2000Mhz). They need 1:1 infinity fabric to DRAM frequency to get the best performance. 3600Mhz is already the sweet spot

3600 is at least a reasonable expectation of something everyone can hit at 1:1, which is a good starting point for benchmarking. IMO the bar should be either that or whatever the spec of the chip is, which would leave Ryzen at 3200 and these at ... 4800?

At least good to see that these can indeed handle faster memory than what Intel is specifying. That table with 4800 only supported in single rank on 1dpc boards, with even 1dpc installed on 2dpc boards being lower? That's pretty terrible.

Dave2D tried. Says it cant be done.

Can probably be done if you're willing to manually configure your power limits to something more sensible, with some undervolting to try and regain some of that performance. Would be interesting to see where this ends up on the benchmarks.