Some strange Intel Core 12900K, 12700K, and 12600K specs have appeared online

Puiu

Posts: 4,867   +3,749
TechSpot Elite
You're underselling the point - that's a massive benefit! Most users are working in that envelope most of the time. Time will tell if the approach is good or bad Vs more normal cores, buts it definitely looks sensible enough on paper. We won't know until we see the real world figures.
No they are not, especially not with high end CPUs. Tell me one 10900k, 11900k or 5900x normal PC that is working under those constraints.

I'm actually overselling this and trying to be nice. Switching high powered cores with low powered cores is the worst "benefit" for any high end PC. These CPUs are still limited to 8 large cores, Intel needs to do much better than that. They could have done 12+4 or something like that. (are they forced to use 8 max because of some weird architecture limit?)

Like I said before, this is mostly for either mobile devices like laptops of very niche PCs (power/cooling constrained builds). I'm sure it will shine in some battery tests.
 
Last edited:

Nanochip

Posts: 43   +47
No they are not, especially not with high end CPUs. Tell me one 10900k, 11900k or 5900x normal PC that is working under those constraints.

I'm actually overselling this and trying to be nice. Switching high powered cores with low powered cores is the worst "benefit" for any high end PC. These CPUs are still limited to 8 large cores, Intel needs to do much better than that. They could have done 12+4 or something like that. (are they forced to use 8 max because of some weird architecture limit?)

Like I said before, this is mostly for either mobile devices like laptops of very niche PCs (power/cooling constrained builds). I'm sure it will shine in some battery tests.
You're not thinking in terms of performance-per-watt-per area. Intel could've gone with larger cores (albeit less of them) for a given area, but more large cores would consume more power in the same area. So instead Intel chose to maximize this area with a blend of high performance and low-performance cores.

If the windows and linux kernel thread schedulers can take full advantage of the two types of cores and efficiently schedule threads, then you get good performance efficiency.

Case in point: if the rumors are true, the PL2 of the 16-core 12900K is 228W, which is less than that of the 8-core Rocket Lake. If (and this is a BIG IF) the chips really do perform well, and within this PL2 power budget, then Intel would've pulled off quite a feat! More cores more performance, and more efficiency. That's good engineering.

Just because a desktop CPU is connected to a beefy power supply doesn't mean the CPU manufacturer(s) shouldn't try to be as efficient as possible. The future of CPU design is going to be tiles/chiplets and blend of different types of compute. We just have to see how the new cores perform in Windows/Linux... to know whether the experiment is worth it or not.

But don't fault Intel for being an empiricist and testing how to breathe life into x86 with new designs. The ARM attack is coming. Apple M2, Qualcomm Nuvia, Ampere Altra, possibly even Nvidia. Intel and AMD have to innovate especially in terms of efficiency to blunt the attack that is coming. Customers want more efficient cores. And Intel is trying to deliver that to the ecosystem.

And besides, this will allow Intel to simplify and unify its cpu-design. They can have one design for both laptop and desktop... unlike 11th gen which has different core designs for both laptop and desktop. If you look at Apple for instance, Apple is able to manufacture one chip, the M1, and put it into many different products: Mac mini, MacBook, iMac, iPad. Much more simple (from a design perspective) than having to expend energy to design, test, and manufacture multiple designs for desktop/laptop/tablet.
 
Last edited:

Puiu

Posts: 4,867   +3,749
TechSpot Elite
You're not thinking in terms of performance-per-watt-per area. Intel could've gone with larger cores (albeit less of them) for a given area, but more large cores would consume more power in the same area. So instead Intel chose to maximize this area with a blend of high performance and low-performance cores.

If the windows and linux kernel thread schedulers can take full advantage of the two types of cores and efficiently schedule threads, then you get good performance efficiency.

Case in point: if the rumors are true, the PL2 of the 16-core 12900K is 228W, which is less than that of the 8-core Rocket Lake. If (and this is a BIG IF) the chips really do perform well, and within this PL2 power budget, then Intel would've pulled off quite a feat! More cores more performance, and more efficiency. That's good engineering.

Just because a desktop CPU is connected to a beefy power supply doesn't mean the CPU manufacturer(s) shouldn't try to be as efficient as possible. The future of CPU design is going to be tiles/chiplets and blend of different types of compute. We just have to see how the new cores perform in Windows/Linux... to know whether the experiment is worth it or not.

But don't fault Intel for being an empiricist and testing how to breathe life into x86 with new designs. The ARM attack is coming. Apple M2, Qualcomm Nuvia, Ampere Altra, possibly even Nvidia. Intel and AMD have to innovate especially in terms of efficiency to blunt the attack that is coming. Customers want more efficient cores. And Intel is trying to deliver that to the ecosystem.

And besides, this will allow Intel to simplify and unify its cpu-design. They can have one design for both laptop and desktop... unlike 11th gen which has different core designs for both laptop and desktop. If you look at Apple for instance, Apple is able to manufacture one chip, the M1, and put it into many different products: Mac mini, MacBook, iMac, iPad. Much more simple (from a design perspective) than having to expend energy to design, test, and manufacture multiple designs for desktop/laptop/tablet.
And there is your answer. It's not about power usage, it's about simplifying production.

Apple is forced to use the M1 for the iMac since they don't yet have the bigger M2, but as you can notice everything else is a mobile device. That's the target device of such an architecture.

It's still very weird that Intel can't scale to above 8 large cores for consumer CPUs. There has to be something else that is limiting them (bad yields, low clocks, die space, or just a limit/bug/problem they couldn't fix when they designed the architecture).
 
Last edited:

Nanochip

Posts: 43   +47
And there is your answer. It's not about power usage, it's about simplifying production. It took a while for you to accept that it's not about saving a few watts in high end CPUs :)
It doesn’t have to be one or the other it could be both. A 16 core chip (alder lake) using the same power as an 8 core (Rocket lake), and getting more IPC? That’s impressive.
 

Puiu

Posts: 4,867   +3,749
TechSpot Elite
It doesn’t have to be one or the other it could be both. A 16 core chip (alder lake) using the same power as an 8 core (Rocket lake), and getting more IPC? That’s impressive.
If it didn't get the same power as Rocket Lake then it would have been a huge fail. We are not talking about just architectural improvements but also a new process node. Lets not ignore the fact that Rocket Lake is very power hungry.
 

Nanochip

Posts: 43   +47
If it didn't get the same power as Rocket Lake then it would have been a huge fail. We are not talking about just architectural improvements but also a new process node. Lets not ignore the fact that Rocket Lake is very power hungry.
Yes Rocket Lake is drunk on power, no argument there. But still, 16 cores (8 Big + 8 Small) consuming the same power as 8 cores is still a good move in the right direction for Intel.
 

Puiu

Posts: 4,867   +3,749
TechSpot Elite
Yes Rocket Lake is drunk on power, no argument there. But still, 16 cores (8 Big + 8 Small) consuming the same power as 8 cores is still a good move in the right direction for Intel.
They needed it a lot. AMD was walking all over them in terms of power efficiency.
 

JamesBlond

Posts: 124   +75
Seriously?

It's already confirmed that amd is doing the same in just a couple more generations of their chips as well.

Intel is just ahead of game in this regard.

Whether it makes them faster is still undetermined but they are making this jump, that amd is absolutely going to be making as well, first.
I am not sure intel are ahead, I think as AMD like to do keep back on new launches just to ensure they don't lose the race... stratergy
 

Theinsanegamer

Posts: 2,661   +4,143
The 5950x pulls only 130w during heavy workloads and crushes rocketlake unless AVX 512 comes in, in which case rocketlake still loses but also pulls insane amounts of power. Intel's solution to this is to staple an atom CPU tot he side in the hopes that will offset power usage instead of fixing AVX 512 so it isnt such a power hog.......

This is typical intel panicking. Reminds me of netburst, and how they stapled a second ALU onto the core desperately trying to keep the processor pipeline from stalling out.
You're underselling the point - that's a massive benefit! Most users are working in that envelope most of the time. Time will tell if the approach is good or bad Vs more normal cores, buts it definitely looks sensible enough on paper. We won't know until we see the real world figures.
Have you ever watched the power consumption of a standard CPU in a consumer machine? 99% of the time they are already sitting between 10-15w.

We're talking a likely difference of 3-4 watts at most. AMAZING! That'll offset the CO2 of the new chip in, what, 400 years?

Intel's problem is insane power usage when tryign to do something that requires heavy processing power. AMD manages double the core count and over double the performance in half the power usage (5950x). Nobody was asking for some atom cores to join the fray.
Are there fundamental architecture differences between a p-core and an e-core? For example, if you took a p-core, turned off hyperthreading, and throttled it to 4 GHz, are there any user-impacting differences between it and the e-core?
Supposedly the e-cores are almost maybe skylake level, but of course this is intel, who claimed the old atom cores were something else. Likely have to wait to see how restricted they are, my guess is much like the pentium cores the gracemont cores will have no AVX support.
 

BadThad

Posts: 617   +657
Personally for the desktop I don't want those weak cores. Since I run a lot of distributed computing, they are nearly worthless to me. I want big, fast as possible cores to get work more done. I understand why they are doing it since most users will probably not even need to run a big core in their daily work but why do us power uses have to pay? The light CPU users should simply buy more efficient CPUs!
 

Danny101

Posts: 1,837   +791
Personally for the desktop I don't want those weak cores. Since I run a lot of distributed computing, they are nearly worthless to me. I want big, fast as possible cores to get work more done. I understand why they are doing it since most users will probably not even need to run a big core in their daily work but why do us power uses have to pay? The light CPU users should simply buy more efficient CPUs!
The choice should certainly be there.
 

hwertz

Posts: 65   +28
Not sure if this helps power use as much as on ARM, but ARM big.little works GREAT in these kinds of configurations. The ARM Chromebook I had (which I but Ubuntu on) ran over 20 hours battery life, by having 4 high-speed ARM cores (2.3ghz max, good instructions per clock, but "only" clock down to like 1ghz), and a low-power ARM (same instruction set, about 1/2 the instructions per clock, only 1.2ghz max, but could clock down to like 50mhz.) It's got to run a CPU to blink the cursor, or handle downloads, or Firefox or Chrome is doing it's thing of using a few % so the CPU can't quite sleep? The low power core would tick over at 50-200mhz and use like 1-10mw power. Those faster ones would kick on in something like 1/100th of a second if I needed speed.
 

Danny101

Posts: 1,837   +791
Not sure if this helps power use as much as on ARM, but ARM big.little works GREAT in these kinds of configurations. The ARM Chromebook I had (which I but Ubuntu on) ran over 20 hours battery life, by having 4 high-speed ARM cores (2.3ghz max, good instructions per clock, but "only" clock down to like 1ghz), and a low-power ARM (same instruction set, about 1/2 the instructions per clock, only 1.2ghz max, but could clock down to like 50mhz.) It's got to run a CPU to blink the cursor, or handle downloads, or Firefox or Chrome is doing it's thing of using a few % so the CPU can't quite sleep? The low power core would tick over at 50-200mhz and use like 1-10mw power. Those faster ones would kick on in something like 1/100th of a second if I needed speed.
At least you know that Linux handles the architecture well. Windows, however...remains to be seen. What did Windows Mobile run on? I've never really looked at it.
 

Puiu

Posts: 4,867   +3,749
TechSpot Elite
You only need to learn to read English, nothing else
in all seriousness, that comment requires some next level dechypering.

"I am not sure intel are ahead, I think as AMD like to do keep back on new launches just to ensure they don't lose the race... stratergy"

Besides the grammar, what exactly did you mean with "to do keep back on launches" - did you try to say that AMD likes to hold back new technologies for future products?
 

JamesBlond

Posts: 124   +75
in all seriousness, that comment requires some next level dechypering.

"I am not sure intel are ahead, I think as AMD like to do keep back on new launches just to ensure they don't lose the race... stratergy"

Besides the grammar, what exactly did you mean with "to do keep back on launches" - did you try to say that AMD likes to hold back new technologies for future products?
You win champ, ill leave you with that, you win, you win, Winner No.1