Intel Core i9-12900K E-Cores Only Performance

[qubit]

I would much prefer Intel to have gone all P core like previous architectures. 16 performance cores would have had so much more performance. If AMD can do it, then why not Intel? Still, it's interesting to see what can be got out of this CPU in this mode.

 

[RandallFlagg]

I would much prefer Intel to have gone all P core like previous architectures. 16 performance cores would have had so much more performance. If AMD can do it, then why not Intel? Still, it's interesting to see what can be got out of this CPU in this mode.

Because at heart, Alder Lake's architecture is designed to be a laptop killer, not a desktop killer. Laptops make up 85% of the client market.

 

[ARF]

I would much prefer Intel to have gone all P core like previous architectures. 16 performance cores would have had so much more performance. If AMD can do it, then why not Intel? Still, it's interesting to see what can be got out of this CPU in this mode.

Because profit margins. Intel simply found a way to extract more "marketing performance" while keeping the die size relatively small.
The largest Alder Lake is in reality a 10-core P-core equivalent. If four E-core complex is as large as one P-core.

A 16-core P-core chip would be quite a bit larger, so the profit margins would be lower.

Also, Intel's 10nm process is definitely not in the best shape or form, so the terrifying power consumption is a consequence.
Alder Lake should have been built on a full-node shrink to be competitive in the 65-95W power envelopes.

 

[W1zzard]

Did you mean 128-bit memory interface? It's only 160-bit when counting ECC bits, which aren't being used anyway and I don't even think the memory controller supports it.

Yeah that's fair reasoning, fixed

 

[Richards]

Because at heart, Alder Lake's architecture is designed to be a laptop killer, not a desktop killer. Laptops make up 85% of the client market.

Raptor lake will increase it to another level... 8+16 crazy performance is coming

 

[Bomby569]

basically a ryzen 5 3600, i'm impressed

 

[defaultluser]

Yes, that's very interesting, and contrary to what I predicted. E-core perf/W will probably be better in the smaller ADL dies, because of less cache, but we won't see those anytime soon.

E-cores still have a solid advantage in perf/mm² so they are not total nonsense, or purely a marketing trick.

It would be interesting to have at least some benchmarks with P-cores only but with HT enabled. Maybe there's a second Wiz2ard around to do all that.

Well anyone that concerned about v\cloud density would be better suited porting their software to Arm Neoverse

 

[bug]

I would much prefer Intel to have gone all P core like previous architectures. 16 performance cores would have had so much more performance. If AMD can do it, then why not Intel? Still, it's interesting to see what can be got out of this CPU in this mode.

Intel doesn't do chiplets (yet), the yields for 16 P cores would have been terrible.

 

[Wirko]

The largest Alder Lake is in reality a 10-core P-core equivalent.

Performance wise, it's anywhere between 5 and 12 P-core equivalent. I'm sure the scheduler can learn new tricks over time, we're just not there yet.

 

[WhoDecidedThat]

The comments about transcoding while gaming (i.e. recording / streaming) with AL E-cores seems spot on. However, it also seems to need DDR5 to effectively both game and transcode.

When you have that combo of AL+DDR5 and in that use case, the results are pretty stunning :

View attachment 225948


View attachment 225949

I wonder how well 5950X would do with DDR5.

So according to Intel, a P core will either be

- more efficient than E core
- more performant than E core

Very interesting. I would love to see a 2 P cores vs 4 E cores comparison though. 4 threads v/s 4 threads.

 

[TheinsanegamerN]

Yeah that would be great!
So we could find the point of diminishing returns so once you go over x amount of watts performance improvement is negligable, especially for gaming

It would be cool to see a graph of, say, 3.5 GHz max to full boost max, in maybe 100 or 200 mhz intervals. Not the entire test bench, maybe just 1 power hungry CPU stress test and 1 game to show power scaling.

 

[qubit]

Intel doesn't do chiplets (yet), the yields for 16 P cores would have been terrible.

I’d love to see 16 P cores in an optimised monolithic design. Shame it’s not gonna happen on a consumer CPU.

 

[ARF]

Intel doesn't do chiplets (yet), the yields for 16 P cores would have been terrible.

This is false.
The die size of one P-core is approximately 14 square millimeters, so 16 of them would result in total of 234 square millimetres.
Intel's job then is to cut the useless for the top variant iGPU.

The current top variant Alder Lake 8/16+8+iGPU is 209 sq. mm.

 

[GURU7OF9]

basically a ryzen 5 3600, i'm impressed

Well yeah and nah. Some things its very good at others not as good.
It is 8 cores verses 6 cores. Its 8 threads verses 12 threads. Not sure how much of a difference those extra 4 threads make ?
For gaming its not as good.
If you could have the best of both worlds, the e cores could run at 5ghz and keep a modest power draw and you could junk the p cores altogether.
Just run 16 e cores @ 5gz .
Actually if you junked the p cores totally, you would have enough room to run 32 e cores ( at 5ghz) and fill the gap with the biggest IGP available.
Now that would be a weapon!

 

[TheinsanegamerN]

This is false.
The die size of one P-core is approximately 14 square millimeters, so 16 of them would result in total of 234 square millimetres.
Intel's job then is to cut the useless for the top variant iGPU.

The current top variant Alder Lake 8/16+8+iGPU is 209 sq. mm.

They tried that, it was called HDET, and nobody bought it. I dont know why you guys keep insisting that intel do what they already tried.

 

[bug]

They tried that, it was called HDET, and nobody bought it. I dont know why you guys keep insisting that intel do what they already tried.

Because Intel is evil and, obviously, some random guy posting on TPU is smarter than all Intel employees put together?

 

[regs]

Can't wait for Raptor Lake which doubles E-core counts and improves the E-cores IPC.

Raptor Lake will have Gracemont logic. So IPC won't change. Clocks are the question though.

 

[Minus Infinity]

Raptor Lake will have Gracemont logic. So IPC won't change. Clocks are the question though.

It has already been reported the Gracemont cores will be getting improvements in performance, so I take that to mean IPC uplift, unless the refined node just means higher clocks.

 

[Bomby569]

Well yeah and nah. Some things its very good at others not as good.
It is 8 cores verses 6 cores. Its 8 threads verses 12 threads. Not sure how much of a difference those extra 4 threads make ?
For gaming its not as good.
If you could have the best of both worlds, the e cores could run at 5ghz and keep a modest power draw and you could junk the p cores altogether.
Just run 16 e cores @ 5gz .
Actually if you junked the p cores totally, you would have enough room to run 32 e cores ( at 5ghz) and fill the gap with the biggest IGP available.
Now that would be a weapon!

You're just making something so simple and increase the complexity level to 100. Those E cores perform like a ryzen 5 3600, that's it.

I think that's amazing and something no one would expect, even more when they aren't clocked that high and i don't know if many of those games scale that well in all 8 cores

 

[InVasMani]

E cores works out to about 18% worse power draw for 47.8% higher multi thread performance per die space area roughly based on w1zzard's charts at least on the power draw and performance. I think that's pretty damn good in the context of things. Intel could've made a 5P 20E chip and would have been insanely powerful at multi-thread performance relative to dollar cost. The power drawn would've gone up a bit further, but you'd have more E core clusters for individual E core multipliers instead of 2 it would be 5 different clusters of E cores and multipliers that you could scale while the P cores could still be scaled individually.

You'll get a cleaner sweep of frequency scaling to power efficiency with more clusters of E cores with multipliers that can be finely adjusted. I can't say how they'll be stock, but as a consumer you'll have the option to tweak them to taste and price will be the only relevant concern in regard to that. Right now you can only adjust 2 individual E core clusters multipliers, but they could easily have as many as 5 in future iteration of the same design. I have to wonder if maybe they should've gone with a 2:1 ratio on the E cores, but it doesn't matter at this stage, but if they insert a mid core it should be 2:1 and maybe 1HT as a in between the P and E cores.

At that point they could rename them H/M/L cores for high/mid/low where high core is 1C 2HT, mid 2C 1HT, and low is 4C 0HT and frequency scales downward linearly along with some of the instruction set differences. The advantage of the mid core is you could fit two of those in place of a low core and thus double the CPU multiplier granularity further to fine tune efficiency. The most power hungry cores would have the highest granularity control over CPU multipliers, but worst efficiency relative to die space occupied when pushed due to a higher peak frequency however they would have the most optimal efficiency when reduced to the base frequency of the low core chip die's.

There is actually room for Intel to put 1P and surround it with 8E cores and have room left over for 1P give or take die space for EDRAM/HBM or to expand upon the already present iGPU in place with the 12900K. Honestly arranging it in such a way would make sense put the P core in the center and 8E cores around it with their own multipliers. The E cores could power and temperature regulate clock wise around the P core so it could turbo constantly at a good frequency while running cool at the same time. You'd also have a ton of multi-threaded performance and if you needed even more drop the P core down to the base frequency of the E cores and overclock those a nudge.

 

[RandallFlagg]

Raptor Lake will have Gracemont logic. So IPC won't change. Clocks are the question though.

Raptor Cove cores replace Golden Cove : +IPC, +200Mhz boost, larger L2 cache, DLVR for higher performance/lower power, faster official DDR5 support (5600), and more E cores - at least that is the current leak.

The efficiency cores (Gracemont) will be the same uArch, just more of them.

Intel Raptor Lake's Digital Linear Voltage Regulator (DLVR) could reduce CPU power up to 25% - VideoCardz.com

Intel DLVR can provide up to 7% more performance with 21% lower voltage An interesting thread has recently appeared on Reddit. It covers the concept of the Digital Linear Voltage Regulator (DLVR) which might be used with the upcoming Intel Raptor Lake CPU series. Intel Raptor Lake Digital...

videocardz.com

Intel Raptor Lake-S Desktop CPU Roadmap & Full Lineup Leaks Out, Core i9 With Up To 24, Core i7 With Up To 16 Core on Raptor Cove Architecture

Intel's Raptor Lake-S Desktop CPUs 13th Gen Core family & feature up to cores based on the Raptor Cove core architecture have leaked.

wccftech.com

 

[Mussels]

What are you talking about? AL is at the top in efficiency. Only the 5900x & 5950x are significantly better.

https://tpucdn.com/review/intel-core-i9-12900k-e-cores-only-performance/images/efficiency-multithread.png

They give the performance of a 3600X, while using more power.


These E-cores that specifically exist exclusively for one goal of power efficiency, are outdone by last gen budget products by the competition in both power consumption and performance.

I like the concept. I like the goal.


But these are still power hungry monsters, and do not achieve what they pretend to be... they're just a method to pad the core count and multi threading results, without needing 500W CPU's.

They are neither low power, high performance, or energy efficient.

 

[Solid State Brain]

In a different Alder Lake thread a graph was linked where a user from a German forum tested the 12900K at various power limit levels and found that (at least in the test chosen) at a 50% power limit performance was 90% of the starting level, so I wonder if to efficiently operate as intended E- and P- cores have to work with power cap rather than in an almost power-unlimited fashion.

Under power-limited conditions the way the CPU throttles E- or P- cores might not necessarily be straightforward as in standard cases, which could be where at least some efficiency gains are "hiding".

Alder Lake CPUs common discussion

For the time being, tight timing DDR4 out performs DDR5 in 99% of use-case-scenario's. That will of course change but not for another 18 to 24 months. Until then, DDR4 with tight timings is the smart choice. No. Just no.

www.techpowerup.com

 

[Bomby569]

They give the performance of a 3600X, while using more power.


These E-cores that specifically exist exclusively for one goal of power efficiency, are outdone by last gen budget products by the competition in both power consumption and performance.

I like the concept. I like the goal.

But these are still power hungry monsters, and do not achieve what they pretend to be... they're just a method to pad the core count and multi threading results, without needing 500W CPU's.

They are neither low power, high performance, or energy efficient.

I think their purpose is to help the power hungry P cores, and they manage to do that.
They do have more cores then the 3600x so depending on the tasks they could be a better fit, i think it would be interesting what Intel would charge for just E cores, a celeron type of thing for 100$ or something like that, it that is possible.

I don't think we will see any super power draw budget cores unless someone does what Apple did. Power keeps going up.

 

[bug]

They give the performance of a 3600X, while using more power.


These E-cores that specifically exist exclusively for one goal of power efficiency, are outdone by last gen budget products by the competition in both power consumption and performance.


View attachment 226154
View attachment 226155


I like the concept. I like the goal.


But these are still power hungry monsters, and do not achieve what they pretend to be... they're just a method to pad the core count and multi threading results, without needing 500W CPU's.

They are neither low power, high performance, or energy efficient.

View attachment 226156

It's really hard to extrapolate from all that. E cores are not meant to run Cinebench and whatnot. Supposedly the scheduler should be smart enough to sent them background tasks and other light loads (wip - the scheduler is not smart enough yet). Running the wrong type of workload could result in worse energy efficiency.
On top of that, we don't know if Intel actually tuned E cores for efficiency on the desktop (where it means little), mobile chips is what we should be looking at. Whenever they may show up.

Energy aside, I think it's revealing how you can get half the performance with 1/16th of the die area.