Some Alder Lake SoCs won't have performance cores

[mongeese]

Recap: At the top end, Alder Lake processors have eight performance cores and eight efficiency cores. As you descend the ranks, you slowly lose e-cores until you get to the i5-12600, which has none. At the bottom of the pile, the Pentium and Celeron parts have just two p-cores.

According to an update to a Linux repository that Intel made themselves, found by Japanese outlet Coelacanth Dream, the Alder Lake-based Atom processors will reverse the trend and have up to eight e-cores and no p-cores.

In the update, which contains a partial boot log of an Alder Lake-N part, Intel describes the Atom processors as having two quad-core clusters of e-cores with 2 MB of L2 cache each and a shared L3 cache of an undisclosed size. They also have a 32 EU GPU, the same as the i5-12600.

These Atom processors will likely end up in SoCs for business applications and as embedded processors in OEM systems, like Chromebooks.

Intel says that e-cores perform about the same as Skylake cores. In our in-game testing, we found that claim to be slightly disingenuous: in some titles, that was true, but in most, they were much worse. However, the discrepancy wasn’t caused solely by the e-cores’ worse throughput — it was primarily due to their poor inter-core latency.

Though games are susceptible to latency, not all applications are. In multi-threaded applications like Blender, Intel’s claim was approximately true. To draw a hasty conclusion, then, an octa-core e-core processor could be well suited to SoC and embedded applications and shouldn’t be dismissed.

	Alder Lake-N*	Celeron G6900
P-cores / threads	-	2 / 4
E-cores / threads	8 / 8	-
L2 cache	2 + 2 MB	2.5 MB
L3 cache	?	4 MB
GPU	32 EU	16 EU
TDP	?	46 W

*Specifications are unconfirmed.

In any case, in comparing an Atom processor with eight e-cores and a Celeron with only two p-cores (although they do have hyperthreading, so four threads), the Atom couldn’t perform all that much worse, plus would at least consume less power and take up far less space. Win-win.

Image credit: Fritzchens Fritz

Permalink to story.

https://www.techspot.com/news/93379-alder-lake-socs-wont-have-performance-cores.html

 

[Dimitriid]

Genuinely interesting for the super power efficient market like you find on the single boards which as you might know, are mostly dominated by ARM chips since you want devices you can basically run on USB power, Power-Over-Ethernet or even batteries only.

So I'd be interesting to see how close intel can get to ARM chips in performance now that they would be aiming for a far more similar power usage target: 8 p cores vs 8 arm cores. While I suspect ARM cores will probably pull ahead, the fact that p-cores are still x86 cores might give them an edge when it comes to software support: People that run some of this appliances would love to have x86 far wider software support because even though arm support is not bad, it's just not as good once you want to move to more advanced/niche stuff like playing around with containers for example.

Also important is how aggressively intel prices their p-core only chips.

 

[Sausagemeat]

Hopefully they will add more types of core so that it’s not consistent or easy to follow and then users will finally stop caring about core counts!

A 5600X = a 3700X at multithreading. A 6 core is the same as an 8 core basically and this is just a modern example. Intels 12100F is a quad core that performs roughly the same on cinebench r23 multithreaded as an R7 1700X which has twice the core count.

Core count is just pushed by marketing departments, it really doesn’t mean much.

 

[zamroni111]

It's indication that intel's fake 7nm doesn't have great yield rate.

 

[Sausagemeat]

It's indication that intel's fake 7nm doesn't have great yield rate.

Intels process for this is named 10nm. And prices indicate that there appears to be more availability for Intels stock than Ryzen 5000. Right now Alder lake is cheaper, faster and more in stock.

 

[Lew Zealand]

"Intel says that e-cores perform about the same as Skylake cores. In our in-game testing, we found that claim to be slightly disingenuous: in some titles, that was true, but in most, they were much worse."

...with a 6900XT at 1080p. But with a GPU price-matched to 8x E-cores like a 3050 or 6600m, they could do very well.

 

[Bullwinkle M]

Intels process for this is named 10nm. And prices indicate that there appears to be more availability for Intels stock than Ryzen 5000. Right now Alder lake is cheaper, faster and more in stock.

Intel's "real" 10nm also has more transistors per given area than AMD's 7nm

Can't wait for Intel's "real" 7nm

 

[ilovelinux]

Soon, they will have pay-to-use P-cores.

 

[Sausagemeat]

Intel's "real" 10nm also has more transistors per given area than AMD's 7nm

Can't wait for Intel's "real" 7nm

It’s also not AMDs 7nm. It’s TSMC’s. AMD would be up sh*t creek without a paddle if it wasn’t for TSMC.

 

[Dimitrios]

Without Microsoft's updated threading schedule for Intel's new architecture they would be up sh*t creek too.

But with me being honest I would still lose against your comment since AMD would be worst off than Intel no matter what while picking straws. ;-)

 

[flee2020]

Correction:
The Celeron G6900 is a two core two thread processor. It does not support hyperthreading. Perhaps the article meant Pentium Gold G7400?

 

[umbala]

Intel keeps churning out more and more garbage and all the Intel fanboys are just eating it up. It's hilarious. Why give people 16 core parts where all 16 are high performance when you can just give them 8 high performance and 8 crap cores and call it innovation? Better yet, forget the high performance cores and just start selling chips with only crap cores!

Fact is, Intel can't compete with ARM when it comes to efficiency and they can't compete with AMD when it comes to high performance (unless they push their chips to like 5GHz and 200+watts). So yeah Intel, keep churning out these Frankenstein chips and pretend you're doing great and leading the pack! Hahahaha.

 

[umbala]

Hopefully they will add more types of core so that it’s not consistent or easy to follow and then users will finally stop caring about core counts!

This wins hands down the dumbest comment of the week award. People should ABSOLUTELY care about core counts, well if the cores are actually good, not the garbage Intel is cranking out lately. They can't compete so your solution is they should release lots of confusing parts so people will stop caring about core counts? Stupid is as stupid does, right Forrest?

 

[defaultluser]

Don't they already sell this model as the Pentium -n and Celeron-n series?

https://www.anandtech.com/show/16388/intel-launches-jasper-lake-tremont-atom-cores-for-all

I'm sure they were already planning on releasing a Gracemont update, but its about time they doubled the core count of these cheap systems!

And no, the Celeron desktop series is still suck in dual-thread land (like it has been for the last fifteen years!

 

[HardReset]

It’s also not AMDs 7nm. It’s TSMC’s. AMD would be up sh*t creek without a paddle if it wasn’t for TSMC.

No. Without TSMC 7nm AMD would have even better GlobalFoundries 7nm.

Don't they already sell this model as the Pentium -n and Celeron-n series?

https://www.anandtech.com/show/16388/intel-launches-jasper-lake-tremont-atom-cores-for-all

I'm sure they were already planning on releasing a Gracemont update, but its about time they doubled the core count of these cheap systems!

And no, the Celeron desktop series is still suck in dual-thread land (like it has been for the last fifteen years!

No, Tremont is not Alder Lake in any way. Tremont is (probably last) of "ordinary" Atom cores that are likely be replaced by Alder Lake (or future) E-cores.

 

[Nintenboy01]

Didn't GF elect to stay at 12nm though?

 

[Fulljack]

No. Without TSMC 7nm AMD would have even better GlobalFoundries 7nm.

cool ideas, but until this day, GloFo hasn't licensed nor developed any 7nm process just yet. so how is it better exactly?

 

[Sausagemeat]

Uh-huh, and guess what, Intel is tapping TSMC too for for their high end chips! If Intel is so great and so ahead of everybody, why can't they manufacture their own chips? Also, Intel *STILL* cannot make a FULLY hyperthreaded 16 core part now, after several years of trying to figure out 10nm chips. Why is that exactly? The best they can do is 8 good cores + 8 crap cores. You fanboys always crack up though, so keep flapping your gums about nonsense.

Lol, look at you rushing to AMDs defence like an extremely valiant and loyal fan, although I wonder why, AMD doesnt care about you, just your money.

I don’t disagree, Intel can’t make a “true” 16 core chip, I don’t think that’s a big problem, the vast majority of users don’t need a 16 core chip and in most use cases Intel chips are both faster and cheaper than AMDs at this current point in time. Also yes Intel are using TSMC, I believe it’s for their GPUs and the chips currently on sale from Intel are not made at TSMC.

Oh and nothing I have said is “nonsense”, please be a little more respectful, I understand you are Emotionally attached to AMD and you are triggered when you hear the fact that Intel
is currently faster and cheaper but there’s no need to get personal. My advice, relax, I have no doubt that AMD will beat Intel again and when they do you can celebrate!

P.S. my comment about core counts isn’t dumb it’s the truth, I’m sorry if you lack the capacity to grasp the points I made.

 

[Sausagemeat]

No. Without TSMC 7nm AMD would have even better GlobalFoundries 7nm.

Global Foundries 7nm doesn’t exist. That’s one of the reasons why we have a chip shortage.

 

[Stardude82]

"Intel says that e-cores perform about the same as Skylake cores. In our in-game testing, we found that claim to be slightly disingenuous: in some titles, that was true, but in most, they were much worse."

...with a 6900XT at 1080p. But with a GPU price-matched to 8x E-cores like a 3050 or 6600m, they could do very well.

Right, really it's performance/Watt. The article doesn't address that or performance per clock.

Or maybe they'll just be used in non-gaming laptops. Oops, the article only looked at games, so there's no indication at all if the chip is a turkey for webrowesing.

 

[Stardude82]

Global Foundries 7nm doesn’t exist. That’s one of the reasons why we have a chip shortage.

Until recently, it was zombie looking to be bought out. Pile Driver really screwed them over as it cut into the demand of their main customer and they couldn't afford the next process node. Now, things are actually looking up for them for making automotive chips.

 

[HardReset]

Didn't GF elect to stay at 12nm though?

They still make 14nm and 12nm stuff yes.

cool ideas, but until this day, GloFo hasn't licensed nor developed any 7nm process just yet. so how is it better exactly?

Global Foundries 7nm doesn’t exist. That’s one of the reasons why we have a chip shortage.

Read again: "Without TSMC 7nm AMD would have even better GlobalFoundries 7nm."

GlobalFoundries had 7nm around six to nine months from mass production phase when they decided to pull plug. One of main reasons (perhaps even biggest) for that was TSMC's 7nm capacity was so large. GF even made their process dimensions quite close what TSMC had so customers could easily change from TSMC to GF and vice versa. However GF's process was designed to reach higher clock speeds than TSMC's process, making it more suitable for many AMD products.

GF surely considered (among everything else) that TSMC's 7nm capacity + GF's 7nm capacity would be near oversupply. And applying that to future ultra-expensive 5nm and 3nm nodes, it was worth abandoning. Despite all development costs (quite huge) already spent.

Yeah, GF clearly underestimated 7nm demand. If they would have continued 7nm development into production phase, there would be much less chip shortage etc. But how many semiconductor companies correctly predicted demand that Covid caused? None?

But basically, without TSMC's 7nm node, AMD would have used GF 7nm node. And surely would not have been abandoned.

 

[Mr Majestyk]

Hopefully they will add more types of core so that it’s not consistent or easy to follow and then users will finally stop caring about core counts!

A 5600X = a 3700X at multithreading. A 6 core is the same as an 8 core basically and this is just a modern example. Intels 12100F is a quad core that performs roughly the same on cinebench r23 multithreaded as an R7 1700X which has twice the core count.

Core count is just pushed by marketing departments, it really doesn’t mean much.

In some cases yes the 5600X is as good, but there are plenty of cases where the 3700X's 8 cores is beneficial and comparing apples to apples 5800X clearly beats 5600X in all multithreading tests. 8 != 6.

 

[LuxZg]

Is this the true return of Atom? Could we expect Windows tablets again that don't have 5 year old CPUs under the hood? While idea of e-core only ADL CPU is laughable at first, I really do hope something comes out of it all. I would really like some tablets and small R-Pi like SBCs with tech that isn't ancient, something comparable to current ARM offerings. x86 is still the king of my hill

 

[Sausagemeat]

Right, really it's performance/Watt. The article doesn't address that or performance per clock.

Or maybe they'll just be used in non-gaming laptops. Oops, the article only looked at games, so there's no indication at all if the chip is a turkey for webrowesing.

They still make 14nm and 12nm stuff yes.


Read again: "Without TSMC 7nm AMD would have even better GlobalFoundries 7nm."

GlobalFoundries had 7nm around six to nine months from mass production phase when they decided to pull plug. One of main reasons (perhaps even biggest) for that was TSMC's 7nm capacity was so large. GF even made their process dimensions quite close what TSMC had so customers could easily change from TSMC to GF and vice versa. However GF's process was designed to reach higher clock speeds than TSMC's process, making it more suitable for many AMD products.

GF surely considered (among everything else) that TSMC's 7nm capacity + GF's 7nm capacity would be near oversupply. And applying that to future ultra-expensive 5nm and 3nm nodes, it was worth abandoning. Despite all development costs (quite huge) already spent.

Yeah, GF clearly underestimated 7nm demand. If they would have continued 7nm development into production phase, there would be much less chip shortage etc. But how many semiconductor companies correctly predicted demand that Covid caused? None?

But basically, without TSMC's 7nm node, AMD would have used GF 7nm node. And surely would not have been abandoned.

You are talking nonsense. GF 7nm does not exist and would not have existed if AMD didn’t choose TSMC. TSMC saved AMDs bacon big time.