E-cores still evolve. But is there a reason for it?

[wheresmycar]

EH cores? E for Effort Cores? Effortless cores?
Maybe nicer names instead!
16 Elegant cores!!
Enchanting Cores!!
Euphoric Cores!! (My personal Favorite)

Sounds like someone's asking for a HUG lol

here u go (... but terrified at the same time)

 

[DavidC1]

Why did Intel abandon HT (which I don't mind at all and it's not to be discussed in this thread) and not E-cores since they already implemented segmental layout? Is there anything real engineers can see going wrong that I don't? Once again, if it's all only limited to cash and marketing then I don't even know what to say.

You are thinking technicals is the issue, which is why it makes no sense. The problem(and the solution) is always the people. Bad management, disgruntled employees, infighting within the organization, etc.

The P core team is in shambles right now, so they could not have HT in Lion Cove, since HT is easy in terms of die resources but difficult in terms of validation, as everything needs to work across two threads without contention, without errors in execution, and now we have to worry about security too.

Since Intel is such a big company, they have parallel design teams, and they happen to have the E core team, which has been executing, far, far better. Thus in the recent years with Tremont, and Gracemont, they caught up significantly. Since the main product line were all P cores and falling behind competition, they decided to include E cores.

That's why we're here today.

That's definitely a statement.

Would've loved to read the thought process behind that, preferrably sans baseless speculations.

Tremont E core came out in 2020, and it performed way, way behind Sunny Cove core in Icelake. Tremont was Ivy Bridge level. Then Gracemont came out and was almost Skylake level in 2021, and Golden Cove was 19% better than Sunny Cove. Golden Cove was 40% faster than Gracemont per clock. Skymont is 2024, and Lion Cove is mere 10-15% ahead of Skymont per clock.

E cores:
Tremont 2020: Ivy Bridge(2014)
Gracemont 2021, 30% faster per clock than Tremont: Skylake(2016)
Skymont 2024, 30% faster per clock than Tremont, 70% faster in FP: Golden Cove(2021)

P cores:
Sunny Cove 2019, 18% faster than Skylake
Golden Cove 2021, 19% faster than Sunny Cove
Lion Cove 2024, 9% faster than Golden Cove

Let me repeat, Skymont is 1/3rd the size of Lion Cove P core, despite being just 10-15% difference in performance per clock. What do you think the next iteration of E cores will do against the next P cores? E cores to 30% every 2 years, versus under 20% for the P cores, sometimes way under during the same timeframe.

 

[Tralalak]

Where is Skymont Enhanced PE-Cores (Performance-Efficiency cores for gaming) with L3 (L4) cache (X3D) 96MB+ ?​

There are rumors of 12C Bartlett Lake built on the LGA 1700 platform only in large cores, but I'm surprised they didn't also introduce 8c Skymont (2x 4c cluster) or 12c Skymont (3x 4c cluster) or 16c Skymont (4x 4c cluster) with L3 (L4) cache (X3D) 96MB and more.


• We found the missing Intel Performance in Games... Big Performance Gains!
source:

• Why are Intel processors starting to lag in gaming? Analysis of the reasons for the regression of the game Core Ultra (from 03:41 is there compare Intel Core Ultra 9 285K 8P + 16E vs 1P + 16E etc.).
source: https://www.bilibili.com/video/BV1tW14YGEmz/


• Intel's Arrow Lake chips aren't winning any awards for gaming performance but I think its new E-cores deserve a gold star
source: https://www.pcgamer.com/hardware/pr...-i-think-its-new-e-cores-deserve-a-gold-star/


• 12c .LITTLE Atom Skymont E-cores beats 8C big Lion Cove P-cores !
source: https://www.techpowerup.com/forums/threads/share-your-cpuz-benchmarks.216765/post-5360549


• Skymont E-cores 3.2GHz up to 4.6GHz Max Turbo Frequency v Cinebench R20 podáva single-thread výkon procesora AMD Ryzen 9 5950X 3.4GHz up to 4.9GHz Max. Boost Clock.
source: https://forum.pctuning.cz/viewtopic.php?p=9907684#p9907684


• In Raytracing-Benchmarks V-Ray Version 6 - Skymont E-core loses 4% to P-core Golden Cove from Alder Lake (and loses 6% to P-Core Raptor Cove from Raptor Lake) but gains up to 62% to Gracemont E -core.
source: https://www.pcgameshardware.de/Core...elease-Specs-Preis-Benchmarks-Review-1458051/

 

[TechBuyingHavoc]

You are thinking technicals is the issue, which is why it makes no sense. The problem(and the solution) is always the people. Bad management, disgruntled employees, infighting within the organization, etc.

The P core team is in shambles right now, so they could not have HT in Lion Cove, since HT is easy in terms of die resources but difficult in terms of validation, as everything needs to work across two threads without contention, without errors in execution, and now we have to worry about security too.

Since Intel is such a big company, they have parallel design teams, and they happen to have the E core team, which has been executing, far, far better. Thus in the recent years with Tremont, and Gracemont, they caught up significantly. Since the main product line were all P cores and falling behind competition, they decided to include E cores.

Great insights. Too many people believe that technical performance is the reason for why a company fails or succeeds but most of the factors are internal to the organization and not visible to non-employees.

I can point to any number of bone-headed, short-sighted, or myopic decisions by the senior management at EVERY company I have ever worked at that has lead to products never appearing on the market at all (most customers never realize this), products underperforming vs expectations (customers are aware but not sure why the underperformance is happening), or the company shrinking slowly over time.

Intel was planting the seeds of its decline since 2012 if you assume a minimum of 3 years for CPU development and Skywell being disappointing. The stock price was up 35% in Nov 2015 vs the middle of 2012. Customers were happy with Intel in 2012. Yet the massive success of Sandy Bridge had started the rot at Intel.

 

[Vya Domus]

What do you think the next iteration of E cores will do against the next P cores? E cores to 30% every 2 years, versus under 20% for the P cores, sometimes way under during the same timeframe.

You understand if you keep increasing the performance of an E core you eventually end up with a very wide P core, right ? This isn't magic, E cores are just built upon regular underclocked cores from years ago which are now much smaller due to node shrinks.

 

[Tralalak]

RUMOURS? from https://x.com/chickenonthepan

source:

https://twitter.com/i/web/status/1851828922310336518

source:

https://twitter.com/i/web/status/1852504317707862365

These are interesting rumors even in the context of the discussion here.

 

[Vayra86]

And I feel like weather on Mars reflects the value of USD a little bit too dishonestly.
I want facts, not feelings.

Feelings are generally the breeding ground for future facts.

I'm also quite convinced that the evolution of the E core is Intel betting on two horses. Either they get a good big little design that keeps the P cores in play without being surpassed left and right by competition (which they are, today, by quite a margin - they need more power, and they're larger, and they don't scale), which is what we've seen up till today, and its clearly NOT working - or Intel moves more and more into E cores and lands at a new, smaller core that can do as much or more as a P core.

Its simple logic. Any other route will leave Intel playing with way too much risk. If anything, I think the 285k release and its results have underlined and given more credibility to the continued development of E cores and slowly removing P cores.

You understand if you keep increasing the performance of an E core you eventually end up with a very wide P core, right ? This isn't magic, E cores are just built upon regular underclocked cores from years ago which are now much smaller due to node shrinks.

Of course not, the development process to get there is completely different, so it opens up new approaches. Its not magic, but it does offer a kind of clean slate.

 

[Vya Domus]

the development process to get there is completely different

It just isn't, the way to obtain more performance is the same. Performance does not scale linearly with size, that's why the P cores seem so much more size inefficient but in reality there is nothing special going on.

It's the same story with how people were impressed by Apple's CPUs and wondered how they were so fast without been clocked very high but in reality the cores were giant, wider than even desktop x86 processors, they weren't doing anything unheard of, it's just a bog standard "trade area for speed or efficiency" approach.

 

[dont whant to set it"']

More or less the same Skylake cores that got DR. Frankensteined.

 

[Vayra86]

It just isn't, the way to obtain more performance is the same. Performance does not scale linearly with size, that's why the P cores seem so much more size inefficient but in reality there is nothing special going on.

It's the same story with how people were impressed by Apple's CPUs and wondered how they were so fast without been clocked very high but in reality the cores were giant, wider than even desktop x86 processors, they weren't doing anything unheard of, it's just a bog standard "trade area for speed or efficiency" approach.

Sure. But in the end it is net performance at X power and Y surface area. The P core is doing a specific balance, the e core does some things differently. I do think it allows intel to create a new thing without throwing away the old. The P cores are based on old principles.

 

[DavidC1]

It's the same story with how people were impressed by Apple's CPUs and wondered how they were so fast without been clocked very high but in reality the cores were giant, wider than even desktop x86 processors, they weren't doing anything unheard of, it's just a bog standard "trade area for speed or efficiency" approach.

You are missing the point. The latest M4 cores are faster than the 5.7GHz Zen 5 and Arrowlake desktop chips when it's clocked at just 4.5GHz. And the cores pretty small at that performance. Nevermind the much, much lower power usage.

You understand if you keep increasing the performance of an E core you eventually end up with a very wide P core, right ? This isn't magic, E cores are just built upon regular underclocked cores from years ago which are now much smaller due to node shrinks.

Sounds like you have a lot to learn.

P cores are 3x the size of an E core. P core team and design just sucks. It needs to be retired.

Mind you, 21% gain over the predecessor is with the hobbled 194ns memory latency.

 

[Vya Domus]

P cores are 3x the size of an E core. P core team and design just sucks. It needs to be retired.

Did you not read the portion of my comment where I explained performance does not scale linearly with size.

Do you believe an E core the size of a P core would just magically be faster ? It would make no sense, the more Intel will improve the performance of E cores the more area and power inefficient they'll get. And by the way E cores still have horrible ST performance, they only redeem themselves in MT, ditching P cores in favor of E cores would be down right idiotic.

 

[Tralalak]

Rumours - Naming of the successor to Arctic Wolf is Golden Eagle.

Griffin a mythical creature with the head and wings of an eagle and the body of a lion, typically depicted with pointed ears and with the eagle's legs taking the place of the forelegs.

Intel's current internal roadmap (plan):
• 2024: Lion Cove: Arrow Lake / Lunar Lake
• 2025: Cougar Cove: Panther Lake
• 2026: Coyote Cove: Nova Lake
• 2027: Griffin Cove
• 2028: "Griffin Cove-next"

The king is dead, long live the king! => Griffin Cove is a "corefusion" E-core and P-core because is a mythical creature known as a half-eagle, half-lion.

 

[JustBenching]

Let's pretend marketing, business and all that economy stuff are completely irrelevant. I'm about to ONLY talk engineering aspects of this phenomenon.

From what I've gathered so far (and I might be totally wrong. Correct me if I am):
• Hybrid structure cries for an impeccable prediction mechanism which can never be invented. At least with our current state of knowledge.
• E-cores are mocked by last gen architectures in gaming even if the game is coded so well E-cores actually improve the experience in all aspects.
• E-cores are mocked by P-cores in terms of performance per watt if you downclock the latters to around 4.3 (Alder Lake) or 4.6 (Raptor Lake) GHz.
• Software development is currently in a state that promotes fast releases but doesn't tolerate actual bug fixing if it takes more than a manhour to deploy. Which means scheduling is virtually thrown outta window.
• It's not impossible to land 16ish properly working P-cores on one die and make them feel at home, likely cutting about a half or two GHz all-core turbo so it actually doesn't go kaboom.
• Average Joes and Janes (and attack helicopters for that matter, too) don't have any idea what these cores are actually good at. They render confused at best.
• There's no evidence that heterogenous architecture helps alleviating background loads any better than just throwing more P-cores.
• It seems it's also more complex and failure prone than a good ol' technique of just having X cores of the same arch.

Why did Intel abandon HT (which I don't mind at all and it's not to be discussed in this thread) and not E-cores since they already implemented segmental layout? Is there anything real engineers can see going wrong that I don't? Once again, if it's all only limited to cash and marketing then I don't even know what to say.

Think about what CPUs would look like 20 years from now. The goal would be to have a CPU akin to the T1000 from the terminator series. Changing it's shape to fit the circumstances. Meaning, combining all the transistors into a big single CORE for single threaded workloads, then dividing into a thousand pieces for MT workloads etc. Sadly we can't do that with current technology, so Ecores are kinda an inbetween solution. The ecores being very space efficient allows Pcores to be big and space inefficient to achieve higher ST performance.

Ecores are in fact mocked by Pcores in performance per watt, but that's an unfair comparison to begin with. Since 1 P core is as big as 3.5 ecores, 3.5 ecores mock 1P core in performance / watt.

To achieve the maximum amount of both ST and MT performance out of a given transistor count, you need to combine small space efficient cores for the MT workloads and big space inefficient cores for the ST / lightly threaded workloads.

You are missing the point. The latest M4 cores are faster than the 5.7GHz Zen 5 and Arrowlake desktop chips when it's clocked at just 4.5GHz. And the cores pretty small at that performance. Nevermind the much, much lower power usage.

Because an m4 cores is more than 50% larger in transistor count. So - to make it a fair comparison, you need to compare 1.5 zen 5 core vs 1 m4 core, and I can tell you the m4 core ain't winning in either performance or efficiency.

EG1. Some actual numbers taken from measuring die shots. An m3 core is 57% larger in transistor count than an Intel Pcore, while it leads by around 15% in performance at the same power. So, I don't know, having to use 60% more transistor for a 15% performance increase ain't what I'd bother writing home about.

 

[FoulOnWhite]

Think about what CPUs would look like 20 years from now. The goal would be to have a CPU akin to the T1000 from the terminator series. Changing it's shape to fit the circumstances. Meaning, combining all the transistors into a big single CORE for single threaded workloads, then dividing into a thousand pieces for MT workloads etc. Sadly we can't do that with current technology, so Ecores are kinda an inbetween solution. The ecores being very space efficient allows Pcores to be big and space inefficient to achieve higher ST performance.

Ecores are in fact mocked by Pcores in performance per watt, but that's an unfair comparison to begin with. Since 1 P core is as big as 3.5 ecores, 3.5 ecores mock 1P core in performance / watt.

To achieve the maximum amount of both ST and MT performance out of a given transistor count, you need to combine small space efficient cores for the MT workloads and big space inefficient cores for the ST / lightly threaded workloads.


Because an m4 cores is more than 50% larger in transistor count. So - to make it a fair comparison, you need to compare 1.5 zen 5 core vs 1 m4 core, and I can tell you the m4 core ain't winning in either performance or efficiency.

EG1. Some actual numbers taken from measuring die shots. An m3 core is 57% larger in transistor count than an Intel Pcore, while it leads by around 15% in performance at the same power. So, I don't know, having to use 60% more transistor for a 15% performance increase ain't what I'd bother writing home about.

Imagine if the Intel and Apple had the same transistor count per core, i bet it would be an interesting comparison then. Not really a fair comparison as it is when the Apple has a MUCH bigger transistor count per core.

 

[Tralalak]

Intel Core Ultra Series 2 Skymont E-cores Shatter Expectations

In this video we’re joined by Greg Boots, Sr. Product Platform Manager at Intel, who shares data highlighting the incredible power and potential of the Skymo...

youtube.com