News Intel Rebranding its Chips With Meteor Lake: Core Ultra 5 Spotted

[bit_user]

If you look at the Meteor Lake image on this story, that CPU tile has enough space for a 0P/18e (0P/20e max) arrangement but not any way to get to 18 physical cores if there are any P cores.

As mentioned a few times, already, the Meteor Lake SoC tile has 2 E-cores. So, that means there are only 16 threads to account for. If the max configuration is 6 + 8, the most logical explanations are that they either disabled 2 P cores or a quad-core E cluster. That yields 16 threads, in either case.

Too many processes depend on having good single-thread performance and a full instruction set to have zero fully-featured cores in your CPU.

I'm not sure what you mean by "fully-featured cores". The E-cores support the exact same instructions as the P-cores.

 

[hotaru.hino]

Intel slapped 'i' in front of numbers long before that though, like the i740 and I'm almost certain I've seen much earlier Intel chips with iXXX model numbers on them.

I think they only did that when they found out numbers were hard to trademark while looking for some schnazzy marketable name.

 

[bit_user]

Intel slapped 'i' in front of numbers long before that though, like the i740 and I'm almost certain I've seen much earlier Intel chips with iXXX model numbers on them.

Oh, you're right. Even back in the late 80's they had the i860. And I also saw the i486 nomenclature, which I think came about when they lost a court case trying to defend a copyright on just "486" or "80486". It came out that you cannot copyright a number. That surely factored into their decision to go with Pentium, instead of 586.

 

[hotaru.hino]

Oh, you're right. Even back in the late 80's they had the i860. And I also saw the i486 nomenclature, which I think came about when they lost a court case trying to defend a copyright on just "486" or "80486". It came out that you cannot copyright a number. That surely factored into their decision to go with Pentium, instead of 586.

You can trademark numbers, such as "007" in the context of film for James Bond and I read there was a NASCAR driver who trademarked the number 1 for purposes of his branding.

It's just that Intel's model numbering thing became under the "generic trademark" label so fast that they couldn't really enforce it.

 

[Sippincider]

With the mention of Pro and Max, the author is clearly suggesting Intel might have a case of Apple-envy. Could be.

Slapping ”Pro“ etc. on everything is one part of Apple the industry should NOT copy.

As for Intel: just show me things that matter, like how hot these chips are going to run. Don’t try to change the subject with marketing…

 

[bit_user]

As for Intel: just show me things that matter, like how hot these chips are going to run. Don’t try to change the subject with marketing…

Arguably, the entire job of marketing is selling products beyond how well they'd do on the basis of their price and merits, alone.

Part of the role good reviewers play is to cut through the marketing BS, and actually asses products on their merits.

 

[TerryLaze]

Oh, you're right. Even back in the late 80's they had the i860. And I also saw the i486 nomenclature, which I think came about when they lost a court case trying to defend a copyright on just "486" or "80486". It came out that you cannot copyright a number. That surely factored into their decision to go with Pentium, instead of 586.

I wouldn't put it past them if the i stands for intel, so intel core intel 5...

As for Intel: just show me things that matter, like how hot these chips are going to run. Don’t try to change the subject with marketing…

They get so hot that you can cool them with a $20 cooler and still get pretty much all of the performance that intel promises.
245 out of 253W which is getting you 94% of the cinebench score of the liquid cooler DeepCool LT720 that is using up 315W

Core i9-13900K Cooling Guide: Testing Intel's Flagship With Budget Air and Big AIOs

The Core i9-13900K can push even the most capable liquid coolers to their limits, but you don’t lose that much performance if you go with budget-priced air cooling.

www.tomshardware.com

The $20 Assassin 120 R SE sustained 5055MHz (an increase of 333MHz) with the CPU consuming an average of 245W.

Finally, when testing with DeepCool’s LT720 liquid cooler, clock speeds increased by another 219MHz. At 5462, this is just 38MHz shy of the maximum clock speed held during multicore workloads. This in turn led to a benchmark score result that’s 1059 points better than the AG620 air cooler. And those high clock speeds don’t come easy in terms of efficiency; power consumption here averaged 315W.

 

[bit_user]

They get so hot that you can cool them with a $20 cooler and still get pretty much all of the performance that intel promises.

No, the subject is Meteor Lake. Different cores on a different manufacturing node. Don't try to pass off anything about Raptor Lake, here.

 

[TerryLaze]

No, the subject is Meteor Lake. Different cores on a different manufacturing node. Don't try to pass off anything about Raptor Lake, here.

Let's wait and see, if intel goes for triangles on their heat sink to one-up ryzens design with the squares then they might be terrible to cool.

 

[oofdragon]

Intel core ultra 😂😂😂

 

[InvalidError]

Intel core ultra 😂😂😂

Lets try throwing every superlative in Intel's current marketing book at this... Intel Core Ultra Pro Max Platinum!

 

[cyrusfox]

I want to praise apple for keeping it simple
M1, M2...
But in the same breath I realize they have the useless adjectives at the end as well (Max, Pro, & Ultra...)

Intel's Core branding was nearly perfect: simple to start and largely told you what performance class you were receiving, granted differences between i5/i7 have always been minute. Core was such good branding that AMD direct copied it (R3/R5/R7/R9). Intel muddied it up starting with Ice Lake, return to decent nomenclature for alder lake. As long as they don't require a decoder ring I will keep hope this isn't going to be a disaster.

If the replacement is more complex, more letters or syllables than the original, hard fail.

 

[bit_user]

Lets try throwing every superlative in Intel's current marketing book at this... Intel Core Ultra Pro Max Platinum!

Hey now! Let's not start with the precious metals!

That was a truly sad day, IMO. For Intel to market server processors using precious metals showed they had entirely stopped being an engineering company selling chips to other engineers.

Intel's Core branding was nearly perfect:

I'd argue it would've been better if they'd dropped the word "Core". That people nearly always do that, in normal conversation, shows how unnecessary it ever was.

At least Core i# was better than the seemingly-redundant "Core 2 Duo". What a train wreck of a name that was...

 

[InvalidError]

At least Core i# was better than the seemingly-redundant "Core 2 Duo". What a train wreck of a name that was...

It isn't redundant: Core 2 merely stood for the 2nd generation of the Core series which was available in Core 2 Solo for single-core, Core 2 Duo for dual-cores and Core 2 Quad for quad-cores models.

 

[TerryLaze]

seemingly = seems like

 

[Giroro]

I'm not sure what you mean by "fully-featured cores". The E-cores support the exact same instructions as the P-cores.

If an E core has exactly the same features as a P core, then why do P cores exist?

As mentioned a few times, already, the Meteor Lake SoC tile has 2 E-cores. So, that means there are only 16 threads to account for. If the max configuration is 6 + 8, the most logical explanations are that they either disabled 2 P cores or a quad-core E cluster. That yields 16 threads, in either case.


I'm not sure what you mean by "fully-featured cores". The E-cores support the exact same instructions as the P-cores.

The pictured tile I'm talking about is clearly 4P + 4e not 6 + 8.
Adding +2 to the possible configurations still doesn't get that to 18 cores in a nice way

 

[InvalidError]

If an E core has exactly the same features as a P core, then why do P cores exist?

Supporting the same instruction set so any thread can run on any core isn't the same as making all cores the same. The E-cores don't have SMT, have smaller caches and a couple of other design simplifications to make the cores smaller and more power-efficient.

 

[TerryLaze]

Supporting the same instruction set so any thread can run on any core isn't the same as making all cores the same. The E-cores don't have SMT, have smaller caches and a couple of other design simplifications to make the cores smaller and more power-efficient.

Smaller and use less power, but if it is more power efficient will depend on what you run on it, not having SMT and having a smaller cache will influence the performance per wat you can get for a lot of apps.

 

[InvalidError]

Smaller and use less power, but if it is more power efficient will depend on what you run on it, not having SMT and having a smaller cache will influence the performance per wat you can get for a lot of apps.

The only time where E-cores are less power-efficient is poorly or lightly threaded workloads where most of the extra cores are being wasted. If all code could thread well, we wouldn't need P-cores.

 

[TerryLaze]

The only time where E-cores are less power-efficient is poorly or lightly threaded workloads where most of the extra cores are being wasted. If all code could thread well, we wouldn't need P-cores.

Lower number = more efficient.
Completing a cinebench run the e-cores use 20% more energy making them 20% LESS efficient.

Intel Core i9-12900K E-Cores Only Performance Review

With Alder Lake, Intel is betting big on hybrid CPU core configurations. The Core i9-12900K has eight P(erformance) cores and eight E(fficient) cores. We were curious and tested the processor running the E-Cores only to see how well they perform against architectures like Zen 2, Zen 3, Skylake...

www.techpowerup.com

We've also added an additional set of results to put the "E-cores-only" results into perspective (which are 8x Gracemont at 3.9 GHz). The "8 P-cores" result imitates that: 8x Golden Cove cores at 3.9 GHz; HyperThreading is disabled, clocked down to match the frequency of the E-cores—this is an apples-to-apples comparison.

 

[InvalidError]

Lower number = more efficient.
Completing a cinebench run the e-cores use 20% more energy making them 20% LESS efficient.

Not really a fair representation of how efficient E-cores can actually be since they are being handicapped by all of the overkill infrastructure overhead required to feed P-cores. A chip designed strictly with E-cores for peak efficiency would have at least twice as many E-cores to offset the lower throughput per core and make better use of the otherwise overkill platform overhead.

 

[bit_user]

It isn't redundant:

That's why I used the word "seemingly". It's just needlessly awkward. It's like tripping over your own feet. Clumsy, in a word.

Core 2 merely stood for the 2nd generation of the Core series which was available in Core 2 Solo for single-core, Core 2 Duo for dual-cores and Core 2 Quad for quad-cores models.

Do you really think that either I don't know that or that the average consumer did?

Back when Core 2 Duo launched, most people hadn't noticed the Core Solo & Core Duo, because they were laptop-only CPUs. Moreover, the Core 2 Quad didn't yet exist and was initially a somewhat niche product until the 45 nm generation.

The word "Core" is bad, because it confuses whether the number following it specifies the number of cores or the generation. Further, using sequential numbering and an ordinal word is potentially confusing about which means what.

If you heard the phrase "Core 2 Duo", for the very first time, and you had no reason to think the speaker knew what they were talking about, you might reasonably presume they meant "Two core duo". That shows just how problematic their usage of the word "Core" is.

I'm convinced that marketing people sometimes hold the view that awkward and confusing names are sometimes good. They count on people to invest in understanding the branding as a means of spreading it. If you have to explain it, that gives it free publicity, and then others feel "in the know". It's probably not too similar to the idea that people have better retention and comprehension of text written in bad handwriting or a less legible font. However, what makes an effective branding strategy can clearly deviate from what makes a sensible one.

 

[bit_user]

The only time where E-cores are less power-efficient is poorly or lightly threaded workloads ...

No, I doubt there's truly a good case where they're less power-efficient. Maybe if you've got a thread on P-core that's waiting in a spinlock, but then that's just bad code. From the link in @TerryLaze 's post, they actually do quite well in single-threaded efficiency:

​

​

The best I can come up with is something that puts them at a disadvantage due to their cache or prefetcher. That's my best guess regarding the Cinebench score in @TerryLaze 's post. Recall that each quad of E-cores is sharing the same L2 cache slice + ring bus stop. In other words, it's a macro-architecture deficiency - not a microarchitecture one.

 

[bit_user]

A chip designed strictly with E-cores for peak efficiency would have at least twice as many E-cores to offset the lower throughput per core

An E-core in Alder Lake performed 54% to 60% as well as a single thread on a P-core, while using only about 30% of the area and about 20% of the power.

 

[TerryLaze]

Not really a fair representation of how efficient E-cores can actually be since they are being handicapped by all of the overkill infrastructure overhead required to feed P-cores. A chip designed strictly with E-cores for peak efficiency would have at least twice as many E-cores to offset the lower throughput per core and make better use of the otherwise overkill platform overhead.

All of that overkill infrastructure overhead required to feed the P-cores is also active when the P-cores do their run...and then they are actually all active, while when the e-cores run at least some of it might be in a sleep state.