Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads

[Tigerick]

As Hot Chips 34 starting this week, Intel will unveil technical information of upcoming Meteor Lake (MTL) and Arrow Lake (ARL), new generation platform after Raptor Lake. Both MTL and ARL represent new direction which Intel will move to multiple chiplets and combine as one SoC platform.

MTL also represents new compute tile that based on Intel 4 process which is based on EUV lithography, a first from Intel. Intel expects to ship MTL mobile SoC in 2023.

ARL will come after MTL so Intel should be shipping it in 2024, that is what Intel roadmap is telling us. ARL compute tile will be manufactured by Intel 20A process, a first from Intel to use GAA transistors called RibbonFET.



Comparison of upcoming Intel's U-series CPU: Core Ultra 100U, Lunar Lake and Panther Lake

Model	Code-Name	Date	TDP	Node	Tiles	Main Tile	CPU	LP E-Core	LLC	GPU	Xe-cores
Core Ultra 100U	Meteor Lake	Q4 2023	15 - 57 W	Intel 4 + N5 + N6	4	tCPU	2P + 8E	2	12 MB	Intel Graphics	4
?	Lunar Lake	Q4 2024	17 - 30 W	N3B + N6	2	CPU + GPU & IMC	4P + 4E	0	8 MB	Arc	8
?	Panther Lake	Q1 2026 ?	?	Intel 18A + N3E	3	CPU + MC	4P + 8E	4	?	Arc	12

Comparison of die size of Each Tile of Meteor Lake, Arrow Lake, Lunar Lake and Panther Lake

	Meteor Lake	Arrow Lake (20A)	Arrow Lake (N3B)	Arrow Lake Refresh (N3B)	Lunar Lake	Panther Lake
Platform	Mobile H/U Only	Desktop Only	Desktop & Mobile H&HX	Desktop Only	Mobile U Only	Mobile H
Process Node	Intel 4	Intel 20A	TSMC N3B	TSMC N3B	TSMC N3B	Intel 18A
Date	Q4 2023	Q1 2025 ?	Desktop-Q4-2024 H&HX-Q1-2025	Q4 2025 ?	Q4 2024	Q1 2026 ?
Full Die	6P + 8P	6P + 8E ?	8P + 16E	8P + 32E	4P + 4E	4P + 8E
LLC	24 MB	24 MB ?	36 MB ?	?	8 MB	?
tCPU	66.48
tGPU	44.45
SoC	96.77
IOE	44.45
Total	252.15

Intel Core Ultra 100 - Meteor Lake



As mentioned by Tomshardware, TSMC will manufacture the I/O, SoC, and GPU tiles. That means Intel will manufacture only the CPU and Foveros tiles. (Notably, Intel calls the I/O tile an 'I/O Expander,' hence the IOE moniker.)

 

[SiliconFly]

...fmax has been increased...

20A's rumored Fmax was around 5.2GHz - 5.4GHz. So, did you mean 20A's Fmax is expected to be beyond that? Sounds a bit unlikely, as 20A is a brand new node with too many new technologies (like GAAFET & Backside Power Delivery). My guess is, Intel will need a miracle for 20A to touch 5.5GHz or 6GHz.

 

[dullard]

Not maybe, he does make stuff up. That's the problem with having a YT show built around leaks, his audience expects him to deliver on time & every time, so he ALWAYS has something for any upcoming product from Intel, AMD, Nvidia. If some bird chirps in his ear it's all good, if not then he'll make do with the wind.

He does make stuff up. But he also ALWAYS indicates what stuff is likely real and what is just wild rumor. To pretend that his stuff is all real or all made up is disingenuous. We can do better than that. Just report it as he does: likely true or likely rumor.

Too many people here just want to deny anything he says, even the true stuff, just because some stuff is wild rumor. I worry that you fall into that category. It would be the same to just dismiss all of this forum because some posts are rumor.

 

[eek2121]

Dayum.
Meanwhile Apple P-cores just use 1/3 of that power.

You are overstating things. The M1 Max, for example, can use up to 15W for 1T workloads.

 

[Schmide]

Let me rephrase. Intel uses foveros which has chiplets on top of an interposer connected using the advanced NOC interconnect fabric. This is far ahead of AMD's dated chiplet design, where the chiplets reside on the substrate and are connected using the very dated infinity fabric.

Like I said, Zen5 is not gonna catch up with ARL in packaging tech. ARL's foveros & NOC interconnect are far too advanced. ARL's only weak point may turn out to be LNC and may lose to Zen5 in ST or MT. Or maybe not. Hard to guess at this time.

Catch up? I don't think that phrase cuts the mustard. Let us look at the bread and butter.

https://chipsandcheese.com/2023/09/23/intels-ponte-vecchio-chiplets-gone-crazy/

It's has potential, but it's not going to leap frog refined technology on it's first release. In some ways it seems like they bit off more than they can chew.

 

[coercitiv]

I worry that you fall into that category.

You know what, you're right. I should be more open minded. I'll just open his video about Meteor Lake.

 

[ondma]

You know what, you're right. I should be more open minded. I'll just open his video about Meteor Lake.

View attachment 93726

Clickbait video. ML may be disappointing, but it is hardly "Bulldozer".

 

[Thibsie]

Clickbait video. ML may be disappointing, but it is hardly "Bulldozer".

Well, that's MLID isn't it ?
Well...

 

[dullard]

You know what, you're right. I should be more open minded. I'll just open his video about Meteor Lake.

View attachment 93726

Any particular spot you want us to watch? We aren't going to do your bidding and just spend eternities watching various videos. Please guide us.

This might help explain my point:

 

[H433x0n]

20A's rumored Fmax was around 5.2GHz - 5.4GHz. So, did you mean 20A's Fmax is expected to be beyond that? Sounds a bit unlikely, as 20A is a brand new node with too many new technologies (like GAAFET & Backside Power Delivery). My guess is, Intel will need a miracle for 20A to touch 5.5GHz or 6GHz.

I wish it was primarily Intel silicon but the bulk of ARL is on TSMC N3B. The top ARL SKU on N3B looks like it should be able to reach ~5.4ghz boost clocks.

On its own terms (relative to Golden Cove era) it looks like it’ll do okay.

 

[coercitiv]

We aren't going to do your bidding and just spend eternities watching various videos. Please guide us.

I didn't ask you to watch anything.

 

[Saylick]

Let me rephrase. Intel uses foveros which has chiplets on top of an interposer connected using the advanced NOC interconnect fabric. This is far ahead of AMD's dated chiplet design, where the chiplets reside on the substrate and are connected using the very dated infinity fabric.

Like I said, Zen5 is not gonna catch up with ARL in packaging tech. ARL's foveros & NOC interconnect are far too advanced. ARL's only weak point may turn out to be LNC and may lose to Zen5 in ST or MT. Or maybe not. Hard to guess at this time.

If Zen 5 loses to ARL in performance or perf/W, it won't be because of advanced packaging or the lack thereof imo. Ultimately, AMD's choice in using a substrate based SerDes interconnect was based on engineering judgement, not a lack of imagination. I'm sure if AMD determined advanced packaging was needed, nothing would prevent them from doing so, especially since TSMC is the leader in it. If anything, the fact that Zen 2 to Zen 5 all successfully use the same interconnection tells you how little it holds back their design in the grand scheme of things.

 

[Glo.]

If Zen 5 loses to ARL in performance or perf/W, it won't be because of advanced packaging or the lack thereof imo. Ultimately, AMD's choice in using a substrate based SerDes interconnect was based on engineering judgement, not a lack of imagination. I'm sure if AMD determined advanced packaging was needed, nothing would prevent them from doing so, especially since TSMC is the leader in it. If anything, the fact that Zen 2 to Zen 5 all successfully use the same interconnection tells you how little it holds back their design in the grand scheme of things.

Zen 5 will not lose to Intel in anything, apart from perf/$.

EOT.

 

[dullard]

I didn't ask you to watch anything.

I see, you just like posting here what you watch. I've been watching a lot of Taskmaster recently. Videos about canyoneering too. I watch bouldering competitions while I'm on the treadmill.

I'm interested in CPUs though, especially in this thread. If you have something relevant to say about CPUs, please do so next time. I'm capable of discussing both likely specs and wild rumors about specs.

Anyone know how Fab 9 opening up affects this discussion of packaging tech (supposedly that fab should allow Intel to finally mass produce Foveros and other packaging technologies)? Fab 9 supposedly recently opened up, but I don't know if that means it is in production or just ramping towards production. Will that give Intel the ability to mass produce Meteor Lake or is it for the the upcoming tiled CPUs? I assume Intel's foundry event next week will go into more details.

 

[coercitiv]

I'm capable of discussing both likely specs and wild rumors about specs.

Considering your poor attempt at being passive-aggressive you're hardly capable of discussing anything. The video I gave as example was from MLID and also discussing Meteor Lake, so exactly on topic. It's a clickbait video, showcasing how little MLID cares about juggling with half truths to attract views. (unless you agree with him that Intel is deceiving consumers)

Meanwhile you're trashing this thread with videos about fan noise and weird stories about your life hobbies. All because someone disagrees with you.

 

[dullard]

All because someone disagrees with you.

Does someone disagree with me? All you've posted so far is "You know what, you're right." and something that you call clickbait that you don't ask us to watch. That is pretty odd that you'd post a link for people not to click on it. But you do you.

MLID posts truths, half-truths, and far-fetched rumors. He marks each one with his interpretation of their reliability. That is far better than (A) people who ignore the fact that something they post is a far-fetched rumor and (B) waiting until the product launches--at least I find it better since it is interesting to discuss.

Here is an example of MLID's key for his Meteor Lake posts:


Do you disagree with me that the items in different shades of colors have different levels of believability?

 

[coercitiv]

Do you disagree with me that the items in different shades of colors have different levels of believability?

Can you map "truths, half-truths, and far-fetched rumors" to "very high confidence, high confidence and mostly confident"?

 

[dullard]

Can you map "truths, half-truths, and far-fetched rumors" to "very high confidence, high confidence and mostly confident"?

Here is MLID's interpretation:


Is he 100% correct in all times? Certainly not. Is he 100% incorrect like you and a few others have implied many times? Certainly not. But we can still discuss his posts without pure outright dismissal of them (well at least some of us can).

Here are other examples of MLID's key where he describes them in more detail. The items in white are not confirmed 100% truths (also marked as estimations or informed opinions).

 

[Hitman928]

Here is MLID's interpretation:
View attachment 93739

Is he 100% correct in all times? Certainly not. Is he 100% incorrect like you and a few others have implied many times? Certainly not. But we can still discuss his posts without pure outright dismissal of them (well at least some of us can).

Here are other examples of MLID's key where he describes them in more detail. The items in white are not confirmed 100% truths (also marked as estimations or informed opinions).
View attachment 93741View attachment 93742

MLID has had many very high confidence or 100% sure "leaks" be wrong. Problem is that he deletes those videos when they get proven wrong so unless someone is archiving all his stuff (why would you?), you can't go back to show how wrong he's been. Even people on Reddit have figured out not to give him any legitimacy.





I will say that he's actually had some good guests on in the past but I realized pretty quickly that he was clueless in what he was trying to talk about and stopped watching his videos so I don't know if he's been able to keep up the interesting guest segments or not.

 

[Thunder 57]

MLID is a clickbait asshat.

 

[Hitman928]

With the release of Hawk Point, Notebookcheck ran some tests on the new AMD chip against their latest MTL numbers. I'm posting a chart here where they re-did their Cinebench r23 scores over power consumption like they did at MTL's release. I'm only including the Intel MTL vs RPL numbers to compare the latest versus release numbers. Overall it's basically the same as the release situation. The release tests did 28 W - 45 W whereas this is a higher range, but overall the results are basically the same. The 155H is roughly 5 - 10% more efficient than RPL depending on what power level you are running at. So whatever updates have come with the latest laptops/firmwares, it hasn't really effected MTL's positioning in compute performance / efficiency according to this test.

 

[igor_kavinski]

Intel 4 was supposed to be 20% more power efficient? Does that mean MTL has pretty leaky compute dies?

 

[DavidC1]

Intel 4 was supposed to be 20% more power efficient? Does that mean MTL has pretty leaky compute dies?

It actually reminds me of Sapphire Rapids where it has too many tiles and they improved significantly by moving to two tiles from four with Emerald Rapids and made quite a decent jump. Even if Meteorlake is better on some implementations, needing post-launch firmware update is proof that something went wrong. It should have been easy enough so everything was there before it even had a chance to be in the hands of testers.

This does not bring confidence with Arrowlake, where it'll use the same many tile approach. Lunarlake on the other hand, cuts it down.

Whether Foveros has an advantage over AMD's approach or not, implementation trumps theory like it always has been.

 

[Tigerick]

It actually reminds me of Sapphire Rapids where it has too many tiles and they improved significantly by moving to two tiles from four with Emerald Rapids and made quite a decent jump. Even if Meteorlake is better on some implementations, needing post-launch firmware update is proof that something went wrong. It should have been easy enough so everything was there before it even had a chance to be in the hands of testers.

This does not bring confidence with Arrowlake, where it'll use the same many tile approach. Lunarlake on the other hand, cuts it down.

Whether Foveros has an advantage over AMD's approach or not, implementation trumps theory like it always has been.

That's pretty objective view of points, unlike someone who keeps praising something irrelevant to the SoC like LPE & Feveros.

4-tile SoC reminds me of old time with four separate ICs; namely CPU, north bridge, GPU and display output. We are back in times, it is all due to IFS incapable of making monolithic SoC. That's why memory latency is real, that's why there is LPE try to cut down power consumption. I have given up hope on ARL due to four tiles solution.

LNL is departure from 4 tiles but too bad it is one time thing and they are available for 4P+4E only.

 

[Abwx]

Intel 4 was supposed to be 20% more power efficient? Does that mean MTL has pretty leaky compute dies?

That s about right, if you look at MTL at 35W then RPL use about 20% more power for the same throughput.
20% better efficency at same throughput amount to about 8% better perf at same power, that is, you trade the better efficency at isoperfs for better perfs at isopower.

 

[dullard]

MLID has had many very high confidence or 100% sure "leaks" be wrong. Problem is that he deletes those videos when they get proven wrong so unless someone is archiving all his stuff (why would you?), you can't go back to show how wrong he's been. Even people on Reddit have figured out not to give him any legitimacy.

I will say that he's actually had some good guests on in the past but I realized pretty quickly that he was clueless in what he was trying to talk about and stopped watching his videos so I don't know if he's been able to keep up the interesting guest segments or not.

Yes, many people have a hard time understanding when someone says there is not high confidence of something that it might not be true--Even those on Reddit. I've seen lists like that plenty of times here. He isn't perfect. But it is ignorant to ignore everything he posts because some things haven't come true. There are saved images of his leaks, no need to store videos. Feel free to give those saved images a proper review if you want (which if you want to exclude someone you should do a proper tally). A lot of his leaks do come true. A lot do not. Below are some of his Meteor Lake images.

Is he correct in everything? Heck no. But 2 and 3 years before the Meteor Lake launch, he actually got some important details right, even in his speculation white font.
Nov 2020:

June 2021:

June 2022:

Nov 2022:

Dec 2022:

May 2023:
June 2023: