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

Tigerick · Aug 22, 2022

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	12 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)	Lunar Lake	Panther Lake
Platform	Mobile H/U Only	Desktop Only	Desktop & Mobile H&HX	Mobile U Only	Mobile H
Process Node	Intel 4	Intel 20A	TSMC N3B	TSMC N3B	Intel 18A
Date	Q4 2023	Q1 2025 ?	Desktop-Q4-2024 H&HX-Q1-2025	Q4 2024	Q1 2026 ?
Full Die	6P + 8P	6P + 8E ?	8P + 16E	4P + 4E	4P + 8E
LLC	24 MB	24 MB ?	36 MB ?	12 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.)

TESKATLIPOKA · Thursday at 1:50 PM

adroc_thurston said:
That's PTL-H.

I know, but I don't think It will be that efficient.

adroc_thurston said:
There isn't an S14 Zenbook with Strix.

But there is S16 Zenbook. The same thing just in a bigger chassis.

adroc_thurston · Thursday at 1:54 PM

TESKATLIPOKA said:
I don't think It will be that efficient.

It inherits the uncore but not the other things. So less, but still.

TESKATLIPOKA said:
just in a bigger chassis.

Yes, "things that cost more money, part 243".

cannedlake240 · Thursday at 1:54 PM

adroc_thurston said:
money-spending options

So apple just bought their way to fame \s

adroc_thurston · Thursday at 1:56 PM

cannedlake240 said:
So apple just bought their way to fame \s

Well, yeah. But they sell exclusively premium products so BOM increments impact them a lot less.

Nothingness · Thursday at 1:59 PM

AMDK11 said:
How did you come to the conclusion that I think so? I have never written this anywhere. I had in mind the discussion with David C1, who starts a given topic and then gives his arguments in a convoluted way, which are not the answer. That's all. He sees Zen5 as a revolution (it is a revolution), which I don't deny, but in LionCove he sees only expansion. Discussion with him is biased, non-technical and non-substantive.

My apologies, I misunderstood you 😀

The predictor in Zen5 is an idea from 30 years ago, but only now its technical implementation makes sense (transistor packing and complexity)

The TAGE predictor was first used by Intel in Haswell, the first working sample of which was presented in 2011.

TAGE was presented around 2005. And as far I know AMD has been using it since Zen2.

Haswell might have used it earlier because Seznec worked for them for some years; but the initial publication made it clear the algorithm would not be patented and could be freely used (it was a rule of the bpred tournament), so everyone could use it

Zen5 uses other old algorithms beyond TAGE, as you pointed out. Most modern high perf CPU have several algorithms running in parallel using more or less complex functions to choose the better one.

As I often say, branch prediction is often the most complex part of a CPU (along with data prefetch).

AMDK11 · Thursday at 2:16 PM

An interesting fact about LionCove in LunarLake is that it can predict 3-4 branches per cycle compared to 2-3 in GoldenCove, but apparently it is inferior to GoldenCove in some respects. We'll see what LionCove will show at ArrowLake. Intel claims that the PBU block is 8x larger than the one in RedwoodCove, so this is an interesting thing to analyze after the launch of ArrowLake.

TESKATLIPOKA · Thursday at 2:47 PM

Abwx said:
LNL has same MT throuhgput and efficency as a 6C/12T Phoenix 2 7540U, i would expect in a 1400€ laptop better MT perf and perf/Watt than laptops sold at 600-650€,
more than 2x the price just because of the GPU.?.

That s the magic of the marketing, actual numbers are not even considered, to the point that a CPU that should be in the sub 1000€ laptop market is accepted as being valuable at 1400€ if not 2000€.

Thanks for once again not answering my original question.

Laptops are not just a CPU(SoC, APU), yet you are only comparing It based on that and only in full nT on top of that, knowing full well how laptop prices don't reflect actual performance difference.

I am looking at that 7540U in a Lenovo Yoga 6 Slim and It costs 769€ and with only 16GB RAM.
BTW, that 7540U is also in a 1549 € business laptop, will you complain about It too?

What 2x price just because of the iGPU?
7540U's iGPU is useless, but even the CPU is weaker If you compare It at the same thread count.
What you are doing is comparing 8 threads vs 12 threads and complain how weak LNL is.
It's hilarious, because If I remember correctly, you were complaining how It was unfair comparing 8C Zen4 to 14C Meteor Lake, because MTL had an advantage in core count and thread count.

TESKATLIPOKA · Thursday at 2:50 PM

adroc_thurston said:
Yes, "things that cost more money, part 243".

Yet the smaller chassis is sold for more in most cases.

MoistOintment · Thursday at 2:52 PM

Abwx said:
LNL has same MT throuhgput and efficency as a 6C/12T Phoenix 2 7540U, i would expect in a 1400€ laptop better MT perf and perf/Watt than laptops sold at 600-650€,
more than 2x the price at same MT perf just because of the GPU.?.

That s the magic of the marketing, actual numbers are not even considered, to the point that a CPU that should be in the sub 1000€ laptop market is accepted as being valuable at 1400€ if not 2000€.

And Phoenix loses in ST.

Why are we sitting here pretending that nT benchmarks are the most important metric when shopping for a thin and light? My laptop use case, order of importance is:

ST > Battery Life in my workloads (mostly web apps and Office Suite) > weight > thickness > keyboard & trackpad quality > screen quality > webcam and microphone quality > speaker quality > then nT performance.

If your order of importance is place nT higher up the list (or at the top) then ARL or Strix is a better choice for you.

Realistically, there's 3 types of laptop buyers:

A) Performance is the most important metric. Once a sufficient level of battery has been reached, the more performance the better. Might not even care about battery life at all.

B) Battery life is the most important metric. Once a sufficient level of performance has been reached, more battery is always better. Workloads are fixed, lightweight tasks like Office and Web.

C) A decent balance of both. Wants a jack of all trade CPU that isn't necessarily the best in battery or performance, but strikes a good balance.

I am in group B. Apple M3 and LNL target group B.

You seem to be in either group A or C. Strix and ARL-H will target group C.
Strix Halo and ARL-HX will target group A.

I think it's important that we at least recognize that these 3 groups of buyers exist, and that the group we belong to isn't the only one. Then maybe we can stop this "LNL sucks because max performance R24 MT score divided by power drawn is the most important metric" discourse.

I mean, these LNL laptops, which supposedly suck in nT, outperform my desktop 3700X in nT by a decent margin and in ST by a large margin. And I don't feel nT constrained at all on my desktop and don't even plan to upgrade until Probably Zen 6 / Nova Lake.

TESKATLIPOKA · Thursday at 2:57 PM

MoistOintment said:
Why are we sitting here pretending that nT benchmarks are the most important metric when shopping for a thin and light?

Because @Abwx must compare in a metric where AMD wins significantly and that one metric becomes the most important for him.

adroc_thurston · Thursday at 3:11 PM

TESKATLIPOKA said:
Yet the smaller chassis is sold for more in most cases.

No?

MoistOintment said:
Strix Halo and ARL-HX will target group A.

Halo is a very own thing.
That's not how laptop segments are sliced anyway.

Magio · Thursday at 3:14 PM

Abwx said:
LNL has same MT throuhgput and efficency as a 6C/12T Phoenix 2 7540U, i would expect in a 1400€ laptop better MT perf and perf/Watt than laptops sold at 600-650€,
more than 2x the price at same MT perf just because of the GPU.?.

That s the magic of the marketing, actual numbers are not even considered, to the point that a CPU that should be in the sub 1000€ laptop market is accepted as being valuable at 1400€ if not 2000€.

Is there a reason besides thread derailment you're avoiding talking about GPU perf or ST perf or battery life thanks to the low power E cores or the other aspects that make LNL the best x86 platform yet for thin and lights?

Of course its low thread count disqualifies it from certain use cases, but it is absolutely leading edge in the aspects it targets.

SiliconFly · Thursday at 3:19 PM

branch_suggestion said:
... i7 is about on par with vanilla N7. ...

Wrong.

N7 is a crappy node compared to Intel 7. Intel 7 is far ahead. Plus, it offers better performance than N7.

Company	Node	HD MTr/mm2	HP MTr/mm2	Metal Pitch	Gate Pitch (CPP)
Intel	7	100.76	67.2	36	54
TSMC	7	91.2	65	40	54

branch_suggestion said:
No, in most metrics it (Intel 4) is just behind vanilla N5 ,

Wrong again.

N5 is a lot worse than Intel 4. According to Anandtech, Semiwiki & Wikichip, Intel 4 is closer to N3 than N5. To be precise, it's in between N4 & N3.

For example, the main HP library density of N5 is only 92.7, but Intel 4 is ~124. Intel 4 is far ahead of N5 and a lot closer to N3.

branch_suggestion said:
18A looks to be around N3P, which is bad considering GAA/PowerVia vs FinFET.

Wrong again.

18A looks to be a lot better than TSMC N3 and very close to N2. Also, 18A is expected to have higher performance than N2 because of BSPDN (higher Fmax than N2, cos N2 doesn't have BSPDN).

branch_suggestion said:
14A is the one that is supposed to beat A16 which is the all in TSMC GAA/BSPDN.

*If* Intel can up the game with High-NA *on time*, they can brutally kick TSMC's bottom with 14A, considering they have more High-NA machines in the pipeline than all other foundries put together.

Technology-wise, Intel has caught up TSMC on most of the fronts. But IFS is still a financial disaster thats threatening the whole ship. Imho, Intel should let go of IFS the first chance they get.

branch_suggestion said:
Kraken Point, the step down from Strix.

Why bother about a silly step-down product when Strix itself is getting hammered this gen?

511 said:
14 is High NA with DSA ( Directed Self assembly) it's a big gamble as well

I think DSA is still more of a rumor as of now.

MoistOintment · Thursday at 3:29 PM

adroc_thurston said:
No?

Halo is a very own thing.
That's not how laptop segments are sliced anyway.

Halo is up to 16 performance cores with chiplets. Definitely sacrifices battery life to get even more performance over standard Strix.

And of course that's how laptop SoCs are segmented. Just that Apple with the launch of M1 showed how much "battery life, ST, and good iGPU / accelerators are the most important to me" customers there are, and not everyone wants to sacrifice light task efficiency for more cores. There's a reason why we see the market settle around a U, H, and HX class. Different products for different customers. Most people look at this and see different brackets for different needs, rather than viewing them as bad, better, best.

SiliconFly · Thursday at 3:33 PM

Abwx said:
Strix is of another class at +60% better perfs at same power, so why should it be priced the same.?

LNL is lacking in the CPU part, in MT that s 6800U level of perf and perf/Watt,
and compared to a 8840U it s worse at 30W but also at 15W, so the low power branding was just marketing as it is used at 30W anyway in all configurations, wich make things even worse.

It would have been good in 2021 before Phoenix was released but at current competition it will be soon exhausted as a fast osoleting part, hstoricaly for laptops the more the throughput the longer the usability over time as generaly there s nothing that could be upgraded in those items.

MT is not that important compared to ST in thin and lights. Apple has already shown that to the world with its Apple Silicon. Lunar Lake is just following the same path. AMD not so much.

AMD Strix is will remain a disappointment this gen.

TESKATLIPOKA said:
And where did I say It should be priced the same?
You are acting like nT performance is some holy grail and the rest doesn't matter, but for ultraportables that's not true. LNL has 8 cores, that's good enough.

Agree.

Abwx · Thursday at 3:39 PM

Magio said:
Is there a reason besides thread derailment you're avoiding talking about GPU perf or ST perf or battery life thanks to the low power E cores or the other aspects that make LNL the best x86 platform yet for thin and lights?

Of course its low thread count disqualifies it from certain use cases, but it is absolutely leading edge in the aspects it targets.

GPU perfs it s not exceptional, not really better than the competion previous gen, and it s not like it can compensate for the low MT perfs, in ST no one wil notice the diference, you are manifying 10% as if it was 100%, and once there s two threads, wich is nothing, the ST perf will collapse by 35% if it s not set at 30-37W, hence why MT perf is that important.

Abwx · Thursday at 3:47 PM

TESKATLIPOKA said:
Because @Abwx must compare in a metric where AMD wins significantly and that one metric becomes the most important for him.

Cinebench is not indicative of real usage but it is indicative of throughput in MT,
if it s weak in CB it wil lbe weak as well in 7 Zip and whatever use 8 cores, reviews are done with clean OS that are not crippled by hundreds of low priority threads, that s the avantage of a lot of cores and SMT, all the background tasks impacts on responsiveness are neutered.

controlflow · Thursday at 3:48 PM

Abwx said:
GPU perfs it s not exceptional, not really better than the competion previous gen, and it s not like it can compensate for the low MT perfs, in ST no one wil notice the diference, you are manifying 10% as if it was 100%, and once there s two threads, wich is nothing, the ST perf will collapse by 35% if it s not set at 30-37W, hence why MT perf is that important.

LNL's iGPU is in general the best performance and most efficient iGPU in the non-Apple ecosystem

The lack of class leading nT performance is just not that important in this segment. I would argue that the faster ST performance of LNL would be felt much more often by users of an ultra portable than the nT advantage of Strix would be.

This doesn't even capture the fact that LNL has far higher efficiency at ultra low power situations thanks to how much further Skymont can scale down to low power compared to Strix.

I'm not saying that Strix is a bad product, but it is very clearly in a different segment than something like LNL. At least make an effort to be objective. The arguments you make are incredibly disingenuous. You cherry pick scenarios that are largely irrelevant in the ultra portable segment to suit your biased narrative.

SiliconFly · Thursday at 3:57 PM

controlflow said:
LNL's iGPU is in general the best performance and most efficient iGPU in the non-Apple ecosystem

The lack of class leading nT performance is just not that important in this segment. I would argue that the faster ST performance of LNL would be felt much more often by users of an ultra portable than the nT advantage of Strix would be.

This doesn't even capture the fact that LNL has far higher efficiency at ultra low power situations thanks to how much further Skymont can scale down to low power compared to Strix.

I'm not saying that Strix is a bad product, but it is very clearly in a different segment than something like LNL. At least make an effort to be objective. The arguments you make are incredibly disingenuous. You cherry pick scenarios that are largely irrelevant in the ultra portable segment to suit your biased narrative.

Very True.

Abwx · Thursday at 3:58 PM

controlflow said:
This doesn't even capture the fact that LNL has far higher efficiency at ultra low power situations thanks to how much further Skymont can scale down to low power compared to Strix.

Look at the numbers at 15W, it s less efficient than a Hawk Point for the CPU part, let alone a 365, dunno from where you got your "far higher efficency at ultra low power", surely from some intel slide.

Edit : Actualy the 365@9W would roughly match the 288V@15W.

And since you re talking GPU :

Intel Core Ultra 7 258V trades blows with Ryzen Z1 Extreme in gaming handheld simulation

The Intel Lunar Lake laptops are here, and soon, MSI Claw 8 AI+ will debut as one of the first handhelds with the new Core Ultra 200 series. To see how well it may perform, ETA Prime has put the Core Ultra 7 258V against Ryzen Z1 Exterme in a gaming handheld simulation.

www.notebookcheck.net

cebri1 · Thursday at 4:07 PM

Abwx said:
Look at the numbers at 15W, it s less efficient than a Hawk Point for the CPU part, let alone a 365, dunno from where you got your "far higher efficency at ultra low power", surely from some intel slide.

24T vs 8T. But any rational discussion with you is impossible.

Raw comparison :

288V: 55p/thread
370: 24p/thread

Ofc hard to compared because intel uses two different archs.

SiliconFly · Thursday at 4:08 PM

Abwx said:
Look at the numbers at 15W, it s less efficient than a Hawk Point for the CPU part, let alone a 365, dunno from where you got your "far higher efficency at ultra low power", surely from some intel slide.

Still trying to compare efficiency using Cinebench Multi Core results... even after many here have clearly stated MT isn't that important in thin and lights??? 🤦‍♂️

Abwx · Thursday at 4:24 PM

SiliconFly said:
Still trying to compare efficiency using Cinebench Multi Core results... even after many here have clearly stated MT isn't that important in thin and lights??? 🤦‍♂️

You could use anything else that stress the CPU and it would be the same, 7 Zip or a Stockfish powered chess game , although for the later case you ll have 70% more chance to win a game against LNL.

H433x0n · Thursday at 4:35 PM

Abwx said:
You could use anything else that stress the CPU and it would be the same, 7 Zip or a Stockfish powered chess game , although for the later case you ll have 70% more chance to win a game against LNL.

Why is this such a sticking point? Nobody is buying LNL to render images.

It’s meant for battery life, web browsing, watching videos and office tasks. It’s also meant to be used an APU for handheld devices.

Tup3x · Thursday at 4:44 PM

H433x0n said:
Why is this such a sticking point? Nobody is buying LNL to render images.

It’s meant for battery life, web browsing, watching videos and office tasks. It’s also meant to be used an APU for handheld devices.

It's not made by AMD. That's the reason. 🫠

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

Senior member

Attachments

Platinum Member

Diamond Member

Member

Diamond Member

Diamond Member

Senior member

Platinum Member

Platinum Member

Junior Member

Platinum Member

Diamond Member

Member

Golden Member

Junior Member

Golden Member

Lifer

Lifer

Member

Golden Member

Lifer

Attachments

Senior member

Golden Member

Lifer

Golden Member

Golden Member