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

 

[SiliconFly]

.

 

[cannedlake240]

Pat also confirmed that NovaLake will be partially outsourced... Newer nodes are losing relevance, especially on desktop so that means there's enough of a PPA difference between 18A and whatever TSMC nodes is out by then to warrant using the newest more expensive option. Or It could be that the architecture can't compete iso node and they need the latest process from TSMC to have an edge

 

[DrMrLordX]

Hope they release Arrow Lake with bugs fixed on 18A in six months and max boost 6 GHz to get consumer and investor confidence back.

6 months? Not happening. See other replies (Nova Lake) but even if Intel were fully committed to rushing something 18a-based out the door, they couldn't have it ready in six months.

intel 7 is probably a little too old, but they maybe could’ve used Intel 3?

Interesting idea, but hold off until we see Arrow Lake-U to get an idea of how Intel 3 really changes things. We should be able to do a nearly 1-to-1 comparison between it and Meteor Lake (Intel 4).

Oh, forgot to mention Wendell had some good results in linux -

Phoronix already showed it getting mostly smoked by the 9950x under Linux (admittedly, it's Phoronix, ymmv on desktop CPU reviews). How does Wendel's analysis hold up? Haven't watched the video.

Hey, you can actually buy Sierra Forrest from an OEM! From Lenovo anyway.

It's very not cheap though. At least the retail price.

That's probably a good sign for Intel. Even if it's about to be massacred.

That's always true. Don't buy Arrow Lake, wait until they ship its 18A follow-on. Don't buy Zen 5 today, wait for the N3E version. Don't X Elite today, wait for their next generation core. Don't buy M4 Macs, wait for next year's M5 with multitronic technology!

M4 seems fine, and tbh Zen5 does too (as long as you get the right CPU for your purposes). Arrow Lake doesn't seem to have any immediate followup. What N3E version? Am I missing something?

@Kocicak

It doesn't look like Intel will have anything new next year to replace Arrow Lake, and AMD will probably ride Zen5 until 2026 at the earliest.

Intel 3 capacity is still quite limited:

Hmmm, interesting. So is 20a and so will be 18a.

 

[DavidC1]

@AMDK11 I noticed that maybe you might have a soft spot for products that are being ignored, and why you were favoring Lion Cove.

The high level technical specifications do not matter, if they do not work together. It can be, and are often compared to cooking. You can have all the "high level" stuff right such as right ingredients. But how you execute them is what makes all the difference.

For whatever reason Intel spent all the transistor resources on Lion Cove, and it did not work out. Tesla presentation said 8-wide decode is impossible on variable width ISA such as x86, but Intel managed to do it anyway. But for what gain? Why did they do so? Well, logically they have plans for much better but weren't able to carry on their vision. Or maybe they spent time/resources on something that weren't going to net them a lot because it's overly complicated(decode).

This tells us that Intel as a company is still in deep trouble.

Hmmm, interesting. So is 20a and so will be 18a.

20A is basically rebranded to 18A.

 

[jpiniero]

It doesn't look like Intel will have anything new next year to replace Arrow Lake, and AMD will probably ride Zen5 until 2026 at the earliest.

Define 'new'. Bartlett Lake should still come out next year.

 

[DrMrLordX]

Define 'new'. Bartlett Lake should still come out next year.

That's a budget segment product. @Kocicak seemed to be implying that we'll be looking at Zen5 and Arrow Lake refresh products next year.

 

[Doug S]

Pat also confirmed that NovaLake will be partially outsourced... Newer nodes are losing relevance, especially on desktop so that means there's enough of a PPA difference between 18A and whatever TSMC nodes is out by then to warrant using the newest more expensive option. Or It could be that the architecture can't compete iso node and they need the latest process from TSMC to have an edge

More likely they simply won't have the volume with 18A to serve all their needs. Especially with the announcement they've snagged two 18A customers. I'm sure that for now, given their current situation, they're going to put third party customers ahead of their internal needs on 18A.

Whether 18A is better or worse than the TSMC node du jour is fairly irrelevant. They can steer the chips going to higher ASP markets to whichever node is better. Doing chiplets/tiles adds even more flexibility - e.g. to save money on a GPU tile in a cheaper mass market SKU you can put it on TSMC N6/N4 or Intel 10nm++++++++

 

[cannedlake240]

More likely they simply won't have the volume with 18A to serve all their needs. Especially with the announcement they've snagged two 18A customers. I'm sure that for now, given their current situation, they're going to put third party customers ahead of their internal needs on 18A.

Whether 18A is better or worse than the TSMC node du jour is fairly irrelevant. They can steer the chips going to higher ASP markets to whichever node is better. Doing chiplets/tiles adds even more flexibility - e.g. to save money on a GPU tile in a cheaper mass market SKU you can put it on TSMC N6/N4 or Intel 10nm++++++++

Yeah this is a more optimistic take on this. People who knew about this however seem to think that NVL top skus will all exclusively use N2P because it apparently has a clear PPA advantage over 18AP

 

[DavidC1]

More likely they simply won't have the volume with 18A to serve all their needs. Especially with the announcement they've snagged two 18A customers. I'm sure that for now, given their current situation, they're going to put third party customers ahead of their internal needs on 18A.

Why do they project they'll have volume problems for so long? Even with the scaled down fab plans, it should be enough for them. They basically have no 3rd party wins.

 

[cannedlake240]

Why do they project they'll have volume problems for so long? Even with the scaled down fab plans, it should be enough for them. They basically have no 3rd party wins.

They do have 6 or 7 external customers, Pats saying 2 new "compute centric" ones added this quarter, but they won't reveal themselves. There was a rumor that Nvidia's in there as well. Probably no truly large deals though...

 

[Doug S]

Why do they project they'll have volume problems for so long? Even with the scaled down fab plans, it should be enough for them. They basically have no 3rd party wins.

Intel's "five nodes in four years" kept too much capacity out of service doing upgrades, plus they've been playing catchup on acquiring EUV scanners. I think that's the real reason they canceled 20A, outfitting for an internal only node that will be used by only one or two products just doesn't make a whole lot of sense. It also causes problems for Intel's "copy exactly" strategy if the nodes aren't around long enough to get copied to other fabs, because those other fabs are being set up for the next node in line.

I didn't read the transcript but from what I heard they claim to have added two customers for 18A. Whether they add meaningful volume I don't know.

 

[DavidC1]

Intel's "five nodes in four years" kept too much capacity out of service doing upgrades, plus they've been playing catchup on acquiring EUV scanners. I think that's the real reason they canceled 20A, outfitting for an internal only node that will be used by only one or two products just doesn't make a whole lot of sense.

That explains a bit, but they've moved significant portions to TSMC, with Novalake rumored to be on it too.

I would almost bet on external deals being negligible, a blip on the radar compared to their own. Saying the customer is Nvidia doesn't matter. It needs to be big, like everything x60 and under using Intel. Client graphics has lot less volume than Intel CPU. What external deals are necessary to replace say, 10 million annum 100mm2 dies?

20A is underperforming too. 18A is basically original 20A. If @Kepler_L2 is right about 5GHz Arrowlake, then that's Cannonlake 10nm level trash.

 

[adroc_thurston]

If @Kepler_L2 is right about 5GHz Arrowlake, then that's Cannonlake 10nm level trash.

Oh don't be silly. CNL was unyieldable.
This is more 10+ ICL-U type of stuff.

 

[DavidC1]

Oh don't be silly. CNL was unyieldable.
This is more 10+ ICL-U type of stuff.

I'm saying it's horse poop, while you are saying it's bird poop. A reason enough to cancel after all that hoopla about "rearview mirror" and "undisputed leadership".

 

[adroc_thurston]

I'm saying it's horse poop, while you are saying it's bird poop. A reason enough to cancel after all that hoopla about "rearview mirror" and "undisputed leadership".

Yeah Pat was too loud for his own good, but 5Ghz on poverty SKUs on a new node is plenty fine. The question would be more in power/yield dimension.
Given it's canned, not promising on both!

 

[MoistOintment]

or Intel 10nm++++++++

That would never happen. There is no chance Intel would port Arc to a node so unoptimized for a tGPU like Intel 7

People who knew about this however seem to think that NVL top skus will all exclusively use N2P because it apparently has a clear PPA advantage over 18AP

Or that they may still rely on N6 to some extent. Intel doesn't have a node like N6: Efficient, cheap, plentiful, and really good low wattage characteristics.
Intel 7 will hang around on RPL-R. Intel 16 is too low-end. Intel 3 is too expensive. And 18A would defeat the purpose of the tile design in the first place.

 

[adroc_thurston]

that NVL top skus will all exclusively use N2P because it apparently has a clear PPA advantage over 18AP

Yeah, 18A derivatives are not even close.

Or that they may still rely on N6 to some extent. Intel doesn't have a node like N6: Efficient, cheap, plentiful, and really good low wattage characteristics.

Pretty sure they announced a cost optimized variant of i3 specifically for that.
Stashed somewhere in the recent foundry PR.

 

[cannedlake240]

Yeah, 18A derivatives are not even close.

N4-N3 gap at best, which shouldn't even matter for desktop chips and 18A density isn't even that much lower

 

[DavidC1]

N4-N3 gap at best, which shouldn't even matter for desktop chips and 18A density isn't even that much lower

If what you are saying is true, then Intel has truly separated the two into separate entities, or that there's something else going on with 18A derivatives which is why the design teams prefer TSMC.

 

[511]

Yeah this is a more optimistic take on this. People who knew about this however seem to think that NVL top skus will all exclusively use N2P because it apparently has a clear PPA advantage over 18AP

N2P won't come until later half of 26 and only for apple it is 18A for majority and Graphics+IO will be TSMC

 

[511]

20A is underperforming too. 18A is basically original 20A. If @Kepler_L2 is right about 5GHz Arrowlake, then that's Cannonlake 10nm level trash.

Cannon Lake level trash means barely yielding and 3ghz max it seems Tiger Lake level 🙂

 

[511]

If it's N2P NVL is not coming before 2027 TSMC roadmap It's either 18AP or N2

I think the CPU would definitely be 18A considering DMR is 18A both uses same Panther Cove would be a duplicate efforts feel free to correct me

 

[dttprofessor]

Pat want everybody forget intel 7,then half products of intel on it ,so intel will suffer next year still.

 

[Thunder 57]

Yeah Pat was too loud for his own good, but 5Ghz on poverty SKUs on a new node is plenty fine. The question would be more in power/yield dimension.
Given it's canned, not promising on both!

I don't understand your need to make silly names for things you don't care about. At least Cinememe sort of made sense. But calling things "poverty SKUs" when they will be fine for most people is just, insert any of many possible words here, elitist?

 

[adroc_thurston]

But calling things "poverty SKUs" when they will be fine for most people is just, insert any of many possible words here, elitist?

But that's what they are. 6+8 ARL was i3 tier only.