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)	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 ?	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.)

 

[Joe NYC]

Does that mean Intel's first Intel 4 CPUs won't begin shipping for another year? Is there anything else on Intel 4 in the meantime? If not, so much for Intel's aggressive process roadmap.

Nope, MTL is the first CPU on Intel 4.

5 nodes in 4 years:

2021 Intel 7
2022
2023
2024 Intel 4, Intel 3
2025 20A, 18A

 

[eek2121]

AMD can't meet demand with Zen3+ and I don't think Zen 4 mobile will fare better from an availability perspective. Intel's fine. They got the whole of 2023 to saturate the market with mobile Raptor Lake.

That is one of the reasons AMD is changing things up with naming. Zen 2/3/4 will all be part of next gen, able to use N6/N5 and possibly even N7, bus alas, this is an Intel thread…

To be fair, Intel's own engineering department did label that a 10nm process, so as far as any of us lay people are concerned, it was a 10nm process. Intel 4 was initially a 7nm process with a later name change to fool tech journalists (and it seems to have worked).

As far as availability goes? If they're projecting Meteor Lake to arrive that late, yes, it may as well be a 2024 product. Which is pretty sad. It should have been out by June/July to secure some back-to-school design wins.

It wasn’t to “fool tech journalists”. Intel 10nm (as originally envisioned) is far superior to Samsung 8N and almost equal to TSMC N7. A ton of folks gave Intel fluff for only being on “10nm” while others were on “7nm” so Intel changed the names to match competitors. In reality, later revisions of Intel’s 14nm process should have been called “10nm”.

If they had done this earlier in the game you never would have known. In the end, a name is a name. Performance characteristics and density are what matter.

Basically folks were claiming Intel has a disadvantage in an area where they do not, simply because of product names from competitors.

EDIT: AnandTech has an article on this. Note that Intel says that “Intel 7” has a higher density than TSMC N7: https://www.anandtech.com/show/1682...nm-3nm-20a-18a-packaging-foundry-emib-foveros

 

[Hulk]

What is the density of TMSC 7 compared to Intel 7?

 

[DrMrLordX]

It wasn’t to “fool tech journalists”. Intel 10nm (as originally envisioned) is far superior to Samsung 8N and almost equal to TSMC N7. A ton of folks gave Intel fluff for only being on “10nm” while others were on “7nm” so Intel changed the names to match competitors. In reality, later revisions of Intel’s 14nm process should have been called “10nm”.

If you say so man, but it's hard to take Intel seriously when they do an about-face after announcing the roadmap and everything. TSMC doesn't even use "nm" in their marketing anymore, and only idiot tech journalists will call N7 a 7nm node (for example). TSMC renaming a tweaked 16nm to 12nm was obviously bogus. It's good that they changed the way they name their nodes so as not to (necessarily) advertise feature size. If they had renamed 16nm to N12 in the middle of a production cycle it still would have reeked of bs.

Whenever some journalist calls Intel 4 a "4nm" node without any context or anything, it just shows that what Intel did, worked.

2024 Intel 4, Intel 3
2025 20A, 18A

Back to the topic of Meteor Lake in particular, that scenario is what should have the x86 world a little worried. Does anyone legitimately think Intel can pull that off? After having no meaningful node changes in two years, that they're going to release 4 seperate nodes in the next 2? Have they ever done that before, even in their best days?

 

[Geddagod]

To be fair, Intel's own engineering department did label that a 10nm process, so as far as any of us lay people are concerned, it was a 10nm process. Intel 4 was initially a 7nm process with a later name change to fool tech journalists (and it seems to have worked).

As far as availability goes? If they're projecting Meteor Lake to arrive that late, yes, it may as well be a 2024 product. Which is pretty sad. It should have been out by June/July to secure some back-to-school design wins.

Ah yes they have fooled the tech journalists who have been reporting on news from these companies, interviewing their executives, talking with their engineers, and running their own tests, for multiple years. Instead, it is the laymen on the internet with no real connection the industry who know the REAL node names.

 

[moinmoin]

tech journalists

You call somebody who calls "Intel 4" "4nm" a tech journalist? Asking for a friend.

 

[DrMrLordX]

Instead, it is the laymen on the internet with no real connection the industry who know the REAL node names.

People on Internet forums don't have to maintain a relationship with marketing people in the industry to keep getting scoops. If someone like TSMC (see above, 12nm was obviously a bs name) or Intel plays stupid games with renames, we're free to criticize them for it. Someone whose entire career revolves around continued access to media departments has to limit their vitriol, unless they're like The Reg and don't give a darn about anything (and I doubt they get invited to many press events but who knows). And that's just the few tech journalists that actually put in the legwork and understand the industry. You get plenty of tech blogger types who have no idea what's going on.

The reality is that Intel 7nm is still Intel 7nm, and it will always be Intel 7nm, and changing the name to Intel 4 doesn't change the feature sizes one bit. If Intel had never rebranded it you'd be calling it a 7nm node, too. Think about that for a minute.

TSMC N4 isn't a 4nm node either (necessarily), think about THAT for another minute.

Regardless, Intel 4 may as well not exist yet since Meteor Lake (which is the actual topic of discussion here) may not even show up until a year from now.

You call somebody who calls "Intel 4" "4nm" a tech journalist? Asking for a friend.

A few good writers have taken to this practice. I think Dr. Cutress has done it a few times. It's easier to just call it 4nm and call it a day.

 

[Doug S]

Whenever some journalist calls Intel 4 a "4nm" node without any context or anything, it just shows that what Intel did, worked.

Back to the topic of Meteor Lake in particular, that scenario is what should have the x86 world a little worried. Does anyone legitimately think Intel can pull that off? After having no meaningful node changes in two years, that they're going to release 4 seperate nodes in the next 2? Have they ever done that before, even in their best days?

It is no more wrong to call Intel 4 "4nm" than it is to call N4 or 4LPP "4nm". Intel stated they were changing their naming scheme since the process naming and minimum drawn gate length had long ago ceased to have any correlation, and to match what the competition was doing with their node names. Theoretically it shouldn't matter, because the "customers" of TSMC and Intel foundries are big companies that know that Intel 7 is neither 7nm nor 4nm, and won't be based their decisions on marketing names.

As for how Intel 4 compares with N4, since it appears we won't see Intel 4 chips for another year or so, we won't know the answer to that for a while. We can't use Intel's density claims anymore than we can use TSMC's - those claims are not reflective of real world CPUs, as the vastly different densities of e.g. Intel CPUs and AMD CPUs / Apple SoCs clearly demonstrates.

I don't think anyone believes Intel will roll out their very first EUV chips in production quantities a year from now, and will then follow that up with production quantities of another major node along with a switch to GAA transistors AND buried/backside power within 24 months.

 

[moinmoin]

It is no more wrong to call Intel 4 "4nm" than it is to call N4 or 4LPP "4nm".

It's no more right either.

 

[DrMrLordX]

It's no more right either.

Bingo.

 

[S'renne]

By the way, so is there any way of knowing when would Meteor Lake be available on the mobile lineup until 2023?

 

[Exist50]

TSMC doesn't even use "nm" in their marketing anymore, and only idiot tech journalists will call N7 a 7nm node (for example).

TSMC's 7nm Fin Field-Effect Transistor (FinFET) (N7) process technology sets the industry pace for 7nm process technology development by delivering 256Mb SRAM with double-digit yields in June 2016. In 2019, in N7 process node's second year of volume production, customers taped out more than 110 new generation products on N7. In addition, 7nm FinFET plus (N7+) technology entered full-scale production in 2019 and delivered customer 7nm products to market in high volume.

Sure sounds like they still use "nm"... It's even in the URL.

7nm Technology - Taiwan Semiconductor Manufacturing Company Limited

TSMC's 7nm Fin Field-Effect Transistor (FinFET) (N7) process technology sets the industry pace for 7nm process technology development by delivering 256Mb SRAM with double-digit yields in June 2016. In 2019, in N7 process node's second year of volume production, customers taped out more than 110...

www.tsmc.com

 

[DrMrLordX]

Sure sounds like they still use "nm"... It's even in the URL.

That's interesting. Everywhere else it's just N7, N5, etc. They don't even use nm on any of their charts or roadmaps that I can recall.

By the way, so is there any way of knowing when would Meteor Lake be available on the mobile lineup until 2023?

Not really. Watch Intel's releases carefully, and take rumours with a grain of salt. Right now it looks like limited release in Q4 2023.

 

[Joe NYC]

By the way, so is there any way of knowing when would Meteor Lake be available on the mobile lineup until 2023?

2023 release is still the plan.

Unless there is some unlikely, dramatic news that come out in next 33 days - no, there is no ways of knowing in 2022, we will have to wait until 2023.

 

[JustViewing]

What is the density of TMSC 7 compared to Intel 7?

Does it really matter? GF14/12nm Zen1 gave better performance and efficiency than Intel 14nm even though Intel 14nm supposed to be superior. Stock 5nm Zen4 on superior node, giving relatively bad efficiency than stock 7nm Zen3. Intel 10nm improved a lot from the initial release. In the end what is delivered that matters to us as consumers.

 

[Hulk]

Does it really matter? GF14/12nm Zen1 gave better performance and efficiency than Intel 14nm even though Intel 14nm supposed to be superior. Stock 5nm Zen4 on superior node, giving relatively bad efficiency than stock 7nm Zen3. Intel 10nm improved a lot from the initial release. In the end what is delivered that matters to us as consumers.

In a tech forum in a thread titled "Meteor Lake and Arrow Lake Discussion Thread?"

Yeah, it matters.

 

[Geddagod]

https://twitter.com/x/status/1597467653094150144

Would be pretty cool if true. I'm guessing the L4 would be on the base tile (Intel 22nm). While I don't think the performance uplift would be anywhere near stacking L3 cache, depending on the size it should bring a nice uplift in gaming, like we saw with Intel Broadwell L4.
Funnily enough, the E-DRAM on Broadwell was manufactured on the same node as the Base Tile of Meteor Lake, 22nm.

 

[Doug S]

Sure sounds like they still use "nm"... It's even in the URL.

7nm Technology - Taiwan Semiconductor Manufacturing Company Limited

TSMC's 7nm Fin Field-Effect Transistor (FinFET) (N7) process technology sets the industry pace for 7nm process technology development by delivering 256Mb SRAM with double-digit yields in June 2016. In 2019, in N7 process node's second year of volume production, customers taped out more than 110...

www.tsmc.com

TSMC was always talking about "10nm" I don't recall ever hearing it referred to as N10 at the time. So I think 7nm/N7 is when they decided to change their nomenclature; seeing some old stuff that calls it 7nm is perhaps not too surprising.

Now if there was something current where they talked about 3nm instead of N3 then I would be very surprised.

 

[Exist50]

Now if there was something current where they talked about 3nm instead of N3 then I would be very surprised.

TSMC’s 3nm technology (N3) will be another full node stride from our 5nm technology (N5), and offer the most advanced foundry technology in both PPA and transistor technology when it is introduced.

https://www.tsmc.com/english/dedicatedFoundry/technology/logic/l_3nm

Frankly, I think this is a ridiculous discussion to be having. On a tech forum, most people probably know that the names are meaningless, so let's stop beating a dead horse. If there's going to be a node discussion, the name shouldn't even factor in.

 

[Exist50]

https://twitter.com/x/status/1597467653094150144

Would be pretty cool if true. I'm guessing the L4 would be on the base tile (Intel 22nm). While I don't think the performance uplift would be anywhere near stacking L3 cache, depending on the size it should bring a nice uplift in gaming, like we saw with Intel Broadwell L4.
Funnily enough, the E-DRAM on Broadwell was manufactured on the same node as the Base Tile of Meteor Lake, 22nm.

They've hinted at the idea before, so it wouldn't be unheard of.

Would be interesting to see how they'd implement it. Shared memory-side/system cache, or dedicated chunks for CPU vs GPU as L4?

Though Intel's never mentioned it in any of their Meteor Lake disclosures, so I have my doubts. Seems like the kind of thing that'd make more sense for a really large GPU config, just like eDRAM was.

 

[Henry swagger]

https://twitter.com/x/status/1597467653094150144

Seems the gpu tile and cpu will share a l4 cache aka system cache 🤔😁

 

[Geddagod]

They've hinted at the idea before, so it wouldn't be unheard of.
View attachment 71925
Would be interesting to see how they'd implement it. Shared memory-side/system cache, or dedicated chunks for CPU vs GPU as L4?

Though Intel's never mentioned it in any of their Meteor Lake disclosures, so I have my doubts. Seems like the kind of thing that'd make more sense for a really large GPU config, just like eDRAM was.

Intel hasn't disclosed a lot about MTL imo
The heavily rumored little cores on the SOC tile hasn't been disclosed (if true) either afaik despite sounding like a bigger deal than an additional L4 cache

 

[DrMrLordX]

Would be pretty cool if true. I'm guessing the L4 would be on the base tile (Intel 22nm). While I don't think the performance uplift would be anywhere near stacking L3 cache, depending on the size it should bring a nice uplift in gaming, like we saw with Intel Broadwell L4.
Funnily enough, the E-DRAM on Broadwell was manufactured on the same node as the Base Tile of Meteor Lake, 22nm.

It would be an interesting use of base tile real estate. Though there's also the possibility that it might be part of the SoC tile, which would explain why it's so big.

Enthusiasts shouldn't get too excited about L4, though. The last consumer L4 product Intel had featured eDRAM that didn't outperform tuned-by-the-standards-of-the-day DDR4. By the time most of the DiY market had moved on to DDR4-2933 CAS/CL14 or better, it provided little benefit. Seems like it could improve some inter-tile communications and cover up some weak DDR5 configs like DDR5-5200 or whatever else OEMs feel like pairing with Meteor Lake laptops.

 

[ashFTW]

So it seems, as I have argued before, the MTL base tile may indeed not be passive, which makes sense given it’s a Foveros based design. Its a huge piece of silicon; it would be a wasted opportunity if it’s just being used as an interposer..

 

[eek2121]

https://twitter.com/x/status/1597467653094150144

Would be pretty cool if true. I'm guessing the L4 would be on the base tile (Intel 22nm). While I don't think the performance uplift would be anywhere near stacking L3 cache, depending on the size it should bring a nice uplift in gaming, like we saw with Intel Broadwell L4.
Funnily enough, the E-DRAM on Broadwell was manufactured on the same node as the Base Tile of Meteor Lake, 22nm.

Curious. This may be Intel's attempt to counter the AMD X3D parts.