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.)

 

[eek2121]

"GPU and other bits" is the majority of the silicon, if you ignore the base die. There's >100mm2 of N6, and another good chunk of N5. That cost is going to dwarf the compute die.

Of course it will, but the entire package will still cost less than it costs AMD for a typical Zen 4 chip.

 

[BorisTheBlade82]

MLID becoming more realistic with the claimed IPC improvements in Meteor Lake? Now he says MTL gains "up to" 10% over RPL-R with some tests in the mid single digits range. It's more realistic right?

There are some other claims. Current boost speed projections are 5.2 Ghz for 28W and 5.4 Ghz for 45W. Another claim: MTL is projected to consume 30-45% less energy than RPL-R at same performance below 45W and it launches between August-October.

As for Arrow Lake he says that both mobile and desktop will launch in Q4 2024. Mobile uses 20A whereas desktop uses TSMC N3+N3E. Recent 6+8 vs 6+8 benchmarks show ARL with 30% higher ST and 40% higher mixed average over MTL.

Yep, regarding MTL that sounds much more down to earth.

 

[Geddagod]

MLID becoming more realistic with the claimed IPC improvements in Meteor Lake? Now he says MTL gains "up to" 10% over RPL-R with some tests in the mid single digits range. It's more realistic right?

There are some other claims. Current boost speed projections are 5.2 Ghz for 28W and 5.4 Ghz for 45W. Another claim: MTL is projected to consume 30-45% less energy than RPL-R at same performance below 45W and it launches between August-October.

As for Arrow Lake he says that both mobile and desktop will launch in Q4 2024. Mobile uses 20A whereas desktop uses TSMC N3+N3E. Recent 6+8 vs 6+8 benchmarks show ARL with 30% higher ST and 40% higher mixed average over MTL.

10% might still be a bit high, but then again he did say it was "up to" and avrg was single digits, so ye seems alright.
Funny though, he claims it's because RWC didn't reach expected IPC targets. Uh huh.
5.4 GHz ST max is reasonable imo. A good sign even for the node considering Intel's 'new' nodes max ST frequency went down for both 14 and 10nm.
30-45% less energy for the same performance is also a reasonable estimate I suppose. To put it into perspective, TGL vs CML was a ~25-30% gain (8c parts).
ARL 2H 2024 should be expected. Intel confirmed ARL is hitting their launch by 2024 timeline. Depending on how TSMC N3 is, they might even push for a 2Q launch to be able to compete against Zen 5 as soon as possible.
I would be a bit surprised about how they are benchmarking ARL right now. I expect ARL to just have been turned on, and maybe Intel announced that in a couple days during their Intel vision event. I also expect clocks to be pretty low right now, so unless they are clock-equalized I don't see how ARL is beating out MTL chips already by such large margins.

 

[Geddagod]

10% might still be a bit high, but then again he did say it was "up to" and avrg was single digits, so ye seems alright.
Funny though, he claims it's because RWC didn't reach expected IPC targets. Uh huh.
5.4 GHz ST max is reasonable imo. A good sign even for the node considering Intel's 'new' nodes max ST frequency went down for both 14 and 10nm.
30-45% less energy for the same performance is also a reasonable estimate I suppose. To put it into perspective, TGL vs CML was a ~25-30% gain (8c parts).
ARL 2H 2024 should be expected. Intel confirmed ARL is hitting their launch by 2024 timeline. Depending on how TSMC N3 is, they might even push for a 2Q launch to be able to compete against Zen 5 as soon as possible.
I would be a bit surprised about how they are benchmarking ARL right now. I expect ARL to just have been turned on, and maybe Intel announced that in a couple days during their Intel vision event. I also expect clocks to be pretty low right now, so unless they are clock-equalized I don't see how ARL is beating out MTL chips already by such large margins.

Also just want to add it would be funny if his IPC estimates for GNR is accurate because it uses LNC lol

 

[eek2121]

Also just want to add it would be funny if his IPC estimates for GNR is accurate because it uses LNC lol

You mean guesses? Guess often enough and eventually you'll get something right.

 

[A///]

you'd be more accurate with a bloody oujia board for petes sake.

 

[Hulk]

I watched that MLID video and actually I think everything he said regarding the Intel timeline and expectations are relatively reasonable. Also the timeline actually makes sense from the perspective of where each part fits into the stack. Specifically why Raptor Refresh will exist and how MTL/ARL compatibility will spur more MTL design wins that it would purely based on out of the box performance.

However I don't think ARL will show 30% IPC, more like 20%. Also RPL-R single core clocks are meaningless except for the fact that nC clocks can be inferred from them. So 6.2GHz from the refresh could mean 5.8GHc nC. 25% power reduction from Raptor to Raptor Refresh at ISO frequency would be nice if true but I'm having a hard time believing that as well.

 

[H433x0n]

Let's give Intel the benefit of the doubt and wet our floors with ample amount of drool.

30-40% (not 30-40% IPC) performance gains isn’t that crazy, it’s happened multiple times on a big node jump. Just look at Golden Cove from Skylake era. Zen 3 was a 30-40% performance jump in some tasks too.

 

[Anhiel]

MLID becoming more realistic with the claimed IPC improvements in Meteor Lake? Now he says MTL gains "up to" 10% over RPL-R with some tests in the mid single digits range. It's more realistic right?

There are some other claims. Current boost speed projections are 5.2 Ghz for 28W and 5.4 Ghz for 45W. Another claim: MTL is projected to consume 30-45% less energy than RPL-R at same performance below 45W and it launches between August-October.

As for Arrow Lake he says that both mobile and desktop will launch in Q4 2024. Mobile uses 20A whereas desktop uses TSMC N3+N3E. Recent 6+8 vs 6+8 benchmarks show ARL with 30% higher ST and 40% higher mixed average over MTL.

Now, these raw and undiluted data points look... perfect. Roughly 2x2% off from my own estimations posted before.
Things've only changed for MTL but not really ARL. It's a ladder system of dependencies so things do carry over.
Either way, it gives a clearer picture, yet, these perfect numbers make me wary of the last time perfect numbers became a disaster. I'm leaning toward these being more reliable judging by how so many fall into place now.
But there seems to be one contradiction here with Intel's measurement. Welp, I'll leave it at that.
Anyhow, it seems ARL will come with L4 die with backside power delivery. This is probably the biggest change in metric from before. Welp, currently this won't count practically yet other than for games or large data sets etc.
Since this only slightly differ from before my earlier conclusion remain the same.

 

[eek2121]

30-40% (not 30-40% IPC) performance gains isn’t that crazy, it’s happened multiple times on a big node jump. Just look at Golden Cove from Skylake era. Zen 3 was a 30-40% performance jump in some tasks too.

Not in a single generation. Did you forget Rocket/Tiger Lake? Also, Zen 3 had both an IPC and clock speed advantage.

Now, these raw and undiluted data points look... perfect. Roughly 2x2% off from my own estimations posted before.
Things've only changed for MTL but not really ARL. It's a ladder system of dependencies so things do carry over.
Either way, it gives a clearer picture, yet, these perfect numbers make me wary of the last time perfect numbers became a disaster. I'm leaning toward these being more reliable judging by how so many fall into place now.
But there seems to be one contradiction here with Intel's measurement. Welp, I'll leave it at that.
Anyhow, it seems ARL will come with L4 die with backside power delivery. This is probably the biggest change in metric from before. Welp, currently this won't count practically yet other than for games or large data sets etc.
Since this only slightly differ from before my earlier conclusion remain the same.

Meteor Lake consuming 40% less power is a given. It is on a smaller process. The question is whether Intel focuses on perf/watt moving forward. Their latest chips are incredibly inefficient. If they can pull off another Core 2 they will be able to turn things around.

 

[A///]

Rocket

it's easier to forget the pain than try to remember it.

 

[H433x0n]

Not in a single generation. Did you forget Rocket/Tiger Lake? Also, Zen 3 had both an IPC and clock speed advantage.

Going from Comet Lake -> Alder Lake is roughly equivalent of what we’d be getting with Raptor Lake -> Arrow Lake (with MTL being that intermediate step).

That was a sizable jump from April 2020 to November 2021.

 

[soresu]

Specifically why Raptor Refresh will exist and how MTL/ARL compatibility will spur more MTL design wins that it would purely based on out of the box performance

Past events have shown us that no matter competitive AMD HW is at any given time that Intel has zero problems getting design wins one way or another.

What they can't get through HW merit they get through contra revenue strategies, because they have the extreme financial grunt to do so, while AMD has never really been in the position to play that card, and likely never will at this point unless Intel somehow makes a truly catastrophic misstep.

 

[jpiniero]

What they can't get through HW merit they get through contra revenue strategies

Intel doesn't do that anymore. AMD's problem right now is that they don't have an answer for Tiger Lake.

 

[soresu]

Intel doesn't do that anymore

You are only fooling yourself if you believe that.

They've been doing it from day one in the Chromebook market.

Doesn't it strike you as odd that so few ARM based Chromebooks exist by comparison?

The reason is very simple - Intel priced most of the ARM vendors out of the market.

They might not be pricing as aggressively in periods where no ARM vendors are making an effort, but you can bet that they are doing it the moment Qualcomm or Mediatek start making announcements to that effect.

This is especially obvious given the fact that Chrome OS supports Android apps which overwhelmingly work better on ARM CPUs due to ARM's near total dominance of the Android device market.

 

[jpiniero]

You are only fooling yourself if you believe that.

They've been doing it from day one in the Chromebook market.

Doesn't it strike you as odd that so few ARM based Chromebooks exist by comparison?

It's tough to compete with a fully depreciated fab in a market where performance doesn't matter. Esp with products fabbed at TSMC, given the high prices.

Even 10 nm, with nobody wanting Sapphire Rapids, there should be plenty of fab time to pump out cheap Jasper Lake and still be profitable.

 

[soresu]

It's tough to compete with a fully depreciated fab in a market where performance doesn't matter. Esp with products fabbed at TSMC, given the high prices.

That too, but it's only part of the problem.

You can't effectively compete with a player willing to sacrifice most of the profits to keep competitors out of the game completely - not unless you have deep pockets of your own, or an entrenched starting position as the ARM vendors had with the Android segments.

 

[DrMrLordX]

Intel doesn't do that anymore. AMD's problem right now is that they don't have an answer for Tiger Lake.

Tiger Lake? Not sure I understand what you're saying here...

 

[soresu]

Tiger Lake? Not sure I understand what you're saying here...

I'm gonna go out on a limb and assume they are referring to Intel's big Little/P+E implementation using Cove/Mont cores.

Arguably Intel owned themselves as much as anyone else by disabling the AVX512 feature with ADL that they had spent years touting despite it's throttling issues - thus removing a performance feature that many software devs had specifically coded for even in the consumer market.

One step forward, 2 steps back 😅

 

[coercitiv]

I'm gonna go out on a limb and assume they are referring to Intel's big Little/P+E implementation using Cove/Mont cores.

No, he's referring to TGL. There's still an abundance of cheap TGL machines on the market. That's not that big of impediment for AMD though, as their product mix still contains many Lucienne, Cezanne and Barcelo products, some of which are directly priced against TGL laptops. The situation is similar to desktop, where AM4 is still around.

On top of that, ADL can also come pretty cheap in some machines, suggesting Intel is indeed attempting to fill the channel with discounted silicon.

 

[eek2121]

No, he's referring to TGL. There's still an abundance of cheap TGL machines on the market. That's not that big of impediment for AMD though, as their product mix still contains many Lucienne, Cezanne and Barcelo products, some of which are directly priced against TGL laptops. The situation is similar to desktop, where AM4 is still around.

On top of that, ADL can also come pretty cheap in some machines, suggesting Intel is indeed attempting to fill the channel with discounted silicon.

Hey, Tiger Lake isn’t a bad chip. If they had been able to push that out on desktop instead of Rocket Lake, AMD would have had a harder time winning marketshare.

 

[coercitiv]

Hey, Tiger Lake isn’t a bad chip.

Who said anything bad about TGL?

 

[eek2121]

Who said anything bad about TGL?

Hah! Nobody. I was just implying that flooding the market with cheap tiger lake chips isn’t necessarily a bad thing.

 

[coercitiv]

I was just implying that flooding the market with cheap tiger lake chips isn’t necessarily a bad thing.

For consumers it's nice. For Intel folk... debatable.

Very unfortunate news, but massive layoffs at Intel coming! Intel's Datacenter and Client computing groups are receiving ~10% budget cuts It's up to divisions to figure out how to cut Given fixed costs, means as much as 20% layoffs in groups LTD (process node) unaffected

https://twitter.com/x/status/1655048861084450816

 

[Geddagod]

For consumers it's nice. For Intel folk... debatable.

https://twitter.com/x/status/1655048861084450816

Well, more evidence of the 'unlimited budget' given to the fab side.
I mean everyone (especially the people working there) should hope this rumor is false, but if it is...
I do hope they don't cut out engineers and instead more of the bloated management I hear often when people talk about Intel.
I also hope the cuts come more out of the client side, considering that client, tbh, has been doing much better than what server has been able to do. Seems like it would hurt less if the cuts come out of client.
However this seems like especially poor timing as well when you consider Intel has diverged into big and little cores, sounds like a lot more work supporting two archs in many skus. However the move to chiplets in client, where you can have a lot of design reuse, could also reduce total engineering hours needed for a new product.