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.)

 

[igor_kavinski]

Idk how many times I say this, and you just ignore it, but again, it's literally only MLID who expects a high performance uplift. Literally everyone else, Xino, Raichu, etc etc all say the perf uplift is gonna be ~10% over RPL.

Intel "may" have a contingency plan to deal with a repeat of a Zen 3 like situation by now. They kinda kept themselves competitive with Raptor Lake by increasing the cache sizes. Who knows what they might do with Arrow Lake. Maybe ditch the original ARL design and go with the Refresh design with increased cache sizes for the ARL debut? They could also include a more powerful iGPU tile with each ARL desktop part and poke fun at AMD's puny iGPU? They could even go with a hybrid strategy where the original ARL design is used in lower end CPUs with disabled cores to avoid chucking the already manufactured dies and keeping the refresh dies for their high end parts.

 

[Geddagod]

ntel "may" have a contingency plan to deal with a repeat of a Zen 3 like situation by now.

wut

They kinda kept themselves competitive with Raptor Lake by increasing the cache sizes.

I don't think the increased cache sizes did much. In gaming the increased L2s just had an increased PPC of like ~5% in gaming. It was mostly just clocks + more cores.

Maybe ditch the original ARL design and go with the Refresh design with increased cache sizes for the ARL debut?

If ARL-R had increased cache sizes they wouldn't be calling it ARL-R, it would be called something else IMO.

 

[tamz_msc]

Idk how many times I say this, and you just ignore it, but again, it's literally only MLID who expects a high performance uplift. Literally everyone else, Xino, Raichu, etc etc all say the perf uplift is gonna be ~10% over RPL.

MLID says 25-35% higher performance than Raptor Lake, which is presumably MT perf. Raichu says 15% IPC improvement, and 10% ST perf uplift over RPL refresh i.e. 14900K.

I don't see how these two claims are incompatible with each other.

 

[igor_kavinski]

It was mostly just clocks + more cores.

And both were helped by the increased cache availability. Their performance uplift would've been smaller with the original Alder Lake caches.

 

[Geddagod]

MLID says 25-35% higher performance than Raptor Lake, which is presumably MT perf. Raichu says 15% IPC improvement, and 10% ST perf uplift over RPL refresh i.e. 14900K.

I don't see how these two claims are incompatible with each other.

Well, first of all, we are talking about LNC performance, not MT perf in general.
Second of all, MLID thinks the ST perf is gonna be "at least 25%" higher than RPL.

And both were helped by the increased cache availability. Their performance uplift would've been smaller with the original Alder Lake caches

The L2 cache increase only netted like an ~5% gain in PPC in gaming.

 

[H433x0n]

If ARL-R had increased cache sizes they wouldn't be calling it ARL-R, it would be called something else IMO.

Remains to be seen IMO. I don’t know if RPL was really a thing 6-8 months before ADL launch.

 

[Geddagod]

Remains to be seen IMO. I don’t know if RPL was really a thing 6-8 months before ADL launch.

RPL was definitely a thing 6-8 months before ADL launch

 

[igor_kavinski]

The L2 cache increase only netted like an ~5% gain in PPC in gaming.

And I'm saying it HELPED towards the total uplift of at least 13% over 12th gen. Otherwise, it would've been lesser and Intel would've been defeated much worse by the Zen 3D parts.

 

[tamz_msc]

Second of all, MLID thinks the ST perf is gonna be "at least 25%" higher than RPL.

No he doesn't.

 

[Geddagod]

No he doesn't.

View attachment 95812

Yes he does.

 

[H433x0n]

RPL was definitely a thing 6-8 months before ADL launch

I looked online and the first MLID leak about RPL was June 2021.

 

[tamz_msc]

Yes he does.
View attachment 95819

Newer info takes precedence over older info. Not that you can infer much from MLID leaks, whether new or old, but yeah, it is what it is.

 

[igor_kavinski]

How can ARL be 25% higher than MTL in MT workloads without hyperthreading? That points to a pretty serious ST uplift or extremely strong performance by them Skymonts. I think the latter seems more likely.

 

[Geddagod]

I looked online and the first MLID leak about RPL was June 2021.

I'm just going of this tbh

Newer info takes precedence over older info. Not that you can infer much from MLID leaks, whether new or old, but yeah, it is what it is.

The older info doesn't contradict the newer info.

 

[tamz_msc]

The older info doesn't contradict the newer info.

I didn't say that it contradicts the older info. The two leaks aren't comparable because they talk about different things. The older leak speaks of ST perf gain over Raptor Lake, and the newer leak is about (presumably) overall performance over MTL at same core count. If we go by this statement, then it is reasonable to assume that MLID is talking about ARL-H and not ARL-S in the second leak.

However, like I said before, newer leaks from a credible leaker usually takes precedence over older leaks from the same leaker. However, since MLID is all over the place when it comes to accuracy of his leaks, it is hard to make a judgment call.

 

[Geddagod]

I didn't say that it contradicts the older info. The two leaks aren't comparable because they talk about different things. The older leak speaks of ST perf gain over Raptor Lake, and the newer leak is about (presumably) overall performance over MTL at same core count. If we go by this statement, then it is reasonable to assume that MLID is talking about ARL-H and not ARL-S in the second leak.

However, like I said before, newer leaks from a credible leaker usually takes precedence over older leaks from the same leaker. However, since MLID is all over the place when it comes to accuracy of his leaks, it is hard to make a judgment call.

Ok.

 

[ondma]

wut

I don't think the increased cache sizes did much. In gaming the increased L2s just had an increased PPC of like ~5% in gaming. It was mostly just clocks + more cores.

If ARL-R had increased cache sizes they wouldn't be calling it ARL-R, it would be called something else IMO.

Raptor Lake technically had "more cores", but they were E cores. I dont think that had much if anything to do with gaming. Hopefully, though, Intel (and the game developers) will eventually find a way to utilize the E cores. Otherwise, Intel will be in serious trouble if cpu demands in future games with ever more powerful GPUs exceed what 8 big cores can deliver.

 

[Joe NYC]

Intel "may" have a contingency plan to deal with a repeat of a Zen 3 like situation by now. They kinda kept themselves competitive with Raptor Lake by increasing the cache sizes. Who knows what they might do with Arrow Lake. Maybe ditch the original ARL design and go with the Refresh design with increased cache sizes for the ARL debut? They could also include a more powerful iGPU tile with each ARL desktop part and poke fun at AMD's puny iGPU? They could even go with a hybrid strategy where the original ARL design is used in lower end CPUs with disabled cores to avoid chucking the already manufactured dies and keeping the refresh dies for their high end parts.

Arrow Lake is already set in stone. Intel most likely already has silicon back from the fab.

 

[SiliconFly]

iGPU is recycled from MTL, as is SOC tile and the rest of the setup.

When I said legacy, I meant pre-MTL (like their 14nm, 10, 7, GLC, Gracemont, Monolithic, etc).

Idk how many times I say this, and you just ignore it, but again, it's literally only MLID who expects a high performance uplift. Literally everyone else, Xino, Raichu, etc etc all say the perf uplift is gonna be ~10% over RPL.

Oh (gedda) god! This is one argument that can actually never be settled until we see real benchmarks. I'm glad the majority has settled on ~10%. Even I think so too. But it's still just speculation and I would prefer to keep an open mind. At this point, I'm not completely ruling out a slight regression either!

That's such an over reaction lol.

Just interpreting Pat's own words. He said they're "all in" with their new stuff. So, if it bonks, they're "all out". Just being logical!

Who said it was low volume?

ARL is their mainstream client. We need to see how and where LNL fits in.

That market has been there for a while, Intel has just been a very bad participant in it. Apple is much better at that.

Very true.

 

[SiliconFly]

Arrow Lake is already set in stone. Intel most likely already has silicon back from the fab.

True. Tape-in to manufacturing easily takes more than year. Can't do major changes this late in the cycle. Everything would be locked in.

I'm just going of this tbh
View attachment 95822

This diagram is missing tape-out.

 

[igor_kavinski]

True. Tape-in to manufacturing easily takes more than year. Can't do major changes this late in the cycle. Everything would be locked in.

What if they have TWO ARL designs? One easier/cheaper and the other harder/expensive and they decide to go with the harder one if they absolutely need to.

 

[Joe NYC]

What if they have TWO ARL designs? One easier/cheaper and the other harder/expensive and they decide to go with the harder one if they absolutely need to.

Intel already has 3 designs for 2024: MTL, ARL, LNL.
Within those 3, they have variants with different P+E cores.

I highly doubt they would have a 4th, separate design. There is always next year (or so), and Panther Lake (or whatever comes after ARL).

 

[dullard]

That makes no sense in context of what Raichu said. He said and I quote "it's easy to get 3 meters".

Could that mean that 3 nm was not as difficult as expected? Could it mean that 3 nm has higher yield than originally expected? Could it mean that the 3 nm foundry is available to use? Could it mean that the 3 nm foundry is underutilized so you don't have to fight other companies for limited space? I could see many ways that the context of that phrase could apply to 3 nm.

 

[dullard]

It literally took me about 20 seconds to find these:

View attachment 95725

View attachment 95726

AI Engine Technology

www.xilinx.com

Thanks, I'll add that to my post. That site did not show up in any of my searches. Not to nitpick or anything, is FP16 covered? It is quite similar to BFLOAT16, but the discussion was specifically about FP16.

 

[reggie_fils_aime]

MLID likes to play both sides with intel leaks, he'll say the best possible thing about prospective performance while also being one of the first to report when the new process has eaten you-know-what on the pavement in front of a school bus of gawking children. I always take those perf figures with a mountain of salt, but he's been pretty rock solid on the actual comings and goings of intel architectures for a minute now.

With that said, you can't give too much creedence to the sky-high projections for IPC/ST. He's definitely better on the negative side of the rumor mill.