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]

How about a nice solid, reliable, CB R23 score with the clock speed and power draw during the run? Those three numbers would tell us about 95% of what we need to know.

And then you find out at launch that you really need a 420mm AIO cooler to keep the CPU below 105C to maintain those numbers!

 

[FlameTail]

Nah, RibbonFET and PowerVia for the win!

 

[Hulk]

And then you find out at launch that you really need a 420mm AIO cooler to keep the CPU below 105C to maintain those numbers!

Power draw would tell us about cooling requirements

 

[Ghostsonplanets]

Has anyone found a review of a laptop with MTL-U? I saw some devices being sold with the U die, but I swear I can't find any reviews or some in-depth analysis of the laptops with the U series.

I find it weird because Intel has already listed a binned MTL U SoC (Core 5 Ultra 115U) but we barely saw any U device in the wild.

Will we have to wait until ARL-U for some "mass" availability of MTL-U? Or did OEMs simply have chosen to forego MTL-U and adopt LNL-MX instead?

 

[SiliconFly]

Is there no more leaks on Arrow Lake?

We're still waiting for Meteor Lake leaks!

 

[H433x0n]

And then you find out at launch that you really need a 420mm AIO cooler to keep the CPU below 105C to maintain those numbers!

Those days are over - the PL2 is <180W.

Ironically it will run hotter than RPL because of heat density but using a big cooler will have minimal effect.

 

[Hulk]

Those days are over - the PL2 is <180W.

Ironically it will run hotter than RPL because of heat density but using a big cooler will have minimal effect.

Yes, exactly. It's not the amount of heat that is the issue but rather trying to increase the heat transfer coefficient. Hopefully they'll improve the TIM or better yet sell delidded versions.

 

[coercitiv]

If Intel feels the pressure on the desktop platform, they will stay at 240W+ even with Arrow Lake. I know we've seen the 180W rating in some leaks, I know heat density is a thing, but I also know that Intel always bins and pushes power if they think they have a chance to win or tie in press reviews. Even today the 14900KS is not really cut out for the 320W power profile, not with conventional cooling anyway.

I expect Intel to go this route with Alder Lake, lowering the base profile to 180W and introducing an "enhanced" profile for 240W+ in the review guidelines. Motherboard makers will probably disregard both spec profiles and go YOLO for all of their Z boards and maybe even their premium B/H gaming boards (so we'll get early reviews with absurd power consumption and temps). The 180W limit should accomplish one good thing though, which is to stop costs from exploding for value/budget boards that aim min spec compatibility and pave the way for more reasonable power profiles in the future.

I just don't see Intel going back to much lower power limits in one swing while not winning decisively on the efficiency side.

 

[H433x0n]

If Intel feels the pressure on the desktop platform, they will stay at 240W+ even with Arrow Lake. I know we've seen the 180W rating in some leaks, I know heat density is a thing, but I also know that Intel always bins and pushes power if they think they have a chance to win or tie in press reviews. Even today the 14900KS is not really cut out for the 320W power profile, not with conventional cooling anyway.

I expect Intel to go this route with Alder Lake, lowering the base profile to 180W and introducing an "enhanced" profile for 240W+ in the review guidelines. Motherboard makers will probably disregard both spec profiles and go YOLO for all of their Z boards and maybe even their premium B/H gaming boards (so we'll get early reviews with absurd power consumption and temps). The 180W limit should accomplish one good thing though, which is to stop costs from exploding for value/budget boards that aim min spec compatibility and pave the way for more reasonable power profiles in the future.

I just don't see Intel going back to much lower power limits in one swing while not winning decisively on the efficiency side.

It’s not going to be physically possible. The CPU tile will be <150mm2 of TSMC silicon that’s not known to be inefficient or leaky.

I’d be shocked if even with uncapped current and power limits we see CPU package power exceed 240W.

 

[SpudLobby]

https://twitter.com/x/status/1776005067117592756

1st Lunar Lake mode ntel(R) Core(TM) Ultra 5 234V 2.10GHz

Intel's Lunar Lake will have multiple SKUs for different segments (based on power) it seems based on both the early and recent leaks. But what's really interesting is these clocks don't look great, and we know LNC is on N3B, uses memory on package and has some low power tech, and isn't going to be a humongous IPC upgrade given how long it's been. It will get them right in line with like Firestorm or the X3/4 at best on IPC and probably not.

So now we've got chips going to about 2.8GHz boost (from the Samsung Galaxy Book 5 Pro with LNL) or 3.1GHz boost which are going to have low teens to mid teens IPC or rather perf/GHz gains over RPC/GLC. If these are Intel's lowest/fanless SKU's, that's funny, because I bet the full platform power on ST isn't great and that was about all they could do.

Also at 3.1GHz with a mid teens perf/GHz upgrade they won't even match the M1 on N5 for fanless ST, and I bet LNL ST at 3.1GHz consumes as much or a bit more power than an M1/2 at the same performance and likewise more than the Snapdragon X Elite (iso-performance).

This would be fine if Intel didn't have a humongous 140mm^2 N3B die and a 46mm^2 IO tile on N6 for a 4+4 part. If you're going to use that kind of area and node for a 5-30W part, the power/performance had better be pretty competitive. It seems like it's just better than anything else Intel has put out and at least "acceptable" to good. Now, there should be some peak 3.8-4GHz SKUs for the 17-30W range, and I'd also like to see what those power/performances look like on ST and NT. Still, I don't think it's going to look impressive in light of the competition or the node/size/resources Intel has thrown into this.

 

[SpudLobby]

Granted it's an "early" stepping but I bet we're pretty close to some of these being finalized if Intel wants a LNL limited release in Q4 2024.

 

[SpudLobby]

Performant in GB6's "MT" (which is really a limited cores test, rather than a reflection of multi core capability) anyway. No info about efficiency in that one.

Yeah this is true and part of why they've kept all the performance/watt curves on MT to GB6, we haven't seen a single one for CB24.

The ST one is utter dominance though.

 

[poke01]

Intel's Lunar Lake will have multiple SKUs for different segments (based on power) it seems based on both the early and recent leaks. But what's really interesting is these clocks don't look great, and we know LNC is on N3B, uses memory on package and has some low power tech, and isn't going to be a humongous IPC upgrade given how long it's been. It will get them right in line with like Firestorm or the X3/4 at best on IPC and probably not.

So now we've got chips going to about 2.8GHz boost (from the Samsung Galaxy Book 5 Pro with LNL) or 3.1GHz boost which are going to have low teens to mid teens IPC or rather perf/GHz gains over RPC/GLC. If these are Intel's lowest/fanless SKU's, that's funny, because I bet the full platform power on ST isn't great and that was about all they could do.

Also at 3.1GHz with a mid teens perf/GHz upgrade they won't even match the M1 on N5 for fanless ST, and I bet LNL ST at 3.1GHz consumes as much or a bit more power than an M1/2 at the same performance and likewise more than the Snapdragon X Elite (iso-performance).

This would be fine if Intel didn't have a humongous 140mm^2 N3B die and a 46mm^2 IO tile on N6 for a 4+4 part. If you're going to use that kind of area and node for a 5-30W part, the power/performance had better be pretty competitive. It seems like it's just better than anything else Intel has put out and at least "acceptable" to good. Now, there should be some peak 3.8-4GHz SKUs for the 17-30W range, and I'd also like to see what those power/performances look like on ST and NT. Still, I don't think it's going to look impressive in light of the competition or the node/size/resources Intel has thrown into this.

If Intel fails with LNL then I got no hope for them for the low power SoCs from them. They need to match or exceed M3 in ST(GB6) because optics.

 

[mikk]

Intel's Lunar Lake will have multiple SKUs for different segments (based on power) it seems based on both the early and recent leaks. But what's really interesting is these clocks don't look great, and we know LNC is on N3B, uses memory on package and has some low power tech, and isn't going to be a humongous IPC upgrade given how long it's been. It will get them right in line with like Firestorm or the X3/4 at best on IPC and probably not.

The detected max frequency on this page is a pretty much meaningless value there. Have a look to Rapor Lake, Alder Lake, Arrow Lake-S, Tigerlake and so on, they are mostly below 3000 Mhz. Dit it mean they only boosted to 3 Ghz in real products? Certainly not. We don't know what it means, is it a PL1 or PL2 max speed and what power. Or is it even an average speed from P+E cores combined. Pretty much meaningless there. You read too much into it.

 

[SpudLobby]

If Intel fails with LNL then I got no hope for them for the low power SoCs from them. They need to match or exceed M3 in ST(GB6) because optics.

They won’t exceed M3 ST on GB6 most likely, and even if they got close to the M2 or the X Elite I bet power won’t be as impressive.

Even at 3.9GHz which should be the top SKU frequency for 17-30W stuff, if Lunar Lake has a 15% IPC upgrade over Meteor Lake on stuff like GB6 or CB24, it would put them a bit short of the M2.

Like at 4.7GHz (look at the .GB6 file if you have a account by appending it) MTL does about 2350: https://browser.geekbench.com/v6/cpu/5605095

Roughly 500 Perf/GHz. This would improve at lower clocks TBF, maybe 5-10%, and more cache would help.

So, we’ll be very charitable and say that until 3.9GHz MTL has about 550 Perf/GHz in GB6. A 10% boost and I’m not even sure if that’s really how much they’d get from lower clocks (but it does get worse at higher clocks and like ADL from 4.8GHz to mid 3GHz loses about 8-9% but then stabilizes). Well, now let’s give them a 15% IPC boost with Lion Cove. That’s also possibly too generous per rumors. So I’m pretty fair here.



550 (MTL GB6 IPC with an adjustment for lower clocks) * 1.15 (15% IPC gains with LNC) = 632.5.

632.5 * 3.9 for the rumored 3.9GHz at the top LNL part, getting around roughly what Apple and QC run their frequencies, = 2,466.75 GB6 ST @ 3.9GHz in LNL.

FOR the rumored top 3.9GHz part, and assuming a 15% IPC gain with LNC in LNL over MTL.

For context, the M1 does about 2350-2380 GB6 ST and the M2 does 2600-2700+. The M3 is 3100-3180.

The X Elite is in the 2500-2900 range on depending on the frequency, but I with a pretty steep power slope so dropping to 9W platform takes off like 10% of the 14W performance and matches Intel/AMD at much higher power still.


And: for Intel we don’t even know what the power will be like either on ST. If the fanless parts are running at 3.1GHz I wouldn’t be surprised if that’s because the boost for that freq runs up to 6-8W.

Above it looks like their best part could match the M1 on ST performance and trail Qualcomm and the M2, but I bet you they’ll use more power for it.

MT I think they will be able to beat the M2 albeit at more power. Battery life will be much better than anything previously Intel but not impressive in a world with the X Elite or even Strix.

 

[SpudLobby]

The detected max frequency on this page is a pretty much meaningless value there. Have a look to Rapor Lake, Alder Lake, Arrow Lake-S, Tigerlake and so on, they are mostly below 3000 Mhz. Dit it mean they only boosted to 3 Ghz in real products? Certainly not. We don't know what it means, is it a PL1 or PL2 max speed and what power. Or is it even an average speed from P+E cores combined. Pretty much meaningless there. You read too much into it.

Nah. I’ve already seen from other sources the top freq for LNL’s lower power parts will be in the upper 2’s to lower 3’s. I have independent sources. They could be wrong, but still.

There will be higher powered parts though.

 

[coercitiv]

If Intel fails with LNL then I got no hope for them for the low power SoCs from them. They need to match or exceed M3 in ST(GB6) because optics.

I think this has been said repeatedly before, but it bears repeating: no single product matters anymore, only execution/iteration speed. LNL can be as mediocre as they come, the important part is how fast they're able to introduce the next meaningful upgrade. Single achievements matter less than cadence nowadays. Folks should not worry when Intel introduces a new chip that fails to meet expectations, they should worry when Intel introduces the refresh of that chip. That's when they fall behind. Big jumps don't cut it anymore, there's just too much to juggle in terms of uncore changes, especially in the context of disaggregation.

Intel as a collective have this modus operandi in heir heads, where their win condition is sheer technological prowess, they introduce something so advanced that the competition simply cannot handle. They don't just make better nodes or better packaging, they make the best node and the best packaging. They go for high cost / high reward / high risk moves. It probably worked well for them in the past so they made this part of their company culture, whether they acknowledge it internally or not (I have no idea if they do, my analysis is 100% from the outside). Lately they've shown some signs of improvement on this front, they seem to have stopped chasing for the next big thing and started working harder on fixing what went wrong yesterday. It's not clear to me if the arrogance is gone, but they seem to emulate humbleness good enough to translate it into action.

So don't lose hope for them, but don't get excited for tomorrow either. It's a process.

 

[mikk]

Nah. I’ve already seen from other sources the top freq for LNL’s lower power parts will be in the upper 2’s to lower 3’s. I have independent sources. They could be wrong, but still.

There will be higher powered parts though.

What other sources? The max frequency on this page is factual meaningless, you can see this on basically every other chip there. With your logic we could apply this to every other past generation which of course is nonsense. Furthermore we have seen a Sisoft entry from Lunar Lake last year with a reported max frequency of 3900 Mhz on an older stepping. How do you explain this?

 

[SpudLobby]

I think LNL will be fine as in the first Intel product that's at least passable from a battery life standpoint and responsiveness without blowing up fans/power. From what I can tell, I'd probably buy it if I had to grab an Intel or AMD part this year, pending Strix.

But I think it won't be really competitive relative to what Qualcomm has (and they will have cut-down 8c/10c parts that will still be very performant, efficient just less peak ST/MT) or Apple, to the extent that Apple is even in comparison here. If people think this is Intel's return to the throne they'll be in for a rude awakening.

 

[SpudLobby]

What other sources? The max frequency on this page is factual meaningless, you can see this on basically every other chip there. With your logic we could apply this to every other past generation which of course is nonsense. Furthermore we have seen a Sisoft entry from Lunar Lake last year with a reported max frequency of 3900 Mhz on an older stepping. How do you explain this?

Go check the chips n cheese discord. They have different SKUs.

 

[mikk]

Go check the chips n cheese discord. They have different SKUs.

What does it say to you? That the entry is useless without knowing what SKU and what power and so on?

 

[SpudLobby]

What does it say to you? That the entry is useless without knowing what SKU and what power and so on?

We can agree to disagree. Low 3GHz has been quoted though for the fanless parts ending in V or 4 or whatever.

 

[SpudLobby]

I’ll let Uzzi for ex come in on this if he wants to.

 

[Ghostsonplanets]

I thought LNL was supposed to only have 4 SKUs and a simplified product stack. Are Intel really doing different SKUs based on TDP? Jesus...

@SpudLobby Where does the 140mm² Compute tile die size comes from? Isn't that a bit big for a 4+4 design + iGPU on N3B? (I also thought the IO Tile would be N3B too. But N6 make more sense from an economic standpoint).

Anyway, I hope that LNL is a success for Intel. They don't need to match AMD or QCOM imo, but they need to execute on LNL promises, which was breakthrough x86 perf/W. If LNL can deliver a fanless x86 experience with great CPU and GPU performance, that's already enough to differentiate itself from others products and enable novel experiences and form factors from their partners. And the 17 - 30W SKU should be okay for those who want the traditional laptop experience but with a bit more of horsepower while still delivering amazing battery life.

 

[DavidC1]

MT I think they will be able to beat the M2 albeit at more power. Battery life will be much better than anything previously Intel but not impressive in a world with the X Elite or even Strix.

Strix technically isn't a big bar to pass if they can eliminate the regressions in battery life between Tiger to Raptor and improve on top of that. The gap between Tiger and ARM platforms are about 50% normalized to battery Whr, and is certainly achievable, as they've done it in the past.

It's been rare thus doubts are reasonable but substantial improvements have been done before. With the Tablet Atoms being comparable with ARM parts. In the Core platform they achieved 50% gain in one generation with Haswell and Broadwell brought another 30% or so for a total of 2x in just 2 years. The gap is such that I pretty much ignore pre-Haswell laptops when looking for used. It's a complete deal breaker for me. Such gains didn't occur since Pentium M in 2003!

Lunarlake from the beginning wasn't known for top perf/W, but it should be much better because TSMC's process itself is more suited for low power envelopes over clock speed and the Lion Cove core in it supposedly has enhancements likely to fit a low power part better.

With the Xe2 graphics part also a substantial perf/W leap, this is one CPU I have hope on.

Still, I don't think it's going to look impressive in light of the competition or the node/size/resources Intel has thrown into this.

They have done worse, such as Sapphire Rapids where it is big in both die size and complex in packaging, while performing like AMD part 2 years ago.

It is better to use die/package resources to squeeze out a win over not meeting it. Die/package costs are comparatively negligible for premium parts. Also, it's a step in the process of gaining true leadership.