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 (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 ?	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.)

 

[Magio]

David's argument was that Intel needs to be significantly better than TSMC to steal business from them

I don't see why. TSMC's cutting edge nodes are essentially Apple exclusive for 1 year, if Intel can offer meaningful capacity on even comparable (not even better, not even necessarily as good) nodes they'll have no issue finding customers (they of course also need to make their cutting edge nodes price competitive, but that's another discussion). If we're talking about stealing Apple from TSMC then yeah that's not even slightly on the roadmap today, but there's plenty of inroads to be made with all other major chip designers.

They're not gonna get juicy contracts on 18A (the first iteration of it) because no major chipmaker is gonna bet their halo products on Intel Foundry today, but if 18A rollout (PTL/CWF/first external designs) pans out well and the node is competitive with N3E/N2 I'd bet they'll start seeing significant interest from most of the top players for 18A-P, 14A and even 3/3E.

 

[cebri1]

The issue for Intel is that, even if their processes are better, the capacity is still very low and will be for the foreseeable future, and a lot of it is going to eaten away by their server products.

 

[Ghostsonplanets]

IMO Intel 18A needs to happen first and on schedule, so that the industry as a whole sees that they're on the right track. Then there will be higher adoption. Intel themselves said that Intel 14A is where they think they can start winning mobile contracts, for example.

 

[static shock]

Skymont is a marvel.

 

[GTracing]

Meteorlake has a low power island, a "proper" one with it on an entirely different die.* It saves an amazing 150mW in video playback, a whopping 4% increase in battery life.

Yeah, Meteor Lake has a low power island, and it does better in video playback than in other battery benchmarks, compared to Phoenix and Raptor Lake mobile.

Assuming you got that number from this video, the number they quote is 150-200mW, which is a 10-15% difference in package power.

 

[SiliconFly]

Twitter/X is abuzz with Zen5 benchmarks. And it’s nowhere near what was suggested before. Even in gaming, it’s only almost on par with 14th gen Raptor refresh.



This gives even more headroom for ARL than expected.

 

[KompuKare]

The big news with Zen5 is efficiency. Hopefully ARL will follow and we won't get 300W CPUs outside of server or workstation parts again - well at least for a few years.

But one CB MT benchmark in today's reviews did show a sizeable gain from 65W stock vs 120W+ so all it takes is one of the x86 vendors thinking that they have to squeeze every last drop to "win" some benchmark and we will be back at square one. Hopefully all the RPL failures will also stop this clock too high nonsense.

 

[SiliconFly]

The big news with Zen5 is efficiency. Hopefully ARL will follow and we won't get 300W CPUs outside of server or workstation parts again - well at least for a few years.

But one CB MT benchmark in today's reviews did show a sizeable gain from 65W stock vs 120W+ so all it takes is one of the x86 vendors thinking that they have to squeeze every last drop to "win" some benchmark and we will be back at square one. Hopefully all the RPL failures will also stop this clock too high nonsense.

Thankfully ARL tops out at 5.7GHz. Hope they dial it down even more in future gens.

 

[Philste]

The big news with Zen5 is efficiency.

Not really, it's slightly more efficient than ZEN4, but nothing surprising. Don't let them food you with the 9700X/7700X comparisons. To see the actual ZEN5 efficiency gains, the 9700X/7700 comparison is better. It's like 10-15% more efficient.

 

[ErichZannMusician]

Videocardz has an article about a Chinese leaker who supposedly attended a closed-door event organized by Asus and Intel.

Allegedly, from the summary of his comments, Arrow Lake uses at least 100 watts less power than Raptor Lake while running at high frequencies and won't have any issues that is plaguing RPL.

Performance is said to be impressive

 

[Hitman928]

Twitter/X is abuzz with Zen5 benchmarks. And it’s nowhere near what was suggested before. Even in gaming, it’s only almost on par with 14th gen Raptor refresh.

This gives even more headroom for ARL than expected.

I think ARL may have a hard time in getting significant gaming improvement over RPL. The clock regression and added latency from the tiles will hurt potential gaming performance. Hopefully I'm wrong though and they are able to negate these issues.

 

[Philste]

I think ARL may have a hard time in getting significant gaming improvement over RPL.

Yeah, ZEN5 is bad but I don't see Intel making a bigger jump. We probably end up at around a draw between ZEN5X3D and ARL. And that is like 5% above current ZEN4X3D lol.

 

[H433x0n]

The big news with Zen5 is efficiency. Hopefully ARL will follow and we won't get 300W CPUs outside of server or workstation parts again - well at least for a few years.

But one CB MT benchmark in today's reviews did show a sizeable gain from 65W stock vs 120W+ so all it takes is one of the x86 vendors thinking that they have to squeeze every last drop to "win" some benchmark and we will be back at square one. Hopefully all the RPL failures will also stop this clock too high nonsense.

As far as I can tell, there is no apples/apples tests for efficiency.

Yeah, ZEN5 is bad but I don't see Intel making a bigger jump. We probably end up at around a draw between ZEN5X3D and ARL. And that is like 5% above current ZEN4X3D lol.

Sadly the most likely outcome.

 

[ondma]

Twitter/X is abuzz with Zen5 benchmarks. And it’s nowhere near what was suggested before. Even in gaming, it’s only almost on par with 14th gen Raptor refresh.

View attachment 104677

This gives even more headroom for ARL than expected.

View attachment 104678

Tom's Hardware shows quite different results, while Anand's limited review is more comparable to this data. tom's review
Bottom line, we clearly need more reviews, including Zen 5 top end models with more cache and slightly higher clock speed, to get a clear estimate of Zen 5 gaming performance. With an estimated 15% IPC uplift, I cant believe the gaming performance will be as bad as this twitter post indicates.

 

[SiliconFly]

I think ARL may have a hard time in getting significant gaming improvement over RPL. The clock regression and added latency from the tiles will hurt potential gaming performance. Hopefully I'm wrong though and they are able to negate these issues.

The clock regression is overblown if you ask me. It's roughly 5% at best. Not a big worry. And the tile penalty shouldn't hurt gaming fps. It may increase the frame lag a bit but not by much i think.

Also, there's a possibility LNC may do better than previous gen in games (clock-for-clock) considering all the architectural improvements.

 

[SiliconFly]

Tom's Hardware shows quite different results, while Anand's limited review is more comparable to this data. tom's review
Bottom line, we clearly need more reviews, including Zen 5 top end models with more cache and slightly higher clock speed, to get a clear estimate of Zen 5 gaming performance. With an estimated 15% IPC uplift, I cant believe the gaming performance will be as bad as this twitter post indicates.

Surprisingly, this generation for both the camps, the expectations vs reality is pretty much a disaster. If ARL final doesn't pull a rabbit out of the hat, there's still a possibility the top sku may lag behind competition's top X3D parts.

 

[Doug S]

David's argument was that Intel needs to be significantly better than TSMC to steal business from them. The fact that laws of physics seem to deny us progress on further nodes does not change the sales and marketing conundrum for Intel when competing with TSMC.

Just as he said, AMD executed very well for many years before they started getting more customers in servers and mobile, and they're still facing a head wind. Inertia is huge in this field.

I don't buy that argument. Inertia applied for Intel and AMD because they were at best equal solutions in the non technical realm. Actually less than equal, since Intel controlled their means of production while AMD had outsourced their fabs, which is the sort of thing OEMs like Dell had to take into account back when AMD was dependent on GF and wouldn't have been able to take significantly larger chunks of the PC market at the time because GF simply lacked such capacity.

The situation between TSMC and Intel is completely different. Not only does TSMC have some geopolitical risk associated with it due to their location (which they can partially defray by putting true leading edge fabs elsewhere rather than N+1 fabs as currently) but Intel has a "made in the USA" advantage. That's not only extremely important to defense/government and defense/government adjacent markets, but it is something that can make a difference in the minds of many end product consumers in the US.

I'm actually kind of surprised Intel hasn't tried to play that up on its PC CPUs. It doesn't seem like it would be that difficult to insure that Dell, HP, etc. PCs destined for sale in the US get CPUs made in US fabs and put a little American flag next to the Intel logo on the holo sticker. They could do that elsewhere too, with lesser impact, though they'd have to insure they don't include that in countries where it would be seen as a negative. That would be a better marketing campaign than "Intel inside" was, and is something AMD (and Qualcomm) can't counter, at least not in the near future.

I think all Intel needs to do is roughly match TSMC in the price/power/performance domain (i.e. being slightly ahead in one and slightly behind in the other two would be good enough in my mind, just don't trail in all three) to be considered a superior solution both due to the "not putting all our eggs in a basket located on an island next to a potentially aggressive superpower that claims its as its own territory" and "rah rah made in the USA" factors.

Intel's biggest challenge is not beating TSMC, significantly or even slightly, it is proving that they can execute on a process roadmap and produce chips in volume at the scheduled time. Then if they achieve both those goals to invest (or rather have invested in the previous several years to be ready) in capacity sufficient to be able to accept the new major customers that would come knocking on their door.

 

[ondma]

Surprisingly, this generation for both the camps, the expectations vs reality is pretty much a disaster. If ARL final doesn't pull a rabbit out of the hat, there's still a possibility the top sku may lag behind competition's top X3D parts.

I dont think there is any doubt of that. I would consider it a pretty good result if ARL can beat Zen 4 x3d in gaming, much less Zen 5.

Edit: and also avoid or greatly reduce the stability and power/temperature issues of 14th gen.

 

[Hitman928]

The clock regression is overblown if you ask me. It's roughly 5% at best. Not a big worry. And the tile penalty shouldn't hurt gaming fps. It may increase the frame lag a bit but not by much i think.

Also, there's a possibility LNC may do better than previous gen in games considering all the architectural improvements.

Not sure what you mean by frame lag, but games are very latency sensitive.

If LNC has 5% regression in frequency, that means the performance drops from mid-double to single digit percentage increase. Add on that a potential tile latency penalty, and it may end up not being a very good gaming improvement, that's all I was saying.

 

[DavidC1]

Assuming you got that number from this video, the number they quote is 150-200mW, which is a 10-15% difference in package power.

10-15% difference in package power is 4-5% in system power. The changes at the package level can induce a change elsewhere, but Meteorlake battery life tests basically confirm they basically are back to the pre-Icelake era. They regressed since then.

Not sure what you mean by frame lag, but games are very latency sensitive.

If LNC has 5% regression in frequency, that means the performance drops from mid-double to single digit percentage increase. Add on that a potential tile latency penalty, and it may end up not being a very good gaming improvement, that's all I was saying.

How about you see the tests on a desktop platform where things can be truly apples-to-apples before settling on the "tile latency" thing? Laptop parts are always slower than Desktop parts, and with Meteorlake we only have laptops with tremendous variations.

 

[Hitman928]

How about you see the tests on a desktop platform where things can be truly apples-to-apples before settling on the "tile latency" thing? Laptop parts are always slower than Desktop parts, and with Meteorlake we only have laptops with tremendous variations.

Am I not allowed to speculate now?

 

[DavidC1]

The situation between TSMC and Intel is completely different. Not only does TSMC have some geopolitical risk associated with it due to their location (which they can partially defray by putting true leading edge fabs elsewhere rather than N+1 fabs as currently) but Intel has a "made in the USA" advantage.

"Made in USA" didn't make a difference big enough for consumers to stop buying endless Chinese stuff.

I think you overestimate the impact since many are actually disappointed in the direction the US is going. Like disregarding Veterans, and parading endless taxes as a Good Thing when even Income Taxes were practically nonexistent prior to WWI. No real reason to be a patriot anymore.

Also even if consumers do, companies wanting to fabricate is a different story. They are also under fire from investors and stockholders to have more profits are they not? Businesses will throw away such stuff when their "business" is on the line. Intel's own projections about future volume is a big evidence that even they know it'll be slow and steady at best.

The only time where they'll be forced en masse when Taiwan takeover actually happens, which by then it won't matter because you have way more important things to worry about(like death, food, water, electricity).

 

[GTracing]

10-15% difference in package power is 4-5% in system power. The changes at the package level can induce a change elsewhere, but Meteorlake battery life tests basically confirm they basically are back to the pre-Icelake era. They regressed since then.

Seems like that depends way to much on stuff other than the SOC to nail it down to a figure as specific as 4-5%.

 

[DavidC1]

Seems like that depends way to much on stuff other than the SOC to nail it down to a figure as specific as 4-5%.

50WHr, 13 hour battery life = 3.84W. Let's say 3.5W since you can't use 100%.

3.5/3.35 = 4-5%. It really doesn't matter. The LPE in MTL is a fail. You are doing more than Local Video Playback anyway. Most have multiple tabs and even other programs open and streaming online from sources such as Youtube = No LPE.

 

[GTracing]

50WHr, 13 hour battery life = 3.84W. Let's say 3.5W since you can't use 100%.

3.5/3.35 = 4-5%. It really doesn't matter. The LPE in MTL is a fail. You are doing more than Local Video Playback anyway. Most have multiple tabs and even other programs open and streaming online from sources such as Youtube = No LPE.

That seems like a over generalization to me. Plenty of people watch Netflix on their laptop in full screen.