Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads

Page 384 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Tigerick

Senior member
Apr 1, 2022
696
602
106
PPT1.jpg
PPT2.jpg
PPT3.jpg



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

ModelCode-NameDateTDPNodeTilesMain TileCPULP E-CoreLLCGPUXe-cores
Core Ultra 100UMeteor LakeQ4 202315 - 57 WIntel 4 + N5 + N64tCPU2P + 8E212 MBIntel Graphics4
?Lunar LakeQ4 202417 - 30 WN3B + N62CPU + GPU & IMC4P + 4E08 MBArc8
?Panther LakeQ1 2026 ??Intel 18A + N3E3CPU + MC4P + 8E4?Arc12



Comparison of die size of Each Tile of Meteor Lake, Arrow Lake, Lunar Lake and Panther Lake

Meteor LakeArrow Lake (20A)Arrow Lake (N3B)Lunar LakePanther Lake
PlatformMobile H/U OnlyDesktop OnlyDesktop & Mobile H&HXMobile U OnlyMobile H
Process NodeIntel 4Intel 20ATSMC N3BTSMC N3BIntel 18A
DateQ4 2023Q1 2025 ?Desktop-Q4-2024
H&HX-Q1-2025
Q4 2024Q1 2026 ?
Full Die6P + 8P6P + 8E ?8P + 16E4P + 4E4P + 8E
LLC24 MB24 MB ?36 MB ?12 MB?
tCPU66.48
tGPU44.45
SoC96.77
IOE44.45
Total252.15

LNL-MX.png

Intel Core Ultra 100 - Meteor Lake

INTEL-CORE-100-ULTRA-METEOR-LAKE-OFFCIAL-SLIDE-2.jpg

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.)



Clockspeed.png
 

Attachments

  • PantherLake.png
    PantherLake.png
    283.5 KB · Views: 24,006
  • LNL.png
    LNL.png
    881.8 KB · Views: 25,490
Last edited:

Hitman928

Diamond Member
Apr 15, 2012
6,186
10,693
136
Generally none of that was what we now call AI. But instead coding programmed by people hoping to appear artificially intelligent. Your examples are 5-year old technology: they certainly didn't have the newer AI models or NPUs to run them (Kroger's started in 2020).

Both McDonald's and Kroger programs were running earlier this year. Yes, it was probably based on "AI" tech that was a couple of years old, but that's how development works. If they want to use hardware and models coming out today, it will be at least a couple of years before you see it in retail stores and then it will need to probably have a second iteration at least and come down in cost before market wide deployment. Maybe I'm being pessimistic, but I just don't see it working well at scale before the end of the decade.
 

SiliconFly

Golden Member
Mar 10, 2023
1,541
897
96
No.

If it matches the 1650 Mobile, it will be good enough already.
Brett's video shows that it's actually well ahead of RTX 2050 & closer to RTX 3050 mobile already. There's a chance LNL final may catch up to 3050 considering these are beta drivers and final drivers will have further performance optimizations.

This changes the game. I never thought an ultra-portable laptop will actually be replacing lower-end and mid-range gaming laptops. And yes, Intel needs to work on their drivers, but it's already pretty good with tons of performance and stability improvements compared to the ones they had last year. Very stable and good. i.e, this time they might actually have a very good product during launch.
 

Ghostsonplanets

Senior member
Mar 1, 2024
691
1,113
96
Brett's video shows that it's actually well ahead of RTX 2050 & closer to RTX 3050 mobile already. There's a chance LNL final may catch up to 3050 considering these are beta drivers and final drivers will have further performance optimizations.
Based entirely on TimeSpy benchmarks, which is an area where Intel iGPs always overperformed when compared to real world performance.

RTX 3050 is a 2048 ALUs GPU with modern uArch, 128-bit memory bus and 112GB/s bandwidth, while running at 1.X GHz clocks (Depeding on TGP).

Lunar Lake is also a modern uArch wirh 1024 ALUs and running at 1.8 to 2GHz, with 128-bit shared UMA giving it ~137GB/s.

it's very unlikely that with two contemporary gpus, the one which is half as wide will match the wider one. Unless there's a serious clock difference between each other.

I don't doubt Lunar Lake top SKU can come close to some low-end RTX 2050/3050 30W SKUs that are clocked at <1GHz. But it's very unlikely that it will consistently outperform it. And this will be even more unfavorable for LNL when compares to higher TGP RTX 3050/2050.

Anyway, I'd argue it's better to wait and see for Lunar Lake performance in real gaming workloads before getting worked up. Meteor Lake iGPU was also overhyped due to ES TS test and in real world, it consistently match or trails behind AMD 780M solution.

I'm sure LNL BMG IP will be fantastic, but let's wait and see first.
 
  • Like
Reactions: HoveringStyle

mikk

Diamond Member
May 15, 2012
4,247
2,311
136
That is what the Golden Pig leak states. Timespy gives 30 W Lunar Lake 4150 points and 54 W Ryzen AI 9 HX 370 gets 4221 points. It is a 1.7% win for AMD over the preliminary Lunar Lake but at almost double the power. The 50 W RTX 3050 gets 4487 Timespy points, which again is a win (8%), but not much of a win for a higher powered discrete GPU.


I don't think it's a win in timespy unless Strix Point also had been tested with 16GB 1R RAM.
 

Geddagod

Golden Member
Dec 28, 2021
1,296
1,368
106
So? It does so at a much higher clock. What part of that don’t you get? How can Lunar be more efficient than M3?

Also the M3 scores much higher than 3100. It does around ~3200.
Well, just because something can clock lower with higher IPC doesn't always mean it will be more efficient... just look at GLC vs Zen 3. But ye, I agree, I highly, highly doubt LNC will be more efficient than a M3 P-core.
Perhaps I will be surprised though :)
 

DavidC1

Senior member
Dec 29, 2023
939
1,470
96
Artificial benchmarks are not necessarily indicative of gaming performance though. Intel is notorious for having poor drivers for games, although they seem to be devoting more resources to this as their gpus become more competitive.
That's true, but Battlemage adds instructions and fixes things that will improve real world gaming performance which the drivers have been hard at work mitigating them.

-Execute Indirect: Used in UE5 games, went from software to hardware
-Fast Clear
-SIMD16 for better compatibility:
When we launch you're gonna see dramatically better game compatibility.
-Another command, Fast Clear, is now supported in the Xe2 hardware, rather than having to be emulated in software as it was on Alchemist.
-Alongside further improvements for bandwidth and, importantly, utilisation.

3DMark benchmarks are not a terrible tool for showing overall performance but can be very limiting.

For example, in 3DMark, for iGPUs it usually doesn't get playable frame rates. Games have to. Therefore, it favors Alchemist where the utilization is only high enough under very demanding workloads. Games by virtue of needing playable frame rates, the demands are lower, thus you lose some performance.
 
  • Like
Reactions: mikk

DavidC1

Senior member
Dec 29, 2023
939
1,470
96
Seriously, numbers like these are utterly insane. This is MTL-U vs LNL package power... Which includes memory for LNL but not for MTL. I would genuinely be extremely surprised if LNL loses to SDXE on battery life, to be frank.

That's a huge selling point for thin and light devices.
Let's recap:
-Local Video Playback 1080P: 532mW

Kabylake, which is actually quite efficient in terms of battery life requires something like 0.9W for SoC and 0.6W for DRAM. The total system consumption on my laptop is 4.3W at 720P.

Based on that result, the system power consumption on my laptop would drop from ~4.3W to 3.2W. This assumes that other parts of the system does not improve, because I can tell you, controlling SoC power affects other parts of the system.

I can only get 0.5W for the SoC if the CPU load is near zero and nothing running. Of course DRAM is not zero.

690mW on Netflix 1080p.

Haswell might be repeated, again: https://www.trustedreviews.com/wp-content/uploads/sites/54/2013/06/haswell-battery-large-3.jpg

The only negative for Lunarlake seems to be they are a one-off chip. I hope they can carry that on for Pantherlake, but I'm not so sure.
 
  • Like
Reactions: mikk

TESKATLIPOKA

Platinum Member
May 1, 2020
2,522
3,037
136
So you have no use case, can't think of one, yet you specifically pick out one and only one of the multithreaded benchmarks in the leak to talk about? It just isn't a very relevant benchmark for this CPU's target market. I can't wait to play Cinebench on a MSI Claw! https://videocardz.com/newz/msi-con...graded-to-lunar-lake-no-plans-for-amd-version
You are the only one who is nitpicking about me using Cinebench.

Cinebench is widely used in reviews of laptops, It will be also used when Lunar Lake will be released, so I find It ridiculous how you claim It's useless for this SoC.
With this argument of yours I can pretty much claim that every single mobile CPU(SoC, APU), unless they have >80W TDP shouldn't be tested in CB, because they are not aimed for rendering etc.
In my books, Cinebench is a good test of nT performance and easy to find results, that's why I chose it and It doesn't matter If I use my laptop for rendering or not, It still shows nT performance, which can be compared.

I originally didn't bother with GeekBench MT for two reasons, I don't really consider It a good bench and the second is that for some reason 17W LNL has higher score than the 30W one.

But ok, If you are unhappy with Cinebench because LNL is behind in performance, so here It is:
LNL 17W vs 8840U 15W Link
5.4 MT 8805 vs 5.5MT 8205
LNL 30W vs 8840U 28W
5.4MT 8653 vs 5.5MT 10014
Wins at 17W but loses at 30W against Zen4.
BTW, not sure, what is the difference in 5.4 vs 5.5 version of Geekbench.

P.S. I honestly don't care what you will use your MSI Claw for
 
Last edited:
  • Like
Reactions: Tlh97 and Hulk

TESKATLIPOKA

Platinum Member
May 1, 2020
2,522
3,037
136
That is what the Golden Pig leak states. Timespy gives 30 W Lunar Lake 4150 points and 54 W Ryzen AI 9 HX 370 gets 4221 points. It is a 1.7% win for AMD over the preliminary Lunar Lake but at almost double the power. The 50 W RTX 3050 gets 4487 Timespy points, which again is a win (8%), but not much of a win for a higher powered discrete GPU.
It would have been more interesting If they compared Strix at 30W.
Not a bad performance in TimeSpy, but I still think RTX 3050 will be faster in games, thanks to much higher BW and let's nor forget Ampere is older tech.
 
Last edited:
  • Like
Reactions: Tlh97

PJVol

Senior member
May 25, 2020
707
632
136
It would have been more interesting If they compared Strix at 30W.
Not a bad performance in TimeSpy, but I still think RTX 3050 will be faster in games, thanks to much higher BW and let's nor forget Ampere is older tech
No one seems to mind that LNL iGPU performance shown in two least representative 3DMark tests, TS and WLE.
At least with ARC, TS score has little to do with gaming performance. The same goes for Nomad

The Fire Strike Extreme is way more suitable for comparing Intel and AMD GPUs in real-world gaming tasks.
 
Last edited:

inf64

Diamond Member
Mar 11, 2011
3,865
4,549
136
View attachment 102100

(From X/Twitter)

LNC has tons of improvement. We can expect a good ST uplift in ARL.
Well we already have ~14% IPC number directly from intel where they compare it to Redwood cove (which is ~3% slower than Raptor cove). Not sure what you consider "a good ST uplift"? If it can clock to ~5.8Ghz ST and does ~15% better than Raptor lake IPC wise (being generous here considering the previous IPC disclosure), it will have ~7-8% better ST performance.
 

SiliconFly

Golden Member
Mar 10, 2023
1,541
897
96
Well we already have ~14% IPC number directly from intel where they compare it to Redwood cove (which is ~3% slower than Raptor cove). Not sure what you consider "a good ST uplift"? If it can clock to ~5.8Ghz ST and does ~15% better than Raptor lake IPC wise (being generous here considering the previous IPC disclosure), it will have ~7-8% better ST performance.
Leaks strongly suggest Arrow Lake's LNC is different from Lunar Lake's LNC. That 14% uplift you mention is Lunar Lake's LNC w.r.t Redwood Cove. Expecting a lot more ST uplift from Arrow Lake's LNC. Even after adjusting for clock regression, we should be left with at least 15% more ST perf (maybe even upto 20%... who knows).
 

inf64

Diamond Member
Mar 11, 2011
3,865
4,549
136
Leaks strongly suggest Arrow Lake's LNC is different from Lunar Lake's LNC. That 14% uplift you mention is Lunar Lake's LNC w.r.t Redwood Cove. Expecting a lot more ST uplift from Arrow Lake's LNC. Even after adjusting for clock regression, we should be left with at least 15% more ST perf (maybe even upto 20%... who knows).
You are basing this all on fate I guess? I have seen zero stuff from anyone reputable that will suggest any big deviation in IPC going to ARL. Even the slides mentioned the optimizations in LNL LionCove were to maximize the ST performance of the core by getting rid of HT related structures.

Just to let you know, in order to get up to 20% ST uplift vs Raptor Cove, you would need almost 30% ST IPC uplift and 5.8Ghz ST clock. 30% IPC uplift is not gonna happen, there is no magic bullet that will make ARL version of LionCove gain 15% more IPC vs LNL version.
 

coercitiv

Diamond Member
Jan 24, 2014
6,677
14,272
136
Leaks strongly suggest Arrow Lake's LNC is different from Lunar Lake's LNC.
Apart from the bigger L2, what else do leaks "strongly suggest" is different?

LNC in LNL benefits from an on-die memory controller. MTL has the memory controller on the SoC. How's Arrow Lake going to fare in this regard, is the overall structural layout the same as MTL, with the mem controller still situated on the other tile?
 

SiliconFly

Golden Member
Mar 10, 2023
1,541
897
96
You are basing this all on fate I guess? I have seen zero stuff from anyone reputable that will suggest any big deviation in IPC going to ARL. Even the slides mentioned the optimizations in LNL LionCove were to maximize the ST performance of the core by getting rid of HT related structures.

Just to let you know, in order to get up to 20% ST uplift vs Raptor Cove, you would need almost 30% ST IPC uplift and 5.8Ghz ST clock. 30% IPC uplift is not gonna happen, there is no magic bullet that will make ARL version of LionCove gain 15% more IPC vs LNL version.
There has been discussion already stating that Lion Cove is not a single architecture, but an umbrella term for a group of designs based on a larger blueprint. Remember "Sea of Fubs to a Sea of Cells".

The P core team in their video clearly said that they've moved not only to industry standard design tools, but now the design is also process agnostic and also now they can rapidly slice and dice the design as they wish.

With so much of power and flexibility at their disposal, I don't think they're gonna just release the same LNC yet again with ARL (defeats the entire purpose). And it also aligns well with the initial rumors as well.

You can always claim they didn't explicitly state ARL's LNC will be different from LNL's LNC. But at the same time, I can also claim that they didn't explicitly state ARL's LNC will be the same as LNL's LNC. The jury is still out on that one.

And looking at all the published info, I believe ARL's LNC is gonna be more different than some assume it to be. 20% IPC uplift over RWC is entirely possible. Maybe more if they've done something cool.
 
  • Like
Reactions: AcrosTinus

ondma

Diamond Member
Mar 18, 2018
3,005
1,528
136
There has been discussion already stating that Lion Cove is not a single architecture, but an umbrella term for a group of designs based on a larger blueprint. Remember "Sea of Fubs to a Sea of Cells".

The P core team in their video clearly said that they've moved not only to industry standard design tools, but now the design is also process agnostic and also now they can rapidly slice and dice the design as they wish.

With so much of power and flexibility at their disposal, I don't think they're gonna just release the same LNC yet again with ARL (defeats the entire purpose). And it also aligns well with the initial rumors as well.

You can always claim they didn't explicitly state ARL's LNC will be different from LNL's LNC. But at the same time, I can also claim that they didn't explicitly state ARL's LNC will be the same as LNL's LNC. The jury is still out on that one.

And looking at all the published info, I believe ARL's LNC is gonna be more different than some assume it to be. 20% IPC uplift over RWC is entirely possible. Maybe more if they've done something cool.
Rental Units would be '"cool". ARL 8+32 would be interesting. Beast Lake 16+32 would be a killer. Royal Core might be cool. Will we ever see any of them? Seems like every time Intel supposedly has something "cool" in the pipeline it either is late, gets cancelled, or simply does not live up to the hype. Point being? Just because Lion Cove is potentially a flexible architecture, considering the way Intel has executed lately, I certainly don't expect them to pull some magic out of the hat and give LC in ARL more than a token increase in IPC compared to the same core in LL. At best perhaps they will be able to make some modifications to Lion Cove in RL-R to make it more performant.
 
  • Like
Reactions: Thunder 57