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

Page 407 - 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
695
601
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 + 4E012 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: 24,004
  • LNL.png
    LNL.png
    881.8 KB · Views: 25,487
Last edited:

Hitman928

Diamond Member
Apr 15, 2012
6,123
10,527
136

It was in response to the "Arrow Lake in December" Rumor

So the date for QS to be sent out was week 34 according to an earlier post, but there was never any manufacturing date given for the QS tested. Could be that the schedule was wrong in the previous post or the sample isn’t quite a QS. If they are launching in October, I would think there are QSs already. They need at least 3 months to start a production run and get product to partners which means the QS samples should have already been made.
 
  • Like
Reactions: Tlh97 and inf64

KompuKare

Golden Member
Jul 28, 2009
1,179
1,448
136
Even if this QS is representative of final performance, there are still 2 very big unknowns. First is what memory config was used, this can make a significant difference in some benchmarks. The second is power/efficiency. I know it says that it was set to 250W, but we don't know how much it is actually using in CB (my guess would be 250W but we don't know for sure) and how much it uses in idle/ST scenarios.
Since ARL 253W can pull way more, a 250W "set" power is currently largely meaningless.
Actual used power is the thing I am interested in.

RPL went crazy, so hopefully ARL can dial that back a fair bit - hopefully current events help. My worry would be if ARL does not perform like they want, that Intel cranks power back up. I want an end to 300W+ CPUs especially since the last 100W+ can gain as little as 1%.

Can't wait for the drama if ARL is a complete wash in performance compared to RPL in single core with all the RPL drama still happening.

Maybe Intel can lower their RPL performance with their new microcode so ARL doesn't look so meh. 🤣
A wash in performance at 100W or so less power is not a bad thing though IMO. Won't get upgraders but a return to power sanity would be nice.
 

Det0x

Golden Member
Sep 11, 2014
1,263
3,998
136
15% in multi with all the hype for the 60% improved IPC of the e cores is even more so
Its also important to remember that +15% MT is vs a 13900k capped at 250w
A 13900k @ 250w only scores like 36-37k in Cinenbench R23

Reviewers have been running those 13900k's and 14900k's @ 350+ watts for them to hit the 40+k numbers we have been seeing the last two years
Reality have finally caught up with Intel it seems
 

SiliconFly

Golden Member
Mar 10, 2023
1,532
882
96
Yes, but we kinda knew the gains based on ARL leak a while ago that people dismissed as sandbagging. And the recent numbers are roughly on par with it.

The best we're going to get in single thread is 5-7%.
Will the final even achieve that 5-7%? It's sad.

I added an edit, but these results don't even show that. It's basically a wash for ST performance at best. Of course, retail samples could perform better but there probably won't be a big difference.
Looking at the benches, it's getting hard to believe whether we can expect more.

From Igorsleak: https://www.igorslab.de/en/intels-i...n-for-raptor-lake-s-refresh-and-arrow-lake-s/

Hmm, I guess it is.

From the Spec 1-copy results though, it only shows 2-3% gain over RPL-S Refresh which would be the 14900K.
It's looking real bad. At least, as of now.

If this QS is retail performance and it is under performing compared to projections, they will probably miss SPEC 1T performance projection as well, so even that 2-3% gain is in question. Not drawing a conclusion yet, we are still a few months off from retail release, just commenting on the results we have.
It definitely looks bad.

The separated core and memory tiles penalty rears its head in the browser benchmarks?
I don't think it's tile penalty. And the new memory subsys penalty is pretty much out of question either. It's just that the IPC is that bad in that specific QS.

Even if this QS is representative of final performance, there are still 2 very big unknowns. First is what memory config was used, this can make a significant difference in some benchmarks. The second is power/efficiency. I know it says that it was set to 250W, but we don't know how much it is actually using in CB (my guess would be 250W but we don't know for sure) and how much it uses in idle/ST scenarios.
Well, the only saving grace in that is there are some unknowns. But, is that enough to save the final. I'm losing hope. Looks like the P core team is full of snit.

(snit: A snit is a really bad mood. Meaning not keeping well at times.)

Did you all miss the fact that it says "preliminary"? Jaykihn himself posted that Arrow Lake QS is Week 34, which is 4 weeks from now. I don't know where this data is from, but definitely not from a QS.
But he really managed to crash the bandwagon. He's been wrong too many times. But when he claims QS, it matters and it's appalling. We can only hope he's wrong, but the chances of that is getting too slim.
 

KompuKare

Golden Member
Jul 28, 2009
1,179
1,448
136
I very much agree, and i have been telling the guys in this thread that have been hyping ~ +20% ST vs 14900K that they would be wise to lower their expectations :wink:
We will see close to launch, but there is a danger this hype train is about to deflate faster than the one for the other x86 vendor did.
15% in multi with all the hype for the 60% improved IPC of the e cores is even more so
But nobody needs SMT any more (as it hindering the huge ST gains ARL is about to unleash...) so Skymont to the rescue!

I do suspect that if there is any technical merit in dropping SMT (aside from it technically helping to close various speculative execution attacks), then it won't be there ARL but rather a few gens later. But that would only happen IMO if Intel completely drops SMT for servers too which seems risky.
 

controlflow

Member
Feb 17, 2015
161
275
136
15% in multi with all the hype for the 60% improved IPC of the e cores is even more so

The ST looks meh but the MT looks pretty reasonable. The ability to get better than RPL MT throughput with 24 instead of 32 threads is a win. Should lead to potentially better gains in mixed workloads which are more than single threaded but not quite stupidly parallel where they saturate all available threads.

If ARL can do all that with much better power efficiency, that makes it a decent improvement overall. Nothing to really write home about though.
 

SiliconFly

Golden Member
Mar 10, 2023
1,532
882
96
The ST looks meh but the MT looks pretty reasonable. The ability to get better than RPL MT throughput with 24 instead of 32 threads is a win. Should lead to potentially better gains in mixed workloads which are more than single threaded but not quite stupidly parallel where they saturate all available threads.

If ARL can do all that with much better power efficiency, that makes it a decent improvement overall. Nothing to really write home about though.
With the kinda amazing gains Skymont provides, an overall measly +18% MT cannot even be considered a decent gain. Something isn't adding up. I'm starting to think something's amiss. Is it really a QS? Or is the QS pushed to max properly? What are we missing? @Jaykihn has been wrong too many times. Is he repeating the same mistakes again? Is he even a real engineer?!?
 

inf64

Diamond Member
Mar 11, 2011
3,864
4,546
136
With the kinda amazing gains Skymont provides, an overall measly +18% MT cannot even be considered a decent gain. Something isn't adding up. I'm starting to think something's amiss. Is it really a QS? Or is the QS pushed to max properly? What are we missing? @Jaykihn has been wrong too many times. Is he repeating the same mistakes again? Is he even a real engineer?!?
Better be preprepared to be disappointed than to expect a miracle after we have decent leaks like the last one. It basically confirms what igorslab leaked one year ago, almost to the T.
 
  • Like
Reactions: Tlh97 and Joe NYC

dullard

Elite Member
May 21, 2001
25,479
3,976
126
We don't know the clock speeds of this QS or whatever it is, what's the point?
The point is to bash Intel. Regardless of power used, clock speeds, price, availability (or lack thereof), or anything else.

I'm a broken record, so I'll say it again. Arrow Lake is a massive iGPU increase, adding an NPU, and lower power. Dominantly better desktop performance isn't the point of Arrow Lake.
 
  • Like
Reactions: controlflow

DavidC1

Senior member
Dec 29, 2023
831
1,314
96
But nobody needs SMT any more (as it hindering the huge ST gains ARL is about to unleash...) so Skymont to the rescue!
It's BECAUSE Skymont has big gains they are able to get 15% MT gains.

In Raptorlake, 1P = 2E in MT remember? So 14900K is 8+8.

8 x 1.3 SMT + 8 = 18.4
8 x 1.15(LNC) + 8 x 1.5 = 21.2, or 15%
15% in multi with all the hype for the 60% improved IPC of the e cores is even more so
The Int gains will be lot less at 30% over Gracemont and Int is basically uarch.