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

Tigerick · Aug 22, 2022

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 (N3B)	Lunar Lake	Panther Lake
Platform	Mobile H/U Only	Desktop & Mobile H&HX	Mobile U Only	Mobile H
Process Node	Intel 4	TSMC N3B	TSMC N3B	Intel 18A
Date	Q4 2023	Desktop-Q4-2024 H&HX-Q1-2025	Q4 2024	Q1 2026 ?
Full Die	6P + 8P	8P + 16E	4P + 4E	4P + 8E
LLC	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.)

511 · May 12, 2025

Tigerick said:
Source ?

I just heard from someone very reliable they are going to waste more diespace on NPUs 75-80 TOPS on Next Gen NPU.

OneEng2 · May 12, 2025

Fjodor2001 said:
As has been said, exactly the same thing that was mentioned by some during the 4C-forever phase. We passed that, and we will pass the 16C-forever phase too.

Well, there is also the problem that currently, the desktop is using DDR6400 (Arrow Lake) dual channel. I have been speculating for a while that Arrow Lake's success at CB2024 is more about bandwidth than a great compute capability (which is why it loses to Zen 5 in CB2023 which is less bandwidth constrained). This is with only 24 cores. More than doubling the core count on the same memory bandwidth seems like it may starve the cores.

Still, for massively parallel loads, Noval Lake will excel I am certain. I just question how many of those there are.

Fjodor2001 said:
This time it’s AMD that is in the lead, but stubbornly stuck at 16C. The exact same arguments are mentioned by the ST perf is all that matters crowd. And I think this time the tables will turn so it’ll be Intel that bumps core count to prove them wrong. Rumors say it’ll be to 52C, although with a mix of P/E/LPE cores.

Zen 6 will move up to 24c/48t, so they aren't "stuck" at 16c. Additionally, thread ripper has way more cores (and more memory bandwidth to go with it).

Nova Lake will reportedly have a CCD consisting of 8P, 16E and 4 LPE shared IOD cores. The 2 CCD design then gets you 52 cores.

Hitman928 said:
If there was really demand for really high core count CPUs, we'd see it in the sales figures pushing the currently highest core SKUs up the sales charts.

This exactly. Furthermore, if you keep improving the cores and adding more cores, you start running into problems with your memory bandwidth. You can only feed so many high performance cores with dual channel DDR6400.

Moving to a 4 channel memory design on the desktop seems really expensive and would be appealing to only a small number of users.

511 · May 12, 2025

OneEng2 said:
Well, there is also the problem that currently, the desktop is using DDR6400 (Arrow Lake) dual channel. I have been speculating for a while that Arrow Lake's success at CB2024 is more about bandwidth than a great compute capability (which is why it loses to Zen 5 in CB2023 which is less bandwidth constrained). This is with only 24 cores. More than doubling the core count on the same memory bandwidth seems like it may starve the cores.

Still, for massively parallel loads, Noval Lake will excel I am certain. I just question how many of those there are.

Zen 6 will move up to 24c/48t, so they aren't "stuck" at 16c. Additionally, thread ripper has way more cores (and more memory bandwidth to go with it).

Nova Lake will reportedly have a CCD consisting of 8P, 16E and 4 LPE shared IOD cores. The 2 CCD design then gets you 52 cores.

This exactly. Furthermore, if you keep improving the cores and adding more cores, you start running into problems with your memory bandwidth. You can only feed so many high performance cores with dual channel DDR6400.

Moving to a 4 channel memory design on the desktop seems really expensive and would be appealing to only a small number of users.

Just a point NVL is DDR5-8000 Standard

Thunder 57 · May 13, 2025

Fjodor2001 said:
Consumer CPUs have maxed out at 16C (or 24C for Intel) for several years, same as when we were stuck at 4C for several years.

In the 4C-forever era it was said by some that nobody needed or wanted more than 4C, and it would be pointless due to Amdahl’s law, so no point providing more than that to the consumer. Then AMD proved that to be wrong, and introduced 8C and soon after 16C CPUs. They did so to break the Intel lead at that time.

This time it’s AMD that is in the lead, but stubbornly stuck at 16C. The exact same arguments are mentioned by the ST perf is all that matters crowd. And I think this time the tables will turn so it’ll be Intel that bumps core count to prove them wrong. Rumors say it’ll be to 52C, although with a mix of P/E/LPE cores.

There really is not many other options to substantialy increase perf than via core count increase, since IPC and frequency increase are only improving perf by a measly ~5-7% per CPU generation currently.

In the 4C era everyone was running 4C. How many people are running 16C now? Very very few (Not counting E-cores since we are talking about AMD). Also, my memory is quite different from yours. I don't recall people saying 4C was fine and that we didn't need more. Quite the opposite in fact, people were tired of Intel keeping the mainstream at 4C with minimal gains generation after generation.

511 · May 13, 2025

Thunder 57 said:
In the 4C era everyone was running 4C. How many people are running 16C now? Very very few (Not counting E-cores since we are talking about AMD). Also, my memory is quite different from yours. I don't recall people saying 4C was fine and that we didn't need more. Quite the opposite in fact, people were tired of Intel keeping the mainstream at 4C with minimal gains generation after generation.

I am glad that era is gone and we are getting actual improvements

Thunder 57 · May 13, 2025

511 said:
I am glad that era is gone and we are getting actual improvements

Amazing what a little competition can do. If only we had a bit more of it in the GPU market.

511 · May 13, 2025

Thunder 57 said:
Amazing what a little competition can do. If only we had a bit more of it in the GPU market.

If you are talking about Nvidia they don't care unless someone put pressure on DC GPU Market.

dr1337 · May 13, 2025

Thunder 57 said:
In the 4C era everyone was running 4C

Yeah 2500k's with 4c/4t, on windows 7, playing games like counter-strike global offensive.

I had a 2700k at that time, and when I 'upgraded' to a 4690k I swear I felt the lack of SMT in every workload.

I also remember that time, that HEDT chips were relatively potent and anyone that had a core extreme chip, ie 3960x or later the 5930k would handily flex (even in gaming) on all the 4c/4t chips being mainstream at the time, despite being 2x-4x the cost. It's not a debate that having halo products matter in a competitive market.

Does everyone in the world need 32c/t right now? probably not, will they need it 10 years from now? More likely so. How do you sell more chips with higher performance? by scaling up your production. I mean just look at how threadripper transitioned from a consumer facing product segment into extremely high end as AMD was able to price it higher due to demand. Without increasing core density per product a company cannot lower their cost per core. And I don't believe anyone here actually thinks intel has ruined themselves with their hybrid p-core/e-core design.

Sure gaming would be better with a 12p core only chip clocked to the moon, but right now core ultra has a lead in productivity and all the usual suspects like Phoronix, Servethehome, and Puget recognize this. Because guess what, just like in the zen 1 times, gamers are not the only consumers of silicon.

Thunder 57 · May 13, 2025

dr1337 said:
Sure gaming would be better with a 12p core only chip clocked to the moon, but right now core ultra has a lead in productivity and all the usual suspects like Phoronix, Servethehome, and Puget recognize this. Because guess what, just like in the zen 1 times, gamers are not the only consumers of silicon.

It seems some people have a hard time grasping this. It's why I don't get why the Core Ultra is crapped on so badly. It pretty damn good in productivity. It's average at best in gaming though so it gets a bad rep.

jpiniero · May 13, 2025

Thunder 57 said:
It's average at best in gaming though so it gets a bad rep.

Which is what DIY cares about.

I think the Core Wars died mainly because it became more obvious that memory bandwidth and cache speed/size/latency matters a lot in gaming performance. If Intel wants to waste a lot of very expensive TSMC silicon on Cinebench Accelerators, that would be very much on brand but also a waste.

511 · May 13, 2025

Thunder 57 said:
It seems some people have a hard time grasping this. It's why I don't get why the Core Ultra is crapped on so badly. It pretty damn good in productivity. It's average at best in gaming though so it gets a bad rep.

the only issue with core ultra is regression in workloads outside gaming that's it otherwise it's fine.

MS_AT · May 13, 2025

dr1337 said:
right now core ultra has a lead in productivity and all the usual suspects like Phoronix, Servethehome, and Puget recognize this.

Do they? I mean Puget is its own thing but on Phoronix says this,

Of course then everyone defines productivity differently you can mix and match but core ultra having a lead in productivity seems a bit too general statement

Thunder 57 · May 13, 2025

jpiniero said:
Which is what DIY cares about.

I think the Core Wars died mainly because it became more obvious that memory bandwidth and cache speed/size/latency matters a lot in gaming performance. If Intel wants to waste a lot of very expensive TSMC silicon on Cinebench Accelerators, that would be very much on brand but also a waste.

That's largely true. It's not just Cinebench though. Look at Handbrake. Core Ultra chews through it. There are places it does well. Just not in gaming.

511 · May 13, 2025

MS_AT said:
Do they? I mean Puget is its own thing but on Phoronix says this,
View attachment 123711
Of course then everyone defines productivity differently you can mix and match but core ultra having a lead in productivity seems a bit too general statement

For Non AVX-512 workloads Phoronix suit contains AVX-512 so ofc AMD is going to win this.

Thunder 57 · May 13, 2025

511 said:
For Non AVX-512 workloads Phoronix suit contains AVX-512 so ofc AMD is going to win this.

To be fair who's fault is that? For awhile Intel had AVX512 and AMD didn't. Should we just pick and choose based on who is supporting what and when?

dr1337 · May 13, 2025

jpiniero said:
Which is what DIY cares about.

I don't disagree with this completely and while I don't wanna get too off topic in a intel thread, but back in the day a lot of people did buy bulldozer for some reason.

MS_AT · May 13, 2025

511 said:
For Non AVX-512 workloads Phoronix suit contains AVX-512 so ofc AMD is going to win this.

You have no idea how much I would like code compilation to be able to benefit from AVX512. But as I said, you can mix and match what is productivity for you, for me it's mostly code compilation and well it is, what it is. https://www.phoronix.com/review/amd-ryzen-9-9950x3d-linux/2

511 · May 13, 2025

Thunder 57 said:
To be fair who's fault is that? For awhile Intel had AVX512 and AMD didn't. Should we just pick and choose based on who is supporting what and when?

It is definitely Intel's fault but they are correcting it with NVL which is too far away.

MS_AT said:
You have no idea how much I would like code compilation to be able to benefit from AVX512. But as I said, you can mix and match what is productivity for you, for me it's mostly code compilation and well it is, what it is. https://www.phoronix.com/review/amd-ryzen-9-9950x3d-linux/2

Yeah i can see that Ryzen is a no brainer CPU for u.

A NVL 52 Core CPU Shipping For R&D from the cache

LightningZ71 · May 13, 2025

I don't get the stanning for 12 full fat cores for gaming. I understand wanting high ST performance and gobs of cache, but, I'm yet to see a game that can keep 8 P cores at 100% that is also held back in any way by the more modest performance of the E-cores. The VAST majority of games have one, maybe up to three threads that are performance determinant, in other words, those threads are keeping their processor cores pegged at 100% and game performance can not improve any more unless those threads get more horsepower. When you look at Intel's 6P+8e products and compare them to the 8p+12e processors and then clock normalize the two processors as the 8p will boost higher generally, the remaining performance difference, if any, can be largely attributed to the slight difference in L3 cache between them and the remainder left to the extra threads available in total.

This isn't the same for AMD as they, to date, don't really have e cores in mainstream products, save for Strix/Kraken/Phoenix2, and those are largely just down clocked normal cores from a logic point of view. Their extra threads are coming from SMT, which lives in contention with other threads and take a bit more of a hit at lower core counts because of it.

I personally want the processor manufacturers to produce more products that look like Strix Point to some degree. I want 4 cores that are just absolutely sold out for ST performance. Large, low latency caches and high clock speeds. Then, I want a sea of slower, fully capable cores that can run secondary and background threads all day long. The biggest thing is that I want them to share the same L3 so that moving threads and data around has minimal latency. Having 4 very high performance threads should be more than enough to keep everything going smoothly in games.

Tigerick · May 13, 2025

511 said:
It is definitely Intel's fault but they are correcting it with NVL which is too far away.

Yeah i can see that Ryzen is a no brainer CPU for u.

A NVL 52 Core CPU Shipping For R&D from the cache

View attachment 123714

36MB X 2 = 72MB L3 seems valid. 3.6Ghz refers to base clock or turbo?

Thunder 57 · May 13, 2025

LightningZ71 said:
I don't get the stanning for 12 full fat cores for gaming....

It's the same people that want a 9590X3D x 2 thinking it would be some godlike CPU but really don't understand that wouldn't be the case. These companies aren't stupid, they know how these types of products perform.

511 · May 13, 2025

Tigerick said:
36MB X 2 = 72MB L3 seems valid. 3.6Ghz refers to base clock or turbo?

Base Clock afaik

Thunder 57 · May 13, 2025

511 said:
It is definitely Intel's fault but they are correcting it with NVL which is too far away.

Yeah i can see that Ryzen is a no brainer CPU for u.

A NVL 52 Core CPU Shipping For R&D from the cache

View attachment 123714

I don't know if that sees the light of day. Between power and memory bandwidth it seems like they would have to make too many comprimises. I don't see how it can work on a mainstream socket.

511 · May 13, 2025

Thunder 57 said:
I don't know if that sees the light of day. Between power and memory bandwidth it seems like they would have to make too many comprimises. I don't see how it can work on a mainstream socket.

I can see few areas to improve power delivery first remove DLVR on Desktop it wastes Power at high Voltages raise the PL2 to 300W make P core efficient as for bandwidth more memory speed like DDR58000 or maybe 10-12K with XMP.

eek2121 · May 13, 2025

Thunder 57 said:
To be fair whose fault is that? For awhile Intel had AVX512 and AMD didn't. Should we just pick and choose based on who is supporting what and when?

Yep, you can’t say a benchmark is invalid because one of the vendors chose not to implement an instruction set. Otherwise, Apple Geekbench 6 results wouldn’t be valid, and AVX-512 is used in far more things.

LightningZ71 said:
I don't get the stanning for 12 full fat cores for gaming. I understand wanting high ST performance and gobs of cache, but, I'm yet to see a game that can keep 8 P cores at 100% that is also held back in any way by the more modest performance of the E-cores. The VAST majority of games have one, maybe up to three threads that are performance determinant, in other words, those threads are keeping their processor cores pegged at 100% and game performance can not improve any more unless those threads get more horsepower. When you look at Intel's 6P+8e products and compare them to the 8p+12e processors and then clock normalize the two processors as the 8p will boost higher generally, the remaining performance difference, if any, can be largely attributed to the slight difference in L3 cache between them and the remainder left to the extra threads available in total.

This isn't the same for AMD as they, to date, don't really have e cores in mainstream products, save for Strix/Kraken/Phoenix2, and those are largely just down clocked normal cores from a logic point of view. Their extra threads are coming from SMT, which lives in contention with other threads and take a bit more of a hit at lower core counts because of it.

I personally want the processor manufacturers to produce more products that look like Strix Point to some degree. I want 4 cores that are just absolutely sold out for ST performance. Large, low latency caches and high clock speeds. Then, I want a sea of slower, fully capable cores that can run secondary and background threads all day long. The biggest thing is that I want them to share the same L3 so that moving threads and data around has minimal latency. Having 4 very high performance threads should be more than enough to keep everything going smoothly in games.

I have seen games use 16-32 cores. Shoot, even 2D games like Factorio. Does every game do that? No. However, if you build it, they will come.

More importantly, PCs aren’t just about gaming. Productivity tasks are a thing.

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

Senior member

Attachments

Platinum Member

Senior member

Platinum Member

Diamond Member

Platinum Member

Diamond Member

Platinum Member

Senior member

Diamond Member

Lifer

Platinum Member

Senior member

Diamond Member

Platinum Member

Diamond Member

Senior member

Senior member

Platinum Member

Platinum Member

Senior member

Diamond Member

Platinum Member

Diamond Member

Platinum Member

Diamond Member