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.)

Thibsie · May 2, 2025

dullard said:
@511 I guess we have vastly different views of professional and prosumer users. I don't know too many casual users that are doing much 3D rendering, game development, and frequent compression/decompression of 100s of GB of data. To me those exactly summarize professional and prosumer users. I believe that LightningZ71 was referring to office workers on Word/Excel, our parents on Facebook doom scrolling, kids doing homework, people watching a movie, etc.

Does kids doing their homework involving VMs qualify prosumers ? Simple secondary school teaching huh. But load the VMs on the wrong cores and you can get 'somewhat' different behaviour. Annoying when introducing VMs...

eek2121 · May 2, 2025

dr1337 said:
Nah, most devs don't have access to build servers and even then a lot of companies straight up don't have resources like that because they're not adobe or oracle. And its much faster for dev work anyways if all your guys can compile independently on their own machines instead of all having to fill the server queue.

???

I’ve never worked for a company that didn’t have access to build servers?

I’ve worked for companies ranging from 4-6,500 employees (for software dev, at least.) Most of my work has been with smaller companies.

Markfw · May 2, 2025

eek2121 said:
???

I’ve never worked for a company that didn’t have access to build servers?

I’ve worked for companies ranging from 4-6,500 employees (for software dev, at least.) Most of my work has been with smaller companies.

delete (sorry mods)

Fjodor2001 · May 2, 2025

gdansk said:
And this time it's true. 4 cores was not ample.

Back then (some) people claimed games could not make use of more than 4C anyway, so going beyond that was pointless. But then 8/16C were introduced, and the new games were adapted to make use of the additional cores.

Same will happen now when going beyond 16C.

Also, not everything is about gaming anyway, as others have mentioned in the thread.

gdansk · May 2, 2025

Fjodor2001 said:
Back then (some) people claimed games could not make use of more than 4C anyway, so going beyond that was pointless. But the 8/16C were introduced, and the new games were adapted to make use of the additional cores.

Same will happen now when going beyond 16C.

Also, not everything is about gaming anyway, as others have mentioned in the thread.

Nope, Amdahl's law is an actual immutable law unlike Moore's observation.

Some day the diminishing returns may still be so cheap and that it is worth it. But the benefit of additional core spam is always diminishing.

MS_AT · May 2, 2025

dr1337 said:
A 16c CPU would be atleast at 50% and that is pretty utilized if you ask me.

I mean I was wondering if running the game on CPU woth fewer threads or more threads affects FPS (avg or 1%) in measurable way. Task manager is not well suited to judge this in my opinion.

dr1337 said:
Do you have any profile/data to backup your claim that you're being limited by ST perf with your compiler? Are you sure its not a you problem?

If you want proof look at benchmark data on phoronix or servethehome.com. Also it is not compiler that is the problem. But the dependency graph unique to each project and the build system that is governing the process, on high level.

Schmide said:
Compilers are fully threaded and can take advantage of multiple cores.

That is not accurate. Build system is launching one compiler process for each translation unit, and this is embarassingly parallel, but compilation of single TU is not yet multithreaded. If it was it would be also a problem for build systems which are not used to this kind of workflow.

And about build servers, they solve different kind of problem. For a day to day development you need fast cycles of build rebuild, builds are mostly incremental and since most of the time the only single TU is being compiled. So ST performance dominates.

For CI/CD workloads, you care mostly about throughput (getting as many builds per unit of time as you can, not caring so much how long a single one of them takes) so number of cores dominates.

Fjodor2001 · May 2, 2025

gdansk said:
Nope, Amdahl's law is an actual immutable law unlike Moore's observation.

Some day the diminishing returns may still be so cheap and that it is worth it. But the benefit of additional core spam is always diminishing.

Exactly the same as was said about the 4C limit.

gdansk · May 2, 2025

Fjodor2001 said:
Exactly the same as was said about the 4C limit.

And? The best gaming CPU is an 8C part with SMT disabled. Notably there is no difference in threads of execution from the 8T gaming champions of 15 years ago. Notably game consoles are still 8C.

The limits to gaming performance remains high speed communication and that is easier to obtain with a small number of cores. The core spam parts you want will be further disaggregated and bring regression in many workloads.You're asking that Intel and AMD cater to your workload rather than build a part that actually benefits most users.

You'll get more cores eventually, no need to rush it at great cost to everyone else. Especially when both Intel and AMD have workstation parts with that in mind.

LightningZ71 · May 2, 2025

The vast majority of games have 4 or less performance critical threads. Most have just one or two. Everything else is used for data movement, AI, compiling shaders, etc. The 6+8 Raptor processors do really well in games, as do the 8 core Ryzens with SMT.

OneEng2 · May 2, 2025

511 said:
I think we are giving AMD the ST Crown for no reason ARL Retains the ST Crown this gen. I fully expect Panther to be at least 20%+ IPC vs Lion Cove improvements and Arctic Wolf to be 25%+ vs Darkmont plus fixing the Latency regression for AMD i can say 15%+ easily for Zen 6 vs Zen 5.

Let me rephrase. ARL gets beaten in multi-core benchmarks by Zen 5 with a few exceptions. ARL gets beaten by Zen 5 in ST or lightly threaded workloads that are latency dependent. For ST that is not latency bound and especially bandwidth dependent workloads, ARL wins... but that seems to be a minority of applications.

So it seems like having enough cores that can work well at low latency wins most of the time.

inquiss said:
Where I disagree here is that gaming drives DIY. You see that in the top processor sales. I'm not saying that DIY is the only part of the market, but it's a much bigger part than 52c would be useful for (as a consumer platforms with consumer prices?)

If course the bulk of the market is just business and grey boxes and non vcache makes sense for volume standard use cases. Tbh that's mostly laptop anyway.

The bulk of the revenue is also in business. DIY is not much in the grand scheme of things.

jpiniero said:
DIY is only a part of the desktop market.

Most of the desktop market is "Corpos who could have bought a laptop but like the desktop/AIO form factor"

Agree.

Fjodor2001 said:
4C 4-ever, since nearly no one will ever need more than that. Used to be Intel™. Now the situation is reversed, so it's 16C 4-ever but AMD™, since 2019.

While software is forever moving forward and demanding more of everything, there are limits to the usefulness.

16 bit was pretty pathetic, but 32 bit was quite good. Yes, 64 bit is better, but for most things, 32 bit did just fine. 128bit? Still no need today. Will we ever need it? Maybe, but I am thinking not in my lifetime.

I think core count is the same. Going from 1 to 2 was HUGE. 2 to 4, was still a pretty big deal. 4 to 8? Not so much for not so many people. 8 to 16, only for a handful of niche cases..... and really, most of those cases are better served by a real workstation with more than 2 banks of memory to serve all the hungry cores with.

52 cores? I am just not seeing it. EITHER the cores are so wimpy that having them is not that big of an impact, OR you need a butt ton more than dual channel memory to feed them.... making it too expensive a platform for desktop and laptop consumers. I'll lump AMD's Zen 6 24c48t in the same boat. Seems like only a few people will benefit and it might be hard to keep the processor fed with only 2 channels.

gdansk said:
It's 32T thread since 2019 and no one yet has surpassed it. If Intel actually does build >32T consumer processor (on a dual channel platform) I have to wonder what they're smoking.

I am thinking the same thing.

Fjodor2001 said:
This time it looks like it’ll be Intel that leads the way forward instead of AMD. So funnily enough, the roles are the opposite this time.

It is definitely ironic considering the history. IIRC, when AMD started pushing the core count and Intel stayed focused on ST or lightly threaded app performance, Intel came out ahead in that situation.

This time around, while the landscape is different, I still think that very few real world apps will benefit much from 52c .... and it seems like the latency is a much bigger issue with ARL vs Zen 5, so I am guessing that this will be where the real battle exists .... not the max core count.

If it is max core count critical, then Thread Ripper will dominate. For people that really need 52c plus, a workstation is a much better value proposition.

dullard said:
@511 I guess we have vastly different views of professional and prosumer users. I don't know too many casual users that are doing much 3D rendering, game development, and frequent compression/decompression of 100s of GB of data. To me those exactly summarize professional and prosumer users. I believe that LightningZ71 was referring to office workers on Word/Excel, our parents on Facebook doom scrolling, kids doing homework, people watching a movie, etc.

Agree.

gdansk said:
Nope, Amdahl's law is an actual immutable law unlike Moore's observation.

Some day the diminishing returns may still be so cheap and that it is worth it. But the benefit of additional core spam is always diminishing.

Yes, but that is at the end of a long exponential curve .... long after I am dust IMO .

fastandfurious6 · May 2, 2025

52 cinebench cores lmao

maybe they find a way to make P cores better than ARL and still use the many extra cores for all that cinebench crunching

511 · May 3, 2025

gdansk said:
Nope, Amdahl's law is an actual immutable law unlike Moore's observation.

Some day the diminishing returns may still be so cheap and that it is worth it. But the benefit of additional core spam is always diminishing.

It is always dependent on workload and core spamming only helps with embarrassing parallel workloads.

OneEng2 said:
Let me rephrase. ARL gets beaten in multi-core benchmarks by Zen 5 with a few exceptions. ARL gets beaten by Zen 5 in ST or lightly threaded workloads that are latency dependent. For ST that is not latency bound and especially bandwidth dependent workloads, ARL wins... but that seems to be a minority of applications.

Latency is the Achilles heel of Arrow Lake thanks to the D2D/L3 Cycles.

OneEng2 said:
So it seems like having enough cores that can work well at low latency wins most of the time.

The bulk of the revenue is also in business. DIY is not much in the grand scheme of things.

Agree.

While software is forever moving forward and demanding more of everything, there are limits to the usefulness.

16 bit was pretty pathetic, but 32 bit was quite good. Yes, 64 bit is better, but for most things, 32 bit did just fine. 128bit? Still no need today. Will we ever need it? Maybe, but I am thinking not in my lifetime.

Yeah we barely are fully utilizing the 64bit limit

OneEng2 said:
I think core count is the same. Going from 1 to 2 was HUGE. 2 to 4, was still a pretty big deal. 4 to 8? Not so much for not so many people. 8 to 16, only for a handful of niche cases..... and really, most of those cases are better served by a real workstation with more than 2 banks of memory to serve all the hungry cores with.

52 cores? I am just not seeing it. EITHER the cores are so wimpy that having them is not that big of an impact, OR you need a butt ton more than dual channel memory to feed them.... making it too expensive a platform for desktop and laptop consumers. I'll lump AMD's Zen 6 24c48t in the same boat. Seems like only a few people will benefit and it might be hard to keep the processor fed with only 2 channels.

Just cause we have the SKUs doesn't mean they are the things that will sell the most they are part of the lineup someone needs it that is why they are increasing the core count also 4LPE Cores are for standby and stuff.

OneEng2 said:
I am thinking the same thing.

It is definitely ironic considering the history. IIRC, when AMD started pushing the core count and Intel stayed focused on ST or lightly threaded app performance, Intel came out ahead in that situation.

This time around, while the landscape is different, I still think that very few real world apps will benefit much from 52c .... and it seems like the latency is a much bigger issue with ARL vs Zen 5, so I am guessing that this will be where the real battle exists .... not the max core count.

If it is max core count critical, then Thread Ripper will dominate. For people that really need 52c plus, a workstation is a much better value proposition.

Workstation cost a lot more than a dual channel High Core count setup.

OneEng2 · May 3, 2025

511 said:
Workstation cost a lot more than a dual channel High Core count setup.

Agreed; however, if you think about the kinds of people that NEED 52 cores badly enough to pay the premium for it, those are MOSTLY people who make money using their computer.

In my experience, this type of computer consumer doesn't mind paying premium money for a real workstation as the ROI for them is an easy calculation.

For the rest of us that use our computers in a way that 24c/24t (8p+16e) is MORE than sufficient, the additional cost just seems silly to me. On the Zen 6 side there will be a 12c/24t part on a single CCD. Again, more than enough for most people.

I am not saying there aren't people who will pay for more cores, only that MOST of those "people" would be the kind of people that would also pay for a real workstation like Thread Ripper having the additional memory channels to feed many cores.

At least that is what I am guessing.

Could someone please make the business case for these high end "desktop" processors?

As per other conversations, there also seems to be some agreement that if the 52 cores have good compute capability, then they will need more bandwidth than can be provided.

If this isn't true, can someone explain why?

511 · May 3, 2025

OneEng2 said:
Agreed; however, if you think about the kinds of people that NEED 52 cores badly enough to pay the premium for it, those are MOSTLY people who make money using their computer.

In my experience, this type of computer consumer doesn't mind paying premium money for a real workstation as the ROI for them is an easy calculation.

For the rest of us that use our computers in a way that 24c/24t (8p+16e) is MORE than sufficient, the additional cost just seems silly to me. On the Zen 6 side there will be a 12c/24t part on a single CCD. Again, more than enough for most people.

I am not saying there aren't people who will pay for more cores, only that MOST of those "people" would be the kind of people that would also pay for a real workstation like Thread Ripper having the additional memory channels to feed many cores.

At least that is what I am guessing.

Could someone please make the business case for these high end "desktop" processors?

I know game farms love to use Desktop processors like 14900K/9950X. I heard about many farms that used 14900K and replaced them with 9950X during the raptor lake debacle.
That is the reason I came to know that these things exists.

OneEng2 said:
As per other conversations, there also seems to be some agreement that if the 52 cores have good compute capability, then they will need more bandwidth than can be provided.

If this isn't true, can someone explain why?

I think someone made a post explaining the bandwidth part also it's not like they are not increasing memory speed to compensate for bandwidth will it be enough that Is something only time can tell.

eek2121 · May 4, 2025

OneEng2 said:
Agreed; however, if you think about the kinds of people that NEED 52 cores badly enough to pay the premium for it, those are MOSTLY people who make money using their computer.

In my experience, this type of computer consumer doesn't mind paying premium money for a real workstation as the ROI for them is an easy calculation.

For the rest of us that use our computers in a way that 24c/24t (8p+16e) is MORE than sufficient, the additional cost just seems silly to me. On the Zen 6 side there will be a 12c/24t part on a single CCD. Again, more than enough for most people.

I am not saying there aren't people who will pay for more cores, only that MOST of those "people" would be the kind of people that would also pay for a real workstation like Thread Ripper having the additional memory channels to feed many cores.

At least that is what I am guessing.

Could someone please make the business case for these high end "desktop" processors?

As per other conversations, there also seems to be some agreement that if the 52 cores have good compute capability, then they will need more bandwidth than can be provided.

If this isn't true, can someone explain why?

A lot of folks who make money using their computer are also gamers. Current solutions sacrifice gaming/single-threaded performance for multicore performance. This is something AMD/Intel could address, but they choose not to.

That being said, if AMD is indeed upping the core count, I suspect that will satisfy many of us. I have no idea about the rumored Intel part because they have had serious execution problems, but if it can also game like a champ and do great at multicore without needing a nuclear reactor for power, they will have a winner.

511 · May 4, 2025

eek2121 said:
A lot of folks who make money using their computer are also gamers. Current solutions sacrifice gaming/single-threaded performance for multicore performance. This is something AMD/Intel could address, but they choose not to.

That being said, if AMD is indeed upping the core count, I suspect that will satisfy many of us. I have no idea about the rumored Intel part because they have had serious execution problems, but if it can also game like a champ and do great at multicore without needing a nuclear reactor for power, they will have a winner.

On the gaming part I doubt any of the vanilla part will beat X3D it will still be slightly behind

OneEng2 · May 4, 2025

I doubt gaming is important to the overall success of a design. AMD has a good product for gamers, but gamers aren't going to need huge core counts IMO.

I am struggling to find a high volume of the market that benefits from high core count that isn't better served by a workstation.

511 · May 4, 2025

OneEng2 said:
I doubt gaming is important to the overall success of a design. AMD has a good product for gamers, but gamers aren't going to need huge core counts IMO.

X3D is a good marketing to buy AMD in DIY Space even though 13/14th gen i5 offer better value nowadays.

OneEng2 said:
I am struggling to find a high volume of the market that benefits from high core count that isn't better served by a workstation.

Should ask Intel/AMDs Product Manager 🤣

reb0rn · May 4, 2025

More cores are always needed.... will they be efficient we gona see, those that think same as 4 core dozen a year ago now think same, just let them live in past

MS_AT · May 4, 2025

OneEng2 said:
I am struggling to find a high volume of the market that benefits from high core count that isn't better served by a workstation.

Not sure if the volume is high enough, but the proposed NovaLake part (2x8P, 2x16E) would be fine for code compilation provided M$ and Intel will finally solve their scheduling issues. After all you don't need memBW for these, nor a lot of PCIe lanes, what makes Threadripper / Epyc / Xeon usually an overkill for such builds, and high clocks are beneficial for the ST bottlenecks you will run into when you scale the build to high enough core counts. At least from my experience with non-trivial C++ projects.

OneEng2 · May 4, 2025

reb0rn said:
More cores are always needed.... will they be efficient we gona see, those that think same as 4 core dozen a year ago now think same, just let them live in past

Didn't work for AMD. ST performance was more important to more people.

I realize that things are much different now than then; however, the question I pose is what percentage of the market needs and will pay for more cores. And of those who do need and will pay for more cores, how many of them would just opt for a workstation?

I am an engineer and a power user. I get along fine with my 6 core/ 12 tread setup at home. Even when I do video rendering (which is only a few times a year), this machine seems just fine (If I did this often, I would definitely be wanting many more cores!).

MS_AT said:
Not sure if the volume is high enough, but the proposed NovaLake part (2x8P, 2x16E) would be fine for code compilation provided M$ and Intel will finally solve their scheduling issues. After all you don't need memBW for these, nor a lot of PCIe lanes, what makes Threadripper / Epyc / Xeon usually an overkill for such builds, and high clocks are beneficial for the ST bottlenecks you will run into when you scale the build to high enough core counts. At least from my experience with non-trivial C++ projects.

I have been thinking that code compilation might be one of those tasks as you have pointed out. I have some non-trivial C++ projects as well, but honestly, even those compile pretty quickly .... even when I force a "Build All". Of course, mostly it is just a single unit in a single object that gets compiled and linked into the project, so despite the size of the project, the compile is pretty quick.

Oddly enough, Flutter at work seems to be more demanding on Android Studio than C++ on the desktop. Seems like somebody got a bit fat over there on Android .

reb0rn · May 4, 2025

So now its 6 core is enough, are you ppl for real

Schmide · May 4, 2025

The reason developers should get the great machines is they run in debug mode. Optimization takes a back seat. If it runs ok on an above average machine, it should run fine on a lesser machine in release build. Running a profiler is not super taxing but if you're doing it over and over those minutes or even few seconds add up.

I probably wouldn't notice a lesser machine for every day compilation. Even incremental builds on large projects wouldn't be horrible. I'd rather spend the extra few bucks for a good processor so when I evaluate a large project or do updates, it takes less time.

The simple truth is. You can dumb down a faster machine (limit threads and such) but you can't smart up a slower one. A developer machine should at least meet the best metrics a program is being designed for and probably exceed it for good measure.

coercitiv · May 5, 2025

eek2121 said:
A lot of folks who make money using their computer are also gamers. Current solutions sacrifice gaming/single-threaded performance for multicore performance. This is something AMD/Intel could address, but they choose not to.

The same will apply to this 52c NVL-S, when compared to an optimized design it will sacrifice gaming perf for productivity perf. Mem controller stays on SoC tile for a start. The tiles are identical 8+16, in order for a dual tile to perform well in gaming it would need to be exclusive P tile and exclusive E tile. This would also increase MT perf since resulting core count would be something like 12P+40E due to distribution on somewhat identically sized tiles. The obvious problem with asymmetrical tiles would be design cost (financial, manpower, time to market). Ironically AMD is in a better position to execute such a setup with a 12+24 chip but I really doubt they'll do it until Intel has something on the shelves that challenges their 3D cache setup.

I agree with @OneEng2 , in the sense that a 52c consumer chip would be in an awkward place in terms of sales potential. I think we can all agree there are people out there that need high core count CPUs at lower prices, but it needs to be cristal clear for everyone here that both Intel and AMD could have addressed this segment with their HEDT lineup. They did not. They want professionals like @Schmide to open up their wallets even for homelab machines.

My take is still the same, I expect such a chip, should it come to market unlike the fabled high core-count Arrow Lake that never saw the light of day, will offer only limited value over the HEDT lineup. I really doubt Intel is willing to start a price war in the professional market out of all places, especially with premium bins from TSMC dies. I'll be there in case they start one though, can't say no to freEEE cores!

Edit: corrected core count for imaginary E core tile

511 · May 5, 2025

On this ARL-S 8+16 N3B tile is 117mm2 which is a lot bigger imo for the performance it offers they need to optimize their P cores first in terms of PPA otherwise it's going to be damm expensive to produce such large dies let alone 2 of them.

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

Senior member

Attachments

Golden Member

Diamond Member

Moderator Emeritus, Elite Member

Diamond Member

Diamond Member

Senior member

Diamond Member

Diamond Member

Platinum Member

Senior member

Senior member

Golden Member

Senior member

Golden Member

Diamond Member

Golden Member

Senior member

Golden Member

Senior member

Senior member

Senior member

Senior member

Diamond Member

Diamond Member

Golden Member