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.)

Wolverine2349 · Oct 11, 2024

Hulk said:
Intel is saying Skymont is +32% IPC over Gracemont so that is where I based my figure. 32% over Gracemont does not equal Raptor Cove. That would be more like 48%.

I think we will see that what Intel means by Skymont~RPC IPC is in some specific best case scenarios (heavy FP) much in the same way Gracemont~Skylake in some use cases.

I have found that taking the more conservative Intel performance estimates will correlate better to actual performance. The more hyperbolic sounding Intel claims, like Skymont = Raptor Cove are generally hard to replicate and rare corner cases.

We'll know the truth soon enough. One thing I can't figure out is how Intel is claiming +18% CB R24 MT for ARL over RPL? That is nuts.

I'm curious to see actual CB R24 MT results and compare them to RPL, stock-for-stock clocks to verify this seemingly outrageous Intel claim.

Well if the Skymont is so much more powerful than Gracemont and Lion Cove has 10-15% better IPC than its possible.

Lack of HT only hurts so much. Logical threads are nothing compared to a physical real core thread. There is a reason its called a logical thread and raw compute power is not really improved with SMT/HT rather more efficient multi tasking ability.

Hulk · Oct 11, 2024

Wolverine2349 said:
Well if the Skymont is so much more powerful than Gracemont and Lion Cove has 10-15% better IPC than its possible.

Lack of HT only hurts so much. Logical threads are nothing compared to a physical real core thread. There is a reason its called a logical thread and draw compute power is not relaly improved with SMTY/HT rather more efficient multi tasking ability.

Sorry, I didn't communicate this correctly. Even if Skymont performs like Raptor Cove in CB R24 MT, Lion Cove would have to be +30% IPC over Raptor Cove to make the total score +18% over Raptor Lake. This assume stock clocks for 14900K and 285K.

It's a bit of a mystery to me right now. Once we get some "real" scores, and by real I mean from people in this forum, we won't know what is going on exactly.

DrMrLordX · Oct 11, 2024

gdansk said:
I don't think it needs to be based strictly upon Skymont as long as it incorporates lessons learned.

The monts are getting pretty fat, and there's a good likelihood that the e-core team already foresaw this possibility and incorporated "lessons learned" into their future development efforts. If they're anything like AMD, they probably have the next 2-3 gens in development to one extent or another. The Haifa team needs to suck up their guts (and their egos) and support other teams or face budgetary cuts.

OneEng2 · Oct 11, 2024

Everyone here seems very concerned about gaming ..... which I guess is cool if that is your thing. For the health of Intel as a company, I am more concerned about how they are doing in the data center where they have been bleeding market share in an open wound for the last several years.

Well, EPYC Turin reviews are out. Looks like Granite Rapids (128 core) is taking a huge beating across the board (with a very few single benchmark wins).

AMD EPYC ‘Turin’ 9005 Series - we benchmark 192-core Zen 5 chip with 500W TDP

Cranking up the power.

www.tomshardware.com

Another thing this article brings up is the value proposition that AMD data center CPU's offer. Much of the server world software licensing is done "per core". With AMD's very potent Zen 5 and Zen 5c core offerings, "Performance per core" is going to be a very big deal.

It is pretty clear that AMD will have another year of unrestrained server market share growth and profit while Intel's Granite Rapids will continue to struggle.

Based on these benchmarks, even the 244 core Xeon coming early next year is going to be in deep trouble.

... and still it looks like Diamond Rapids will be late in 2025 or even 2026 from the leaks I have been reading. And Intel is moving sockets again while AMD stays put with Turin which remains socket compatible with the last 3 versions of EPYC. Note: The 500W version requires more power than was available on some (maybe all) server main boards, but development of such a MB from previous designs should be fairly straight forward vs a completely new socket.

And on the point of Arrow Lake, if this is all the better an Intel slide presentation can slant the data to make it look, I can only imagine what real world benchmarks are going to look like. For those of you who are gamers, It seems like AMD will likely slip in samples of 9800X3D for reviewers to use on game comparisons which will very likely show a considerable gap between Arrow Lake and that chip for gamers.

So from a core architecture point of view (which I see as key to a companies long term viability), Intel is currently holding a strong card with Lunar Lake in thin and light laptops, but they are doing so on an expensive process.

In the desktop, Arrow Lake doesn't seem all that impressive either..... not that desktop market is worth shaking a stick at these days anyway.

It's Arrow Lake for the full size laptop (desktop replacement) that still intrigues me. Intel could score a win here that AMD has no answer for until Zen 6 (when they move to a 16 core CCD).

In the data center, AMD's Turin variants appear to be way more than a match for Granite Rapids in any variation that Intel can release.

Det0x · Oct 11, 2024

OneEng2 said:
Based on these benchmarks, even the 244 core Xeon coming early next year is going to be in deep trouble.

And AMD haven't even shown their highest offering yet..

Level1Techs hinted strongly for a 256core Z5c offering in 2025 🫣

Abwx · Oct 11, 2024

.........................

OneEng2 said:
It's Arrow Lake for the full size laptop (desktop replacement) that still intrigues me. Intel could score a win here that AMD has no answer for until Zen 6 (when they move to a 16 core CCD).

The answer is in january or so with Strix Halo s 16C/32T.

Det0x · Oct 11, 2024

Some Arrow Lake and Zen5X3D leaks from MSI

Full slidedeck here:

https://imgur.com/a/zc5JpY5

Det0x · Oct 11, 2024

More leaks

GTracing · Oct 11, 2024

OneEng2 said:
Everyone here seems very concerned about gaming ..... which I guess is cool if that is your thing. For the health of Intel as a company, I am more concerned about how they are doing in the data center where they have been bleeding market share in an open wound for the last several years.

Well, EPYC Turin reviews are out. Looks like Granite Rapids (128 core) is taking a huge beating across the board (with a very few single benchmark wins).

AMD EPYC ‘Turin’ 9005 Series - we benchmark 192-core Zen 5 chip with 500W TDP

Cranking up the power.

www.tomshardware.com

Another thing this article brings up is the value proposition that AMD data center CPU's offer. Much of the server world software licensing is done "per core". With AMD's very potent Zen 5 and Zen 5c core offerings, "Performance per core" is going to be a very big deal.

It is pretty clear that AMD will have another year of unrestrained server market share growth and profit while Intel's Granite Rapids will continue to struggle.

Based on these benchmarks, even the 244 core Xeon coming early next year is going to be in deep trouble.

... and still it looks like Diamond Rapids will be late in 2025 or even 2026 from the leaks I have been reading. And Intel is moving sockets again while AMD stays put with Turin which remains socket compatible with the last 3 versions of EPYC. Note: The 500W version requires more power than was available on some (maybe all) server main boards, but development of such a MB from previous designs should be fairly straight forward vs a completely new socket.

And on the point of Arrow Lake, if this is all the better an Intel slide presentation can slant the data to make it look, I can only imagine what real world benchmarks are going to look like. For those of you who are gamers, It seems like AMD will likely slip in samples of 9800X3D for reviewers to use on game comparisons which will very likely show a considerable gap between Arrow Lake and that chip for gamers.

So from a core architecture point of view (which I see as key to a companies long term viability), Intel is currently holding a strong card with Lunar Lake in thin and light laptops, but they are doing so on an expensive process.

In the desktop, Arrow Lake doesn't seem all that impressive either..... not that desktop market is worth shaking a stick at these days anyway.

It's Arrow Lake for the full size laptop (desktop replacement) that still intrigues me. Intel could score a win here that AMD has no answer for until Zen 6 (when they move to a 16 core CCD).

In the data center, AMD's Turin variants appear to be way more than a match for Granite Rapids in any variation that Intel can release.

There's so much in this comment that it's hard to address it all, but I will say a few things. People are discussing desktop/gaming because Intel announced desktop CPUs literally yesterday; desktop is a larger portion of Intel's overall revenue than you might think, 21% last quarter; and Intel's current financial difficulties are almost entirely due to their manufacturing division. Their server offerings are definitely uncompetitive though.

DavidC1 · Oct 11, 2024

Hulk said:
Intel is saying Skymont is +32% IPC over Gracemont so that is where I based my figure. 32% over Gracemont does not equal Raptor Cove. That would be more like 48%.

Remember 9% for Lion Cove is combined Int/FP, while Skymont is 32%/72% in Int and FP respectively.

Assuming 65/35% split like in Geekbench, that'll result in an average of 46% gains.

Now tell me in what world would a 32% Int, 72% FP gain even against the 12th Gen Gracemont would result in Lion Cove being 20-30% faster per clock over Skymont?

If you take Raptormont figure, that's 7.2% and 3.5% lead in Int/FP. If it's over Gracemont it's 1% and 3% behind. So with a meagre 9% gain, Lion Cove is as low as 2.9% faster over Skymont.

I think we will see that what Intel means by Skymont~RPC IPC is in some specific best case scenarios (heavy FP) much in the same way Gracemont~Skylake in some use cases.

Lunarlake presentations said Skymont on ring will outperform Raptor Cove by 2% in both SpecInt and FP.

Skymont straight up doubles the number of FP units, so unlike the new instructions which take eons for software to adopt, it'll benefit all the applications dating back to the Pentium 4 era immediately.

In many applications, Skymont has straight up better FP throughput as it has twice the amount of FP units compared to Raptor Cove.

I certainly prefer this approach over the P core teams' almost Elitist approach of shutting down competition by forcing them to adopt new ISA, and for the entire world to cater and recompile for them.

Det0x · Oct 11, 2024

Regarding the pictures ive posted above

9950X stock PPT limit is 200w
285K is running unlimited at 303w here (?)

Kinda seems like MSI really wanted the 285k to win R23 MT 🤣

Dont know if its relevent for this convo, but i can hit 49k in R23 on 9950X with manual static OC @ 250w using watercooling
Stock boosting gets ~42k for the 9950X @ 200w (in my opinion stock V/F curve on vanilla Z5 is pretty bad as it overvolt alot ->atleast gets alittle better on X3D)

Magio · Oct 11, 2024

Wolverine2349 said:
Well if its true that Skymont has almost the came IPC as Lion Cove, and Arctic Wolf e-cores take another big jump, why wouldn't e-cores already be ready to replace P cores unless P cores can take another huge jump though that seems to be struggling to happen in the Israel Design Center P core team.

Or is there more to the story than just IPC/ Is it that Skymont has IPC in some areas close to lion Cove but not ready tot ake over by itself even with a big jump? Is there some specific limitation the Austin Atom team cores have regardless of jumps right now that prevent them from being primary core? Are they like dependent on P cores for all around functionality?

Like much more to the situation and story than this oh e-cores ready to take over P cores with Arctic Wolf in a few years or longer?

Right now, Skymont just doesn't scale beyond a certain power/clock speed. Its sweetspot is at low power so you start redlining it at relatively low overall performance and it just won't scale beyond that.

The IPC is already nice but the e-cores will have to widen their range of operation quite a lot before they're a credible replacement for P cores. That's also why I think a unified core won't be born out of nowhere all of a sudden, it's more realistic to slowly build up the e-cores over several gens and to make the switch once it's viable.

Det0x · Oct 11, 2024

LOL so much for power efficency

Guess what kind of "design motherboards" will be used for reviews ?
Hint: Its the ASUS APEX/HERO -> recommended settings is "P1=250w P2=295w" ?
At bare minimum its P1=250w P2=250w

DavidC1 · Oct 11, 2024

DrMrLordX said:
The monts are getting pretty fat,

It's bigger but it's still only 1.15mm2, and full 3x the core size difference from their P.

If they didn't double FP units, it would have been under 1mm2, as FP blocks typically take 20-25% of the core area. Since Gracemont adopted FMA, it would have been enlarged already and it's straight up doubling of that.

This is the original Atom "Bonnell" core. The FPC is the Floating Point Cluster.

The FPC takes roughly 18% of the core. In Goldmont they went 128-bit and fully pipelined it, while going out of order FP. In Gracemont they added FMA.

Here was my analysis a while ago about processes:

Crestmont N6 - 1.46mm2
Crestmont Intel 4 - 1.01mm2

N6 - 16% over N7
N5 - 1.84x over N7, but according to Angstronomics, it's actually 1.5x
N3 - 1.7x over N5

Hypothetical Crestmont N7 - 1.69mm2
Hypothetical Crestmont N5 - 0.92mm2, 1.12mm2(Angstronomics)
Hypothetical Crestmont N3 - 0.54mm2, 0.66mm2(over Angstronomics)

Intel 3 - 1.1x density over 4: 0.92mm2
Intel 18A - 1.3x density over 3: 0.7mm2

According to the above info, Skymont being 1.15mm2 on N3B means it got 74% bigger. If we assume FP block took up 20%, then without the FP block, the core size growth would have been 45%, quite reasonable as it has 32% Int gain and clocks increased slightly. If you assume FP blocks are 25%, then the core size increase is 39%.

MS_AT · Oct 11, 2024

DavidC1 said:
Lunarlake presentations said Skymont on ring will outperform Raptor Cove by 2% in both SpecInt and FP.

Skymont straight up doubles the number of FP units, so unlike the new instructions which take eons for software to adopt, it'll benefit all the applications dating back to the Pentium 4 era immediately.

In many applications, Skymont has straight up better FP throughput as it has twice the amount of FP units compared to Raptor Cove.

I certainly prefer this approach over the P core teams' almost Elitist approach of shutting down competition by forcing them to adopt new ISA, and for the entire world to cater and recompile for them.

Weren't the slides talking about IPC difference, not actual performance difference? Now, I wouldn't call AVX2 to be elitist, it is on the market for 12 years already. Oh and Raptor Cove has 3 SIMD execution ports. So it's 3 vs 4, not 2 vs 4. With P core having 256b wide regs

DrMrLordX · Oct 11, 2024

OneEng2 said:
It is pretty clear that AMD will have another year of unrestrained server market share growth and profit while Intel's Granite Rapids will continue to struggle.

Granite Rapids' best trait is that it isn't as bad as Sapphire Rapids relative to the rest of the market. It actually released on time (I guess) and it doesn't seem plagued by as many problems. That being said, it barely competes with previous gen competition, and it has to face the new generation which is quite a beast. So yes, they will likely continue to bleed marketshare.

@DavidC1

True, but my main point is that monts keep on taking on traits of their larger siblings, to the point where they won't (necessarily) be so small or targeted towards power/area efficiency anymore. I'm pretty sure the monts can replace the P-cores without too much outside interference. Could be wrong, and with the power/clockspeed scaling of Skymont maybe we're looking at one of the last major limitations of the e-cores that the design team will need to overcome before they can go primetime.

DavidC1 · Oct 11, 2024

Yes you are right, they were talking about perf/clock.

I am pointing out that @Hulk is saying that Skymont is 20-30% behind Lion Cove because Skymont gains 32% over Gracemont while the gap between Grace and Golden is ~45%. Which I refuted by saying Skymont doubles FP units thus gets 72% in FP, which with 65/35% split as with GB will result in the ~45% required to be on par with Raptorlake.

MS_AT said:
Weren't the slides talking about IPC difference, not actual performance difference? Now, I wouldn't call AVX2 to be elitist, it is on the market for 12 years already. Oh and Raptor Cove has 3 SIMD execution ports. So it's 3 vs 4, not 2 vs 4. With P core having 256b wide regs

Golden Cove architecture only has two units capable of FMA, and full throughput. The third pipe only deals certain much less used instructions and don't count as part of FP throughput: https://chipsandcheese.com/p/lion-cove-intels-p-core-roars

That has not changed from Core 2, where it had two FP pipes capable of doing the important things and you have others that impact it a lot less such as FP Shuffle.

Intel's Merom Unveiled

Introduction Intel’s Developer Forum (IDF) for Spring 2006 has a rather obvious focus: the announcement of Intel’s next generation microarchitecture.

www.realworldtech.com

cannedlake240 · Oct 11, 2024

DavidC1 said:
Here was my analysis a while ago about processes:

It should be noted that TSMC itself added a new metric to their N3E slide above called “Chip Density” which the company derives using “50% [Logic Density] + 30% [SRAM Desnity] + 20% [Analog Density].” TSMC says the “chip density” is ~1.3x versus the 1.6x logic density for N3E vs N5.

N3E Replaces N3; Comes In Many Flavors

TSMC outlines a number of 3-nanometer class node changes and flavors.

fuse.wikichip.org

MS_AT · Oct 11, 2024

DavidC1 said:
Yes you are right, they were talking about perf/clock.

I am pointing out that @Hulk is saying that Skymont is 20-30% behind Lion Cove because Skymont gains 32% over Gracemont while the gap between Grace and Golden is ~45%. Which I refuted by saying Skymont doubles FP units thus gets 72% in FP, which with 65/35% split as with GB will result in the ~45% required to be on par with Raptorlake.

Golden Cove architecture only has two units capable of FMA, and full throughput: https://chipsandcheese.com/p/lion-cove-intels-p-core-roars

That has not changed from Core 2, where it had two FP pipes capable of doing the important things and you have others that impact it a lot less such as FP Shuffle.

Intel's Merom Unveiled

Introduction Intel’s Developer Forum (IDF) for Spring 2006 has a rather obvious focus: the announcement of Intel’s next generation microarchitecture.

www.realworldtech.com

Well that depends on what you count as important. They can do 3 SIMD adds per cycle and 2 muls/fmuls according to throughput tables for Golden Cove, I guess Raptor Cove did not regress on that. And the fact Coves were bottlenecked on shuffles on port 5 was a problem in the past, so if Skymont can do shuffles on more units, then it will have an advantage. It will be definitely interesting to see the comparison when reviewers will publish ArrowLake reviews.

DavidC1 · Oct 11, 2024

MS_AT said:
Well that depends on what you count as important. They can do 3 SIMD adds per cycle and 2 muls/fmuls according to throughput tables for Golden Cove, I guess Raptor Cove did not regress on that. And the fact Coves were bottlenecked on shuffles on port 5 was a problem in the past, so if Skymont can do shuffles on more units, then it will have an advantage. It will be definitely interesting to see the comparison when reviewers will publish ArrowLake reviews.

I think the fact that the combined performance is a meagre 9% is a testament to whatever they did not working, because if they boosted FP in a general way like on Skymont, the FP portion would have done lot better.

Not getting good gains on perf/clock is what I count as important. The P core team for forever talked a lot about what's a bottleneck or whatever but aside from Pentium M, Core 2, and Sandy Bridge, it was usually disappointing. 10% for Haswell, 10% for Skylake in an era where it was far easier to get big gains on process. What the hell were they doing?

Skymont is capable of legacy and FMA execution on all 4 ports. Even Lion Cove is only FMA on 2 pipes and other 2 are FP Add, so the gains won't be universal.

H433x0n · Oct 11, 2024

Det0x said:
Regarding the pictures ive posted above

9950X stock PPT limit is 200w
285K is running unlimited at 303w here (?)

Kinda seems like MSI really wanted the 285k to win R23 MT 🤣

View attachment 109274

View attachment 109275
Dont know if its relevent for this convo, but i can hit 49k in R23 on 9950X with manual static OC @ 250w using watercooling
Stock boosting gets ~42k for the 9950X @ 200w (in my opinion stock V/F curve on vanilla Z5 is pretty bad as it overvolt alot ->atleast gets alittle better on X3D)

They’re both running at unlimited power limits, says so right in the bottom left corner.

Det0x · Oct 11, 2024

H433x0n said:
They’re both running at unlimited power limits, says so right in the bottom left corner.

Yeah i know about the note in bottom left, thats why i have marked it on the screen 🤷‍♂️
And thats really not how i read it

"Intel core ultra proccesor cpu performance with powerlimit unlocked outperform ryzen 9000"
We know this also from comparing the 14900k vs 9950X score..

Its 9950X @ 200w PPT vs 285K @ 303w PPT they are comparing here

Abwx · Oct 11, 2024

H433x0n said:
They’re both running at unlimited power limits, says so right in the bottom left corner.

With PBO the 9950X for sure outmatch the 14900K/KS by 10% and by 7-8% at stock, wich is not the case at all in those pics, so not only it s not ocked but likely that all intel chips are.

Cinebench R20, R23 und 2024 - Seite 3 - Hardwareluxx

Hardwareluxx hat den neuen AMD Ryzen 9 9950X ausführlich getestet.

www.hardwareluxx.de

DavidC1 · Oct 11, 2024

Det0x said:
We know this also from comparing the 14900k vs 9950X score..

Its 9950X @ 200w PPT vs 285K @ 303w PPT they are comparing here

That sounds right. 285K probably has a greater gain over 14900K on the default power limits compared to unlimited, because Intel processes are optimized for higher frequencies.

Motherboard vendors are there to sell their own boards, so they won't really care other than the newest product outperforming others.

H433x0n · Oct 11, 2024

Abwx said:
With PBO the 9950X for sure outmatch the 14900K/KS by 10% and by 7-8% at stock, wich is not the case at all in those pics, so not only it s not ocked but likely that all intel chips are.

Cinebench R20, R23 und 2024 - Seite 3 - Hardwareluxx

Hardwareluxx hat den neuen AMD Ryzen 9 9950X ausführlich getestet.

www.hardwareluxx.de

The 14900K/S will score 41K with uncapped power limits.

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

Senior member

Attachments

Senior member

Diamond Member

Lifer

Senior member

Golden Member

Lifer

Golden Member

Golden Member

Member

Golden Member

Golden Member

Member

Golden Member

Golden Member

Senior member

Lifer

Golden Member

Senior member

Senior member

Golden Member

Golden Member

Golden Member

Lifer

Attachments

Golden Member

Golden Member