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

[Tigerick]

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	8 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 (20A)	Arrow Lake (N3B)	Lunar Lake	Panther Lake
Platform	Mobile H/U Only	Desktop Only	Desktop & Mobile H&HX	Mobile U Only	Mobile H
Process Node	Intel 4	Intel 20A	TSMC N3B	TSMC N3B	Intel 18A
Date	Q4 2023	Q1 2025 ?	Desktop-Q4-2024 H&HX-Q1-2025	Q4 2024	Q1 2026 ?
Full Die	6P + 8P	6P + 8E ?	8P + 16E	4P + 4E	4P + 8E
LLC	24 MB	24 MB ?	36 MB ?	8 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.)

 

[Det0x]

W1zzard from TPU said Arrowlake will be faster than both zen 5 and Zen 4x3d:

AMD Zen 5 Technical Deep Dive

AMD has finally shared technical details for their next-gen Zen 5 microarchitecture, which will power the new Ryzen 9000 desktop processors and the Ryzen AI 300 Series laptop CPUs. We detail all the technical innovations, including overclocking, and how AMD achieved +16% IPC.

www.techpowerup.com

Where does he say that ?
I can only find this:

AMD Ryzen 9000 series CPU core counts range between 6-core and 16-core, something that hasn't changed for AMD since the Ryzen 3000 Zen 2. AMD needs to work on this, because Intel Arrow Lake is about to hit AMD with not just those high-IPC Lion Cove cores, but also the Skymont E-cores where Intel has managed to pull off a miracle, with a 50% IPC increase over the previous E-core. Whatever core-count Intel may come up with for Arrow Lake may prove competitive against Granite Ridge. Sooner or later, AMD needs to reconsider core counts.

 

[Goop_reformed]

Where does he say that ?
I can only find this:

"We're really happy with what we're seeing with the Ryzen AI 300 series mobile processors. The Ryzen 9000 desktop processors are exciting, too, but we really wish it came with an NPU and beat the 7800X3D, because Arrow Lake-S will probably do both, at unknown power levels though."

This has been edited. The original was without "at unknown power levels though".

 

[ryanjagtap]

ARL-S tops up at 8P+16E/24T, right? Does that mean the Ryzen 9 9950X with 16C/32T will dominate the nT benchmarks, or will the uplift from Lion Cove and Skymont make up for the loss of thread counts (HT threads)?

 

[SiliconFly]

ARL-S tops up at 8P+16E/24T, right? Does that mean the Ryzen 9 9950X with 16C/32T will dominate the nT benchmarks, or will the uplift from Lion Cove and Skymont make up for the loss of thread counts (HT threads)?

HT "threads" are just emulation. The loss of thread count won't bring MT perf down by much cos each physical core tends to gain ST perf when HT is removed which compensates well.

 

[vanplayer]

ARL-S tops up at 8P+16E/24T, right? Does that mean the Ryzen 9 9950X with 16C/32T will dominate the nT benchmarks, or will the uplift from Lion Cove and Skymont make up for the loss of thread counts (HT threads)?

For Skymont maybe, but for LionCove I'm afraid is just beyond its limit. ES2 ARL LionCove is a thing like Geekbench6 ST 3100ish thing, internal data. You might be able to see 3300ish when it comes out if lucky. Without HT ARL might lose 30% performance from Pcore side and LionCove cannot boost that higher to mitigate the HT loss and IPC is not optimistic at the meanwhile. All the bet just put on E core currently.

 

[SiliconFly]

For Skymont maybe, but for LionCove I'm afraid is just beyond its limit. ES2 ARL LionCove is a thing like Geekbench6 ST 3100ish thing, internal data. You might be able to see 3300ish when it comes out if lucky. Without HT ARL might lose 30% performance from Pcore side and LionCove cannot boost that higher to mitigate the HT loss and IPC is not optimistic at the meanwhile. All the bet just put on E core currently.

Nope. First off, HT loss in ARL isn't 30%. It's upto 30% on very specific cases. Generally, it's around 15% (and maybe even upto 20% on occasions). But when HT is removed from a core, ST gets a solid boost which offsets the HT loss. So, the effective loss due to lack of HT is not that much. Instead, we can expect better ST boost than normal. Thats the reason Lion Cove dropped HT.

In short, ARL is expected to post a solid Geekbench 6 MT score compared to previous gen and compared to competition. All those low MT perf numbers that are projected are pretty much fake/rumors.

 

[mikk]

Interesting, where is this info from?

What exactly? Integrated IMC and LPE? Can't remember all sources but here you go

Page 282 - Discussion - Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads

Page 282 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Page 349 - Discussion - Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads

Page 349 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Some common sense also helps. Because LPE cluster moves into the compute tile they can go down from 6 to 4 P-cores at the same time, because 4 LPE like on LNL can be used properly for real workloads and overall will be 15-20% faster than ARL-H with a 6+8 configuration. PTL is made from the LNL team, it's using the second tile generation unlike ARL which uses the MTL layout with all its flaws. ARL-H is a failure in many ways, they need PTL badly. LNL and PTL is the real deal for Intel.

 

[Henry swagger]

Intel Core Ultra 9 285K rumored to boost up to 5.7 GHz - VideoCardz.com

Flagship Core Ultra 9 SKU with 5.7 GHz boost A rumor shared by OneRaichu suggests clocks will be lower on next-gen desktop CPU series. The Intel Core Ultra 9 285K is currently known as the fastest upcoming Arrow Lake-S processor. The series is said to make use of Lion Cove Performance Cores and...

videocardz.com

Lion cove 5.7ghz boost . Skymont 4.7ghz boost and 4.6 all core 😁🤔.. 285k will have the multi core crown with ease

 

[mikk]

2 passive dies for PTL. Looks like Intel is converging on a typical configuration for their tiled architectures now, where one die is for CPU compute, another die is for the GPU, and the last die is basically IO and everything else. Oh, and a base tile.

https://twitter.com/x/status/1812898063502938260

PCD only 3mm² bigger than on LNL. There is no room for the NPU or LPE CLuster in this (if someone doubted it) , they need a little more size for more PCIe lanes or upgraded wireless/wired connectivity or security engines. Same tile layout as on LNL, only that the GPU goes into a separate tile this time.

 

[Ghostsonplanets]

To confirm my understanding:
Given the similarity of the platform controller to Lunar Lake's, the compute tile has the LP E cores (software can directly use them?) and memory controller, not like in Meteor Lake. But, Arrow Lake still has the old style.

Correct. Arrow Lake inherits the MTL tile layout. But without the Crestmont LPE cores on Desktop while they are retained on Mobile H.

 

[Det0x]

So we're back to:

Thats a 9% or 13% clockspeed deficit depending on if you compare against 14900K or the KS, that you need to overcome with purely IPC

• ..then it need a IPC increase of ~26% to reach the 14900K +20%
• ...then it need a IPC increase of ~30% to reach the 14900KS +20%

14900K +20% ST performance which some speculated earlier not gonna happen

 

[Josh128]

Intel Core Ultra 9 285K rumored to boost up to 5.7 GHz - VideoCardz.com

Flagship Core Ultra 9 SKU with 5.7 GHz boost A rumor shared by OneRaichu suggests clocks will be lower on next-gen desktop CPU series. The Intel Core Ultra 9 285K is currently known as the fastest upcoming Arrow Lake-S processor. The series is said to make use of Lion Cove Performance Cores and...

videocardz.com

Lion cove 5.7ghz boost .

Just like I called it a few days ago. This seems to be TSMCs current frequency limit for high perf designs. N3B does seem to have seen much if any frequency gains compared to N4P/N4X.

 

[inf64]

So we're back to:

14900K +20% ST performance which some speculated earlier not gonna happen

5.7Ghz x 1.15x (IPC) / 6.2Ghz ~= 1.057 or roughly 6% higher ST performance.

 

[dullard]

So we're back to:

14900K +20% ST performance which some speculated earlier not gonna happen

We very well might not get 20% ST performance gain over the 14900K (unless that July 2 leak was not faked). At this point, I think you are correct with that statement.

But, it seems like each time you post you move the goalposts a little further. The original post you quoted was not 20% and was not comparing to the 14900K or 14900KS. (1) It was a range of 15 % up to 20%, (2) it was against Raptor Lake not Raptor Lake Refresh chips, and (3) it mentioned Raptor Lake in general and in no way was that quote solely based on the top two Raptor Lake chips. Here is what you originally quoted:

I am expecting a good 15-20% total single core uplift (ipc + clocks) over Raptor Lake.

15% to 20% ST over a randomly chosen Raptor Lake chip seems quite doable.

 

[Abwx]

5.7Ghz x 1.15x (IPC) / 6.2Ghz ~= 1.057 or roughly 6% higher ST performance.

The guy on Xwitter was apparently right, that s 3% better ST perf at 5.7 than the 14900K.

Maths is 1.14 x 57/60 x 0.955 = 3%.

The 1.14 is the announced improvement, 57/60 the frequency ratio and 0.955 MTL s loss of IPC relatively to RPL, because the 14% improvement is in respect of MTL, not RPL.

 

[Elfear]

We very well might not get 20% ST performance gain over the 14900K (unless that July 2 leak was not faked). At this point, I think you are correct with that statement.

But, it seems like each time you post you move the goalposts a little further. The original post you quoted was not 20% and was not comparing to the 14900K or 14900KS. (1) It was a range of 15 % up to 20%, (2) it was against Raptor Lake not Raptor Lake Refresh chips, and (3) it mentioned Raptor Lake in general and in no way was that quote solely based on the top two Raptor Lake chips. Here is what you originally quoted:


15% to 20% ST over a randomly chosen Raptor Lake chip seems quite doable.

Technically you're right but don't you think the vast majority of desktop users will compare Arrow Lake with the current best available Intel desktop chips (rather than chips launched almost 21 months ago)?

 

[dullard]

Technically you're right but don't you think the vast majority of desktop users will compare Arrow Lake with the current best available Intel desktop chips (rather than chips launched almost 21 months ago)?

I think most would compare i9 to i9, i7 to i7, i5 to i5. I haven't yet seen i7 and i5 rumored speeds. So, not much to go on there at this point in time to know if 15% to 20% is wildly over optimistic or achievable.

You are right that comparing to Raptor Lake Refresh would be a better comparison. But, the 20% number that has been subject to a series of rants wasn't involving the Refresh line.

 

[Henry swagger]

Just like I called it a few days ago. This seems to be TSMCs current frequency limit for high perf designs. N3B does seem to have seen much if any frequency gains compared to N4P/N4X.

N3B has more mask layers and has less performance than n3E.. intel will use n3E for arrow refresh i think 🤔

 

[coercitiv]

Technically you're right but don't you think the vast majority of desktop users will compare Arrow Lake with the current best available Intel desktop chips (rather than chips launched almost 21 months ago)?

The original reply by @eek2121 was about ST performance uplift over Raptor Lake, but in the context of whether AMD can have a shot at leading in single core performance. It was clearly aimed at the flagship parts, so I think a comparison with 13900K(S)/14900K(S) is warranted.

That being said, the exact value is irrelevant at this point, the important part is we should probably adjust expectations from healthy double digit to high single digit at best. The 15-20% guess is more likely to be 10% or lower. Whether the frequency curve favors lower SKUs such as the i5 remains to be seen, there are a lot of factor involved here, some of which may favor such a resolution.

 

[ErichZannMusician]

Geekbench 6 average ST score for 13/14900K = 3100 at 5.7 GHz

Lunar Lake ES Geekbench 6 ST score = 2700 at 4.9 GHz

Geekbench 6 ST PPC uplift on Lion Cove vs Raptor Cove based on ES silicon = 1% or within margin of error.

Add 200 pts ST to the ES score for the final product = 2900 ST, and recalculate PPC uplift = 9% higher PPC for Lion Cove in Lunar Lake.

Normalize clock speeds, then at 5.7 GHz, assuming linear sclaing, Lion Cove in ARL-S would be 16 - 25% faster PPC than Raptor Cove in 13900K.

16% - being conservative, if there is little change between ES and final silicon.
25% - being cautiously optimistic.

Incidentally, this seems to be right in line with what MLID speculated.

The naysayers are in for a surprise.

 

[mikk]

Geekbench 6 average ST score for 13/14900K = 3100 at 5.7 GHz

Lunar Lake ES Geekbench 6 ST score = 2700 at 4.9 GHz

Geekbench 6 ST PPC uplift on Lion Cove vs Raptor Cove based on ES silicon = 1% or within margin of error.

Add 200 pts ST to the ES score for the final product = 2900 ST, and recalculate PPC uplift = 9% higher PPC for Lion Cove in Lunar Lake.

You shouldn't compare a desktop with a mobile chip, IPC won't be comparable even if it's from the same architecture.

 

[dullard]

You shouldn't compare a desktop with a mobile chip, IPC won't be comparable even if it's from the same architecture.

Also, are those comparing Geekbench 6.0 to Geekbench 6.3? We just went over this, they aren't comparable. https://www.geekbench.com/blog/2023/06/geekbench-61/

Thanks to these changes, Geekbench 6.1 single-core scores are up to 5% higher, and multi-core scores are up to 10% higher than Geekbench 6.0 scores. As a result of these methodological differences, which have a non-trivial impact on scores, we recommend users not compare Geekbench 6.1 scores against Geekbench 6.0 scores.

 

[ErichZannMusician]

You shouldn't compare a desktop with a mobile chip, IPC won't be comparable even if it's from the same architecture.

Okay then compare against the Strix scores which should be near-final silicon.

Lion Cove in LNL = Zen 5 in Strix, with Lion cove better at integer workloads than in fp workloads.

 

[Abwx]

Also, are those comparing Geekbench 6.0 to Geekbench 6.3? We just went over this, they aren't comparable. https://www.geekbench.com/blog/2023/06/geekbench-61/

That s precisely GB 6.3.0.

Intel Corporation Lunar Lake Client Platform - Geekbench

Benchmark results for an Intel Corporation Lunar Lake Client Platform with an Intel Core Ultra 7 268V 2.20GHz processor.

browser.geekbench.com

Geekbench Search - Geekbench

The 185H score 2522 pts at 5.087Ghz, while this LNL score 2713-39 at 4.887Ghz, do the maths, that s close to 14%.

Okay then compare against the Strix scores which should be near-final silicon.

Lion Cove in LNL = Zen 5 in Strix, with Lion cove better at integer workloads than in fp workloads.

LNL is better at none, Zen 5 is 7% faster per clock in Geekbench and 5% faster per clock as well in Cinebench wich is FP, if it was weaker in INT then it would be less than 5% difference in GB since this test include both INT and FP.

 

[ErichZannMusician]

LNL is better at none, Zen 5 is 7% faster per clock in Geekbench and 5% faster per clock as well in Cinebench wich is FP, if it was weaker in INT then it would be less than 5% difference in GB since this test iclude both INT and FP.

Entry with Highest ST score for Zen 5 Strix =



Entry with Highest ST score for LNL ES:



Do the math.

Perf/clock is within margin of error even on an ES sample in which the LNC core-only power is consuming perhaps 0.5x-0.6x power of a single Zen 5 core in Strix running ST Geekbench 6.