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 (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 ?	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.)

MarkizSchnitzel · Aug 6, 2024

A leak on notebookcheck said 1200-1300€ for A er Swift 14.

Magio · Aug 6, 2024

Lunar Lake lives and dies by the quantity and quality of its design wins in the premium thin and light segment. LNL should be within expectations in terms of performance (with even decent gaming capabilities) and battery life for that segment but Intel will have to rely on its OEM partners to deliver laptops attractive enough (design, build quality, ...) to make people opt for them over MacBooks which won't be priced that much higher.

The big marketing angle will probably be AI but it's a huge TBD whether consumers actually want Copilot+/NPUs.

FlameTail · Aug 6, 2024

Serious question;

Do you guys think Lunar Lake could beat the battery life of Snapdragon X Elite?

Hitman928 · Aug 6, 2024

FlameTail said:
Serious question;

Do you guys think Lunar Lake could beat the battery life of Snapdragon X Elite?

Technically, it's possible, meaningfully, not so sure.

gdansk · Aug 6, 2024

FlameTail said:
Serious question;

Do you guys think Lunar Lake could beat the battery life of Snapdragon X Elite?

Yes. Not all laptops. But there will be some LNL laptops with better battery life than some SDXE laptops.

Magio · Aug 6, 2024

FlameTail said:
Serious question;

Do you guys think Lunar Lake could beat the battery life of Snapdragon X Elite?

If I had to bet today I'd say it will win on some loads (medium loads and video playback), lose on others (low loads and idle) and be largely equal in a "whole picture" scenario.

GTracing · Aug 6, 2024

FlameTail said:
Serious question;

Do you guys think Lunar Lake could beat the battery life of Snapdragon X Elite?

I think Lunar Lake might match the X Elite in some workloads. I don't think it'll beat it all around.

Intel's QuickSync engine combined with the low power island and better manufacturing nodes should give lunar lake excellent battery life in video playback. I think this is where Lunar Lake will be at its best.

In mixed light/medium usage, they'll be pretty close. I would guess the X Elite will pull ahead. Though as CPUs get more efficient, they become less and less of a factor in the overall battery life of a device in light workloads. If it's 15 hours vs 14 hours then it will only matter for bragging rights.

In heavy single threaded load, the X Elite will win unless Intel puts a sane cap on the single core power draw. If you limit both chips to 8W power draw, then the efficiency will be similar.

In heavy multi threaded benchmarks the X Elite will win hands down, assuming similar power limits. If the X Elite is given an 80W power limit then lunar lake will obviously "win" in efficiency.

dullard · Aug 6, 2024

Since no one posted it, Intel discussed a little bit about 18A today (Panther Lake and Clearwater Forest).

Intel Foundry Achieves Major Milestones

Intel 18A powered on and healthy, on track for next-gen client and server chip production next year.

www.intel.com

and

Kevin O’Buckley Talks Progress on Intel 18A

New Foundry Services leader reveals why he came to Intel, excitement for the foundry business and why Intel 18A matters to the future of computing.

www.intel.com

Key takeaways (these are direct quotes from the links above):

Both Panther Lake and Clearwater Forest booted without additional configurations
Panther Lake DDR memory performance already running at target frequency
First external customer is expected to tape out on Intel 18A in the first half of next year.
The Panther Lake client processor is powered on and booting Windows, yielding well, in use inside Intel and ahead of schedule on product qualification milestones
Clearwater Forest for data center is powered on, booting operating systems, in use inside Intel and performing well

Reading between the two lines above, they did not specifically mention yielding well for Clearwater Forest, nor did they specifically mention performing well for Panther Lake. That doesn't necessarily mean anything. But it is interesting to have those left out.

Kosusko · Aug 6, 2024

AVX10.2

Intel® Advanced Vector Extensions10.2

Architecture Specification

July, 2024

Revision 1.0

source: https://cdrdv2.intel.com/v1/dl/getContent/828965

LLVM/Clang 20 Compiler Begins Seeing Intel AVX10.2 Support

source: https://www.phoronix.com/news/LLVM-Clang-20-Intel-AVX10.2#google_vignette

Ghostsonplanets · Aug 6, 2024

dullard said:
Since no one posted it, Intel discussed a little bit about 18A today (Panther Lake and Clearwater Forest).

Intel Foundry Achieves Major Milestones

Intel 18A powered on and healthy, on track for next-gen client and server chip production next year.

www.intel.com

and

Kevin O’Buckley Talks Progress on Intel 18A

New Foundry Services leader reveals why he came to Intel, excitement for the foundry business and why Intel 18A matters to the future of computing.

www.intel.com

Key takeaways (these are direct quotes from the links above):

Both Panther Lake and Clearwater Forest booted without additional configurations

Panther Lake DDR memory performance already running at target frequency

First external customer is expected to tape out on Intel 18A in the first half of next year.

The Panther Lake client processor is powered on and booting Windows, yielding well, in use inside Intel and ahead of schedule on product qualification milestones

Clearwater Forest for data center is powered on, booting operating systems, in use inside Intel and performing well

Reading between the two lines above, they did not specifically mention yielding well for Clearwater Forest, nor did they specifically mention performing well for Panther Lake. That doesn't necessarily mean anything. But it is interesting to have those left out.

Good to see that Panther Lake is yielding well and ahead of schedule. I genuinely believe it will be a good product. And the variants of PTL, that will release later, are going to provide Intel much needed competitiveness in some key markets.

DavidC1 · Aug 6, 2024

GTracing said:
Intel's QuickSync engine combined with the low power island and better manufacturing nodes should give lunar lake excellent battery life in video playback. I think this is where Lunar Lake will be at its best.

Meteorlake has a low power island, a "proper" one with it on an entirely different die.* It saves an amazing 150mW in video playback, a whopping 4% increase in battery life.

In Lunarlake, there are other factors that will give it real gains:
-System Cache, which will cut traffic to memory and reduce power. Keeping it as close as possible to the core is touted as being a necessity to lower power in all future compute.
-The Skymont E core actually scales really well. It seems like at least half the battery life improvement will come from having workload reside in Skymont. They claim Skymont's ability to scale is what enables Microsoft Teams to reside entirely on the E core cluster, when Meteorlake LPE was too slow to do more than basic(and don't forget about the 0.15W improvement).
-A more sane 2-tile setup from a 4-tile one in Meteorlake.
-Unlike previous implementation that just has LPDDR support, Lunarlake actually optimizes it to lower power. They say the PHY power is also 40% lower than the predecessor, matching the SoC gains.

*Meteorlake's tile approach and the LPE core may be an example of what sounded good in high level not working in reality. Having it in another tile means it needs to actually hand off workloads between two tiles without both of them being active for a significant amount of time.

Ghostsonplanets said:
Good to see that Panther Lake is yielding well and ahead of schedule. I genuinely believe it will be a good product. And the variants of PTL, that will release later, are going to provide Intel much needed competitiveness in some key markets.

Color me skeptical.
-Does Pantherlake have a System Level Cache?
-It has more tiles again.
-Why are they separating E and LPE again? Will LPE be on another tile like Meteorlake? If separation of E and LPE exist, then at least it should be on the same tile. High communication latency is responsible for practical battery life not being high as planned.
-What about a proper low power optimized LPDDRx PHY? I doubt they'll do it for the entire lineup, because it'll likely hurt performance in some way.

Performance-wise it might be fine. What about battery life? Will Lunarlake be a one-hit wonder with no successor like many previous Intel products that was supposed to be great?

Are they so cautious that they can't dedicate a Lunarlake successor? Because without such specialization it will suffer in either market, or be mediocre in both. The cynic in me says there's nothing after Lunarlake because they are confident Lunarlake will render the WoA market dead, and thus a lesser successor will be enough.

DavidC1 · Aug 6, 2024

Intel Foundry needs a big advantage over TSMC in technicals and do it successfully for many years before they'll take any significant amount of share.

How long did AMD successfully execute over Intel until now? Think about why they are saying growth won't come until 2030. This is assuming they execute. Maybe they'll get 5% at best. 10% will be a huge win. 2030 is about 5 years away now. Not a long time.

Having industry standard tools is just a start. You still need to work as a customer to move things from TSMC to Intel, no? Like if you are used to working on Office for many years, would you move to OpenOffice? What about Blender to other 3D design programs? Even if the alternative gave you a real easy way to migrate you still need to take a day or few to migrate them, because it's disruptive to your workflow.

With them saying 14A offers a mere 15% performance and 20% density advantage, this is going to be a painful many years before the Foundry achieves financially material impact.

Relying on Taiwan being taken for Intel Foundry to succeed is the biggest delusion there is. Because if Taiwan is taken, they'll instantly lose 20% of total revenue on China being gone as a customer. And surely a World War 3 is bullish right?

Ghostsonplanets · Aug 6, 2024

DavidC1 said:
-Does Pantherlake have a System Level Cache?

AFAIK yes

DavidC1 said:
-It has more tiles again.

Correct. It needs to cover more markets (U to H) and thus needs more flexible tiles than an ultra-specialized part like LNL. But PTL only has one extra tile than LNL. It's no MTL, be assured.

DavidC1 said:
-Why are they separating E and LPE again? Will LPE be on another tile like Meteorlake? If separation of E and LPE exist, then at least it should be on the same tile. High communication latency is responsible for practical battery life not being high as planned.

LPE and E live on the same tile. And they have the new core topology (Which I believe debuts with ARL? Not sure) which allows LPE cores to be used on any workloads. It's a proper, Arm like, multi-cluster scheduler scheme this time. Unlike MTL weird one.

DavidC1 said:
-What about a proper low power optimized LPDDRx PHY? I doubt they'll do it for the entire lineup, because it'll likely hurt performance in some way.

That I do not know.

DavidC1 said:
What about battery life?

They're aiming at better than Lunar Lake last I heard. But without any specifics or if it only applies to certain SKUs.

DavidC1 said:
Will Lunarlake be a one-hit wonder with no successor like many previous Intel products that was supposed to be great?

LNL is a new swimlane that has no proper successor so far, yes. PTL-U is the closest thing to a successor due to 4P + 4LPE and >50 TOPs NPU and some SKUs that might feature on-package memory. But the iGP is half the size.

I think, if LNL proves a success, Intel will integrate them into the standard generation product line-up. Maybe we'll see something proper with Nova Lake Mobile or Razor Lake (Lake Next after NVL) Mobile.

DavidC1 said:
Are they so cautious that they can't dedicate a Lunarlake successor? Because without such specialization it will suffer in either market, or be mediocre in both. The cynic in me says there's nothing after Lunarlake because they are confident Lunarlake will render the WoA market dead, and thus a lesser successor will be enough.

As I said, I think it's a matter of new swimlane, different priorities and Intel going on a faster schedule.

Lunar Lake will need to prove it to be popular for Intel to create further iterations. And Intel is on a faster schedule where they're trying to bring new mobile designs faster to compensante for the lethargy of past years. So some SKUs will co-exist.

And by priorities, I mean there's only so much projects they can tackle while there's a lot of markets they need to cover. They're doing Panther Lake U and H. That's already a lot of SKUs that need to be ready.

But they identified a gap into their product line-up that is the fact they have no SKU/Product that can attend the budget market outside of RPL 282. Thus, they choose to do Wildcat Lake, which is a "Panther Lake U Lite" for the budget market. Same technologies/IP (Core, GFX, NPU) as PTL, but in a much smaller configuration.

I'm sure by NVL and/or RZL they'll have everything sorted out to give every segment a new product in the same cadence.

DavidC1 · Aug 6, 2024

Ghostsonplanets said:
Lunar Lake will need to prove it to be popular for Intel to create further iterations. And Intel is on a faster schedule where they're trying to bring new mobile designs faster to compensante for the lethargy of past years. So some SKUs will co-exist.

Lunarlake is currently the only promising client part for Intel. Arrowlake seems nothing special. 5% ST? We used to get that with Ticks.

Ghostsonplanets said:
They're aiming at better than Lunar Lake last I heard. But without any specifics or if it only applies to certain SKUs.

Unless Pantherlake has something that covers deficiencies, don't count on it.

Lunarlake proves that they need a separate lineup for low power. They had to do that for server right?

LPE and E live on the same tile. And they have the new core topology (Which I believe debuts with ARL? Not sure) which allows LPE cores to be used on any workloads. It's a proper, Arm like, multi-cluster scheduler scheme this time. Unlike MTL weird one.

By logic LPE can't be "used on any workloads". It is limited by it's performance. Lunarlake's two core setup can already use the E cores for boosting performance. That means the "LPE" core is going to be low frequency again.

Also 4+8+4LPE isn't gonna be a performance leader, similar to not counting Meteorlake as 6+8+2. 6+8 is already low. How will they cover the -H market in 2026?

Correct. It needs to cover more markets (U to H) and thus needs more flexible tiles than an ultra-specialized part like LNL. But PTL only has one extra tile than LNL. It's no MTL, be assured.

I hoped they would continue the excellent battery life that Cherry Trail Tablets had. In Braswell they completely gave that up. Why? Because by eliminating that lineup they could save extra few hundred million. They need a specialized lineup again not a generic one.

SoCs aimed at respective markets
-Server P
-Server E for Cloud/VM
-Client High end mobile to Client Desktop: 45W-125W
-Client ultra battery life<--9W to 35W

511 · Aug 7, 2024

DavidC1 said:
Meteorlake has a low power island, a "proper" one with it on an entirely different die.* It saves an amazing 150mW in video playback, a whopping 4% increase in battery life.

In Lunarlake, there are other factors that will give it real gains:
-System Cache, which will cut traffic to memory and reduce power. Keeping it as close as possible to the core is touted as being a necessity to lower power in all future compute.
-The Skymont E core actually scales really well. It seems like at least half the battery life improvement will come from having workload reside in Skymont. They claim Skymont's ability to scale is what enables Microsoft Teams to reside entirely on the E core cluster, when Meteorlake LPE was too slow to do more than basic(and don't forget about the 0.15W improvement).
-A more sane 2-tile setup from a 4-tile one in Meteorlake.
-Unlike previous implementation that just has LPDDR support, Lunarlake actually optimizes it to lower power. They say the PHY power is also 40% lower than the predecessor, matching the SoC gains.

*Meteorlake's tile approach and the LPE core may be an example of what sounded good in high level not working in reality. Having it in another tile means it needs to actually hand off workloads between two tiles without both of them being active for a significant amount of time.

Color me skeptical.
-Does Pantherlake have a System Level Cache?
-It has more tiles again.
-Why are they separating E and LPE again? Will LPE be on another tile like Meteorlake? If separation of E and LPE exist, then at least it should be on the same tile. High communication latency is responsible for practical battery life not being high as planned.
-What about a proper low power optimized LPDDRx PHY? I doubt they'll do it for the entire lineup, because it'll likely hurt performance in some way.

Performance-wise it might be fine. What about battery life? Will Lunarlake be a one-hit wonder with no successor like many previous Intel products that was supposed to be great?

Are they so cautious that they can't dedicate a Lunarlake successor? Because without such specialization it will suffer in either market, or be mediocre in both. The cynic in me says there's nothing after Lunarlake because they are confident Lunarlake will render the WoA market dead, and thus a lesser successor will be enough.

Meteor Lake is way efficient than the Intel 7 RPL-H the battery life advantage is really nice but that doesn't come from LP-E as most workload go to the normal E cores and are kept there for the most part
As for PTL tiles i think the LPE and E exists on the same die 18A for PTL it is NPU+CPU+Media IP+IMC on 18A GPU on N3E/I3 misc on N6

511 · Aug 7, 2024

DavidC1 said:
With them saying 14A offers a mere 15% performance and 20% density advantage, this is going to be a painful many years before the Foundry achieves financially material impact.

For this part sadly it is the truth from TSMC as well node shrinks will be limited to 20% and PPW by 10-15% this is true to industry you can look at N2/A16

Doug S · Aug 7, 2024

DavidC1 said:
With them saying 14A offers a mere 15% performance and 20% density advantage, this is going to be a painful many years before the Foundry achieves financially material impact.

Maybe you need to take a look at what TSMC is promising for upcoming shrinks before using "mere" to describe what 14A is offering. That's pretty much par for the course for now on unless/until we figure out how to shrink SRAM cells or replace them with something that will shrink.

Henry swagger · Aug 7, 2024

Doug S said:
Maybe you need to take a look at what TSMC is promising for upcoming shrinks before using "mere" to describe what 14A is offering. That's pretty much par for the course for now on unless/until we figure out how to shrink SRAM cells or replace them with something that will shrink.

Well said

coercitiv · Aug 7, 2024

Doug S said:
Maybe you need to take a look at what TSMC is promising for upcoming shrinks before using "mere" to describe what 14A is offering. That's pretty much par for the course for now on unless/until we figure out how to shrink SRAM cells or replace them with something that will shrink.

David's argument was that Intel needs to be significantly better than TSMC to steal business from them. The fact that laws of physics seem to deny us progress on further nodes does not change the sales and marketing conundrum for Intel when competing with TSMC.

Just as he said, AMD executed very well for many years before they started getting more customers in servers and mobile, and they're still facing a head wind. Inertia is huge in this field.

Ghostsonplanets · Aug 7, 2024

DavidC1 said:
Unless Pantherlake has something that covers deficiencies, don't count on it.

Lunarlake proves that they need a separate lineup for low power. They had to do that for server right?

I do agree it's unlikely. Specially as LNL had to be pretty specialized to get the battery life gains it had. I can only assume they're banking on further improved fabrics, idle and better tuned load consumption due to BSPDN and RibbonFET

DavidC1 said:
By logic LPE can't be "used on any workloads". It is limited by it's performance. Lunarlake's two core setup can already use the E cores for boosting performance. That means the "LPE" core is going to be low frequency again.

Yes. What I meant is that it can be leveraged on MT workloads and work as an proper core. Rather than MTL P + E -> LPE

DavidC1 said:
Also 4+8+4LPE isn't gonna be a performance leader, similar to not counting Meteorlake as 6+8+2. 6+8 is already low. How will they cover the -H market in 2026?

That honestly I don't know. Maybe Desktop 35W silicon will be brought to Laptops to cover the 45W/55W luggable/High-end Gaming Laptops markets.

DavidC1 said:
I hoped they would continue the excellent battery life that Cherry Trail Tablets had. In Braswell they completely gave that up. Why? Because by eliminating that lineup they could save extra few hundred million. They need a specialized lineup again not a generic one.

I do agree with you. Been saying for years that the killing of Atom and Core M is one of the reasons Intel got so behind in power efficiency. Short-sighted gains back them translated into a struggling future currently.

DavidC1 said:
SoCs aimed at respective markets
-Server P
-Server E for Cloud/VM
-Client High end mobile to Client Desktop: 45W-125W
-Client ultra battery life<--9W to 35W

- Agreed. Intel will need to be very agressive here

- According to Exist, that's dead after CWF. Dunno how Intel will handle the server many cores after the merging of E into P. Maybe Zen C like but for Intel P core?

- Agreed

- Agreed. Though there's need to many SoCs/dies/tiles in this specific segment as there's varying levels of performance requirements but also pricing. WCL, for example, is basically due to Intel tiled + advanced packaging approach being expensive and thus they needed a different approach to provide budget markets a successor (I believe AMD will also struggle here with Zen 6).

FlameTail · Aug 7, 2024

DavidC1 said:
Also 4+8+4LPE isn't gonna be a performance leader, similar to not counting Meteorlake as 6+8+2. 6+8 is already low. How will they cover the -H market in 2026?

Considering how strong Skymont E-core is, I'd guess 8E+4LPE = 8P.

Then 4P+8P+4LPE = 12P

FlameTail · Aug 7, 2024

Ghostsonplanets said:
According to Exist, that's dead after CWF. Dunno how Intel will handle the server many cores after the merging of E into P. Maybe Zen C like but for Intel P core?

How is P-core and E-core team merged, if they are based in different countries?

Ghostsonplanets · Aug 7, 2024

FlameTail said:
How is P-core and E-core team merged, if they are based in different countries?

? You're aware core design is made on multiple countries, right? And also remote work.

Besides, what Exist (and others) were trying to say is that both teams will be merged into one to develop a single core design with influences from both + Royal leftover work (supposedly). To be fair, he also did say it's not 100% certain that E Team will be under P Team. Could be the opposite. But it's very likely that P Team wins the politics at Intel.

dttprofessor · Aug 7, 2024

Intel must increase more profits first through LNL/ARL/SRF/CWF/PTL（maybe 10-20B$），whatever products on TSMC or internal fab or any other fab,before the next Grand Strateg。 Pat must buy time for intel & himself.

511 · Aug 7, 2024

FlameTail said:
Considering how strong Skymont E-core is, I'd guess 8E+4LPE = 8P.

Then 4P+8P+4LPE = 12P

Apparently darkmont is to skymont what Crestmont was to Gracemont also at lower Power i bet E core wins vs P core as evident with Crestmont Redwood Cove they need to bring Quite better PPA than Strix Point 12/24T

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

Senior member

Attachments

Senior member

Member

Diamond Member

Diamond Member

Platinum Member

Member

Member

Elite Member

Member

Intel® Advanced Vector Extensions10.2​

Architecture Specification​

July, 2024​

Revision 1.0​

LLVM/Clang 20 Compiler Begins Seeing Intel AVX10.2 Support​

Senior member

Senior member

Senior member

Senior member

Senior member

Senior member

Senior member

Platinum Member

Senior member

Diamond Member

Senior member

Diamond Member

Diamond Member

Senior member

Member

Senior member

Intel® Advanced Vector Extensions10.2

Architecture Specification

July, 2024

Revision 1.0

LLVM/Clang 20 Compiler Begins Seeing Intel AVX10.2 Support