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	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)	Arrow Lake Refresh (N3B)	Lunar Lake	Panther Lake
Platform	Mobile H/U Only	Desktop Only	Desktop & Mobile H&HX	Desktop Only	Mobile U Only	Mobile H
Process Node	Intel 4	Intel 20A	TSMC N3B	TSMC N3B	TSMC N3B	Intel 18A
Date	Q4 2023	Q1 2025 ?	Desktop-Q4-2024 H&HX-Q1-2025	Q4 2025 ?	Q4 2024	Q1 2026 ?
Full Die	6P + 8P	6P + 8E ?	8P + 16E	8P + 32E	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.)

Hulk · Feb 28, 2024

inf64 said:
With imminent launch of 14900KS, which has around 5-6% higher ST boost versus 13900K, I wonder how will Arrow Lake do against this SKU. Igor's lab and other leak (slide) pointed to ~5% ST increase versus 13900K. If that is still true, then Arrow Lake will have a hard time beating 14900KS in ST tasks.

edit: my bad, I checked the Igor's lab leak and it was normalized to 13900K. Geomean of ST should be ~12% versus 13900K, which should still put arrow Lake at 5-6% faster than 14900KS. So the projection is still valid I guess.

I have been wondering the same thing. I suppose once Intel started binning 14900K silicon they were finding quite a few golden samples that could hit 6.2GHz and couldn't resist milking the "KS" cash cow one more time.

Of course as you mentioned it pushes the ST performance bar ever so slightly higher for ARL. I wonder if the all core frequency for the 14900KS will be higher than 14900K?

SiliconFly · Feb 28, 2024

DavidC1 said:
Raptorlake was very well executed.

Calling Raptor Lake a project is a stretch I believe. It's nothing but golden cove with some minor tweaks to ensure that it isn't golden cove.

SiliconFly · Feb 28, 2024

Geddagod said:
bro what lol

Actually it's true. Apple SoC is both power efficient and area efficient with good performance, unlike the power hungry bloated Intel monsters.

Geddagod · Feb 28, 2024

SiliconFly said:
Actually it's true. Apple SoC is both power efficient and area efficient with good performance, unlike the power hungry bloated Intel monsters.

uh, he was talking about conroe vs the first iphone cpu

DrMrLordX · Feb 28, 2024

Hulk said:
I still think 5% over Raptor Cove in performance meaning 15% increase at ISO frequency from Raptor Cove to Lion Cove due to decrease in clocks and tiles, which kind of makes sense.

If it's the same leak we've all seen, it's isopower (actually isoPL, since the PL value listed for the 8+16 Arrow Lake is PL2 = 251W).

Hulk · Feb 28, 2024

DrMrLordX said:
If it's the same leak we've all seen, it's isopower (actually isoPL, since the PL value listed for the 8+16 Arrow Lake is PL2 = 251W).

Just to be clear, the leak is implying that ARL is performing 5% better than RPL at the same power?

dullard · Feb 28, 2024

Hulk said:
Just to be clear, the leak is implying that ARL is performing 5% better than RPL at the same power?

There are two leaks. The first one is here:

Intel "Arrow Lake-S" Desktop Processor Projected 6%-21% Faster than "Raptor Lake-S"

Intel's future-generation "Arrow Lake-S" desktop processor is already being sampled internally, and to some of the company's closest industry partners, and some of the first performance projections of the processor, comparing it with the current "Raptor Lake-S" (Core i9-13900K), have surfaced...

www.techpowerup.com

That leak claims that for 8+16+1 chips, Arrow Lake-S is 2.4%* to 17.4%* faster than Raptor Lake-S Refresh at PL1=PL2=~250 W power. Presumably this is the 14900K they are comparing. Note: the Arrow Lake chip was using 3 W less power (250 W vs 253 W), but they are roughly iso-power. That leak also says 4.5%* to 19.0%* faster than the Raptor Lake 13900K. Then the graphics (3rd slide in that link) is supposed to blow both Raptor Lake and Raptor Lake Refresh out of the water.

* Note: I'm using the midpoint of the estimations for this whole post, since they are a range.

The second leak is from our very own uzzi38 poster with 5% faster single-thread and 15% faster multi-thread:

Page 105 - Discussion - Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads

Page 105 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Maybe he has context as to which Raptor that is talking about, but if so, I don't remember it.

DrMrLordX · Feb 28, 2024

Hulk said:
Just to be clear, the leak is implying that ARL is performing 5% better than RPL at the same power?

Yes, if it's the old Intel slide. Though the Intel slide showed numbers slightly different than 5%/15%.

Hulk · Feb 28, 2024

Hulk said:
Just to be clear, the leak is implying that ARL is performing 5% better than RPL at the same power?

dullard said:
There are two leaks. The first one is here:

Intel "Arrow Lake-S" Desktop Processor Projected 6%-21% Faster than "Raptor Lake-S"

Intel's future-generation "Arrow Lake-S" desktop processor is already being sampled internally, and to some of the company's closest industry partners, and some of the first performance projections of the processor, comparing it with the current "Raptor Lake-S" (Core i9-13900K), have surfaced...

www.techpowerup.com

That leak claims that for 8+16+1 chips, Arrow Lake-S is 2.4%* to 17.4%* faster than Raptor Lake-S Refresh at PL1=PL2=~250 W power. Presumably this is the 14900K they are comparing. Note: the Arrow Lake chip was using 3 W less power (250 W vs 253 W), but they are roughly iso-power. That leak also says 4.5%* to 19.0%* faster than the Raptor Lake 13900K. Then the graphics (3rd slide in that link) is supposed to blow both Raptor Lake and Raptor Lake Refresh out of the water.

* Note: I'm using the midpoint of the estimations for this whole post, since they are a range.

The second leak is from our very own uzzi38 poster with 5% faster single-thread and 15% faster multi-thread:

Page 105 - Discussion - Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads

Page 105 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Maybe he has context as to which Raptor that is talking about, but if so, I don't remember it.

Hmm. The first two slides are quite interesting. I am not familiar with those workloads but they could have something to do with the difference between RPL and ARL at iso-power. Here's why. Let's assume that the ARL sample tops out at 5.4 or 5.5GHz and we know the 14900K tops out at 6GHz MT and about 5.7GHz multicore. Under a lighter workload it is possible that both the RPL and ARL will hit their max clocks, meaning RPL will have a significant frequency advantage and thus narrow the performance gap between the architectures.

With a heavy workload it is not difficult for RPL to hit power levels much higher than 253 Watts. So it could be possible for RPL under a heavy load at 253W to "only" achieve a frequency of 5.3GHz all core. ARL on the other hand may be much more efficient and be able to maintain its top frequency even on the demanding workloads, thus only benefitting from better IPC over RPL but also higher realized clock speeds.

This could be why we're seeing the spread from 5% (could be light loads) to 17% (heavy loads), in addition to the normal architectural differences we normally attribute to IPC differences between processor generations.

Anyone familiar with how those tested applications saturate the cores?

Honestly this is probably me looking for trends that aren't there.

dullard · Feb 28, 2024

Hulk said:
Anyone familiar with how those tested applications saturate the cores?

Honestly this is probably me looking for trends that aren't there.

I'm not extremely familiar with how saturated those chips would be in those benchmarks. But what you are saying does make theoretical sense. If a chip is running at turbo for extended periods of time, then a more power-efficient chip could be able to run turbo for longer. Otherwise if they have the same number of cores, why would there be better multi-threaded performance gains than single-threaded?

Hulk · Feb 28, 2024

dullard said:
I'm not extremely familiar with how saturated those chips would be in those benchmarks. But what you are saying does make theoretical sense. If a chip is running at turbo for extended periods of time, then a more power-efficient chip could be able to run turbo for longer. Otherwise if they have the same number of cores, why would there be better multi-threaded performance gains than single-threaded?

I'm taking my 14900K performance. I have my power limit set to 200W and P/E frequency capped at 5.5/4.3. For most of my applications like Photoshop, Presonus Studio One, Topaz Video/Photo ai, and others the processor will run at 5.5/4.3 well below 200W. But frameserving in Vegas Pro, or transcoding in Handbrake, or testing in CB it will hit that 200W limit and hold at about 5.1/4.1 or so depending on the application.

In the leaks I'm wondering if RPL is hitting the power limit before max clocks, even before getting into the turbo territory. I have found 1-2 core turbo to be all but useless in my 14900K, which is why I cap frequency at 5.5GHz. I don't need 5.8 or 6GHz for a few seconds now and then and a huge amount of voltage for practically no benefit in performance. This of course is a subjective decision based on personal preference and usage but going over 5.5 for my chip/usage was an exercise in futility. At the end of the day I just capped the voltage at 1.3 with LLC 4. I never hit 1.3V, full load is generally about 1.15V and temps are high '70's with great stability. I know, OT!

AMDK11 · Feb 28, 2024

Hulk said:
I don't know how I missed the fact that Golden/Raptor Cove is not 5+1 but actually 6 simple decoders. Thanks. The Anandtech article didn't have a definitive answer so I guess I assumed 5+1.

The 20% IPC improvement from Skylake to Golden Cove was the result of an additional decoder, 4 additional ports, and a host of other logic changes. It was a significant overhaul. I have to admit I would be surprised if Intel could gain another 20% with the addition of 1 decoder simply because with every evolution of the core there is less "low hanging fruit" as Anand used to say. I should rephrase that and say I would be "happily" surprised.

I am not doubting you but where did you read 3x3 for the Skymont front end? Do we have any reviews that were able to isolate Meteor Lake Crestmont performance iso frequency vs Gracemont or are you quoting Intel for that 4 to 6% increase?

Meteor Lake reviews have been sketchy in my opinion,and I still don't have a good idea of how Redwood Cove and Crestmont in tiled form compare at iso-frequency to Raptor Cove and Gracemont?

Intel did not answer whether the decoder in GoldenCove is 6 simple or 1 complex + 5 simple. It is difficult to give a clear answer to this, and many websites provide different interpretations.

Similarly, in the Intel slides for Skylake, the graphics show 4-way decoding, but without specifying whether it is 1+3.

It seems that Intel provides the decoder width without going into details.

I will only add an Intel graphic that clearly shows the 4-way decoder in Skylake. Nothing has changed in this respect at SunnyCove.

https://en.wikichip.org/w/images/f/f7/skylake_sp_buffer_windows.png

https://images.anandtech.com/doci/14514/Blueprint%20Series_May%2016-2019_COMBINED%20FINAL_AnandTech%20%282%29-page-026.jpg

ondma · Feb 29, 2024

dullard said:
I'm not extremely familiar with how saturated those chips would be in those benchmarks. But what you are saying does make theoretical sense. If a chip is running at turbo for extended periods of time, then a more power-efficient chip could be able to run turbo for longer. Otherwise if they have the same number of cores, why would there be better multi-threaded performance gains than single-threaded?

Could be because the E cores are more improved than the P cores, or dont show a frequency regression.

itsmydamnation · Feb 29, 2024

People are overthinking the complex decode is for things like microcoded instructions etc , things with horrible performance. That complex decoder can still do simple decoding and the vast majority of instructions are simple. So just think of it as 6wide decode

FlameTail · Feb 29, 2024

What is the expected Single Thread performance of Lunar Lake?

Meteor Lake does around 2400 points in Geekbench 6 ST.

If Lunar Lake wants to beat Snapdragon X Elite/Apple M3, it will have to exceed 3000 points.

igor_kavinski · Feb 29, 2024

FlameTail said:
If Lunar Lake wants to beat Snapdragon X Elite/Apple M3, it will have to exceed 3000 points.

It will be a miracle if it does 2800 fanless.

Hulk · Feb 29, 2024

FlameTail said:
Meteor Lake does around 2400 points in Geekbench 6 ST.

Any idea of the frequency it was running to achieve that score?

dullard · Feb 29, 2024

ondma said:
Could be because the E cores are more improved than the P cores, or dont show a frequency regression.

That could be it. Thanks.

Hitman928 · Feb 29, 2024

Hulk said:
Any idea of the frequency it was running to achieve that score?

5 - 5.1 GHz.

SiliconFly · Feb 29, 2024

FlameTail said:
What is the expected Single Thread performance of Lunar Lake?

Meteor Lake does around 2400 points in Geekbench 6 ST.

If Lunar Lake wants to beat Snapdragon X Elite/Apple M3, it will have to exceed 3000 points.

I believe it'll be slower than Meteor Lake due to the low power design it targets. Think same applies to Panther Lake as well.

dullard · Feb 29, 2024

SiliconFly said:
I believe it'll be slower than Meteor Lake due to the low power design it targets. Think same applies to Panther Lake as well.

While your post is mostly correct, it is incomplete. There is some power overlap on some models. The ultra-low power Meteor Lake MTL-U9 chips (currently the 134U and 164U) are 9 W base and 30 W turbo. That is pretty close to the Lunar Lake rumors of 8 W base and 17 W to 30 W turbo.

Meteor Lake might still win in the single thread comparison since Lunar Lake has twice as many P-cores to support with that low power window though.

Also for that reason, I don't think there is a single answer to this question:

FlameTail said:
What is the expected Single Thread performance of Lunar Lake?

With almost a 4X power window range, there will be a massive range of single thread performances across devices depending on their design power.

Tigerick · Feb 29, 2024

I am not sure this is final clock speed. Based on leaks, LNL should have 1.8 GHz base clock with turbo up to 2.8GHz...

Hulk · Feb 29, 2024

Hitman928 said:
5 - 5.1 GHz.

That is right about what Golden/Raptor Cove scores at that frequency. Whatever minor modifications Intel made to Redwood Cove it appears the intent was to offset the latency of the tiles.

dullard · Feb 29, 2024

Tigerick said:
I am not sure this is final clock speed. Based on leaks, LNL should have 1.8 GHz base clock with turbo up to 2.8GHz...

At 8 W power or 30 W or something else? Unfortunately the key piece of data necessary to evaluate those speeds isn't there in that leak.

Hulk · Feb 29, 2024

dullard said:
At 8 W power or 30 W or something else? Unfortunately the key piece of data necessary to evaluate those speeds isn't there in that leak.

We still don't even that have kind of data for Meteor Lake!

You would think it would be relatively easy to have good data for a released product. You know, benchmark score and corresponding frequency and power. Test at 5 or 6 frequencies and boom, review finished.
Nope, we can't really compare anything. This CPU in this chassis has one thermal design, frequencies are all over the place, no idea of actual power usage. We're being gaslit!

I still have no idea of how MTL performs other than similar to RPL but perhaps more efficient somewhere in the curve?

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

Senior member

Attachments

Diamond Member

Golden Member

Golden Member

Golden Member

Lifer

Diamond Member

Elite Member

Lifer

Diamond Member

Elite Member

Diamond Member

Senior member

Platinum Member

Platinum Member

Platinum Member

Lifer

Diamond Member

Elite Member

Diamond Member

Golden Member

Elite Member

Senior member

Diamond Member

Elite Member

Diamond Member