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)	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.)

 

[Geddagod]

How can you know? Afaik Royal Core is more like a project name. The old reddit leak claimed Panther Cove is an entirely new architecture from the ground up.

Raichu claimed that old reddit leak was wrong- almost certainly bcuz it stated that PTC was a new arch from the ground up.
Xino also said a week ago that RYC was no longer part of NVL.
PNC was also apparently supposed to be used in PTL at first, and PTL was never supposed to use RYC originally anyway.

I'm guessing PTL got downgraded to Cougar Cove (LNC+? ++?) and NVL was going to use either PNC (which would be the next "new" P-core arch) or whatever the first "RYC" core is. PTL finished the definition stage, and it looks like it got locked in for PNC.

Kinda reminds me of how MTL might have been the first "next gen architecture" with Ocean Cove, but that got scrapped for continuing the regular Cove line. Perhaps RYC is facing similar pushback.

 

[mikk]

Raichu claimed that old reddit leak was wrong- almost certainly bcuz it stated that PTC was a new arch from the ground up.
Xino also said a week ago that RYC was no longer part of NVL.
PNC was also apparently supposed to be used in PTL at first, and PTL was never supposed to use RYC originally anyway.

I'm guessing PTL got downgraded to Cougar Cove (LNC+? ++?) and NVL was going to use either PNC (which would be the next "new" P-core arch) or whatever the first "RYC" core is. PTL finished the definition stage, and it looks like it got locked in for PNC.

Kinda reminds me of how MTL might have been the first "next gen architecture" with Ocean Cove, but that got scrapped for continuing the regular Cove line. Perhaps RYC is facing similar pushback.

It wouldn't surprise me if RYC is dead like Ocean Cove. Or maybe they really need more time but how long do they need, another 5 years or what. I mean Nova Lake comes later than expected two years ago, in the reddit leak it was 2025 and now it seems to be 2026. Even with this delays RYC didn't make it, something went wrong I guess.

 

[HoveringStyle]

There is no mention at what TDP did i7-13700H achieve that score.
MTL will have up to 2x more powerful IGP(33% more units, 50% higher clockspeed at 2.25GHz), so It could theoretically reach 47FPS, but at an unknown TDP.
I don't think It should matter. It's not like you can play a lot of PC games using Apple SoC.

Almost certainly less than 45W real usage, testing on my own 12500H laptop (80EU Iris Xe) I get 21.24 fps while limited to 28W, any higher doesn't improve it beyond margin of error. The 13700H has a theoretical 38% better compute but it's better binned and it's clearly bottlenecked in some way as it's only 11% faster.


I wouldn't put that much stock on Wild Life performance though, from what I've seen, Intel performs much better relative to AMD and Nvidia on it, Time Spy graphics aligns much better with actual gaming performance. IDK the relative standing of the arm competitors, the new Steel Nomad benchmark might help with that.

 

[coercitiv]

I found out Ultra 7 155U has only 2.7GHz base, but It is a 4P+8E+2LE model.
What is more surprising is that Ultra 5 135U with 2P+8E+2LE cores has only 2.1GHz base clock. Logically, It should have a higher base clock than 155U.

In this particular case I reckon there's more to it than binning: let's not forget these chips are TDP configurable. That being said, the screenshots you posted yesterday also show a 500Mhz delta between the major SKU tiers. Normally we would expect to see this delta in boost clocks due to a combo of binning and segmentation, but base clocks should be closer.

We'll have to wait for i5 vs. i7 reviews on the same notebooks to gauge the impact. My guess is nowadays Intel is binning more aggressively at the top as a way to be more competitive and also justify the i7 purchase over i5.

 

[TESKATLIPOKA]

In this particular case I reckon there's more to it than binning: let's not forget these chips are TDP configurable. That being said, the screenshots you posted yesterday also show a 500Mhz delta between the major SKU tiers. Normally we would expect to see this delta in boost clocks due to a combo of binning and segmentation, but base clocks should be closer.

It looks like Intel really wants to make the difference between tiers bigger.
4P+8E Ultra 5 125h -> Base: 3.3GHz (100%)
6P+8E Ultra 7 155h -> Base: 3.8GHz (115%)
Yet with Raptor It was the exact opposite
4P+8E i5 1360P -> Base: 2.2GHz (115%)
6P+8E i7 1370P -> Base: 1.9GHz (100%)
But this base doesn't really matter, see below

We'll have to wait for i5 vs. i7 reviews on the same notebooks to gauge the impact. My guess is nowadays Intel is binning more aggressively at the top as a way to be more competitive and also justify the i7 purchase over i5.

CB R15 MT loop:
Lenovo ThinkPad T14s G4 i7-1365U: Ø1122 (1097.76-1428.93)
Lenovo ThinkPad T14 G4 i5-1335U: Ø1062 (993.99-1540.3)
Both of them have ~22.5 PL1.
Despite having a much higher base, the real difference is only 10%(lowest score), which can be attributed to better binning.

Probably this is the reason why Ultra 7 155U has also 2 extra P-cores, so the difference in performance will be larger.

 

[TESKATLIPOKA]

In the same review, you can also find Lenovo ThinkPad T14s G4 with AMD Ryzen 7 7840U at 25W PL1: Ø1811 (1798.56-1931.99)
That's a huge 64% difference.
i7-1365U would have needed ~3.6GHz P-cores and ~2.6GHz E-cores for a comparable performance.

The question is If 2P+8E MTL can go that high with 25W PL1 or a bit lower depending on IPC gains.
I am kinda optimistic thanks to Ultra 5 125H and Ultra 7 155H.
This would be a great performance, considering 2P+8E is more or less equal to 6P cores in performance (Intel vs Intel).

 

[mikk]

Seems like we should see some real availability this month. This is different to many other mobile launches in the past, even Raptor Lake mobile wasn't available for some time.

Spotted by @Tech_Reve, the Samsung Galaxy Book 4 Ultra laptop was leaked over at a Chinese 3rd party seller, GooFish, and comes packed with the Intel Core Ultra 7 155H CPU.

Samsung Galaxy Book 4 Ultra Laptop With Intel Core Ultra 7 155H CPU & RTX 4050 GPU Being Sold For $1765 US

Samsung's upcoming Galaxy Book 4 Ultra laptop featuring the next-gen Intel Core Ultra "Meteor Lake" CPUs has been spotted.

wccftech.com

 

[Abwx]

In the same review, you can also find Lenovo ThinkPad T14s G4 with AMD Ryzen 7 7840U at 25W PL1: Ø1811 (1798.56-1931.99)
That's a huge 64% difference.
i7-1365U would have needed ~3.6GHz P-cores and ~2.6GHz E-cores for a comparable performance.

The question is If 2P+8E MTL can go that high with 25W PL1 or a bit lower depending on IPC gains.
I am kinda optimistic thanks to Ultra 5 125H and Ultra 7 155H.
This would be a great performance, considering 2P+8E is more or less equal to 6P cores in performance (Intel vs Intel).

2 + 8 MTL wont match a 7840U, the gap is too big, they would need 2x better perf/watt at isoperf, that s in the NBC article you linked :

What's exciting here is that the AMD processor offers almost the same multi-core performance at 12 watts as the Core i7-1355U at 24 watts.

 

[TESKATLIPOKA]

2 + 8 MTL wont match a 7840U, the gap is too big, they would need 2x better perf/watt at isoperf, that s in the NBC article you linked :

Sorry, just because you said so doesn't mean It can't happen.
The difference is big, true, but I already said why I think It could happen based on what we already know about MTL.

Raptor Lake is seriously limited by power, which is evident from very low base clocks, but MTL has them much higher at the same TDPs.
It's still unknown If the same is also true for E-cores.

At worst, I am wrong and It will be weaker, doesn't really matter. It's not like this 2P+8E is aimed against the 8C Phoenix, more like the 6C one or PHX2.

 

[HoveringStyle]

Here's a more direct comparison, the i7-1255U gets 1062 points at 20W in CB15 MT, vs 1025 for the 7840U at 12W, so the latter is 60% more efficient at roughly iso performance, a large gap still but not 2x.

Besides Alder Lake and derivates suffer at low power limits because the uncore uses ~4W by itself (AMD's mobile chips use ~1W) if that can be improved, it'll help move the needle a bit for Intel.

 

[SiliconFly]

MLID talks about MTL in this new video, and it's hysterical: youtube link

He says the usual stuff like MTL is bad, Intel is stupid, AMD is better, Ryzen rocks, etc., etc. But after a while he starts to talk about MTL's efficiency. And after watching the segment, it's clear he's actually stupid and doesn't understand tech that much. He's convinced MTL is gonna be only 8% more efficient than RPL in normal usage scenario (his own words, but carefully includes a lots of if and buts in between to cover his behind just in case).

I think it's centered on his outmoded belief that the CPU tile is the one thats going to drive MTL when in reality it's gonna be the SoC tile thats gonna be active most of the time for most of the users in the world.

Anyways, next thursday is going to a very interesting day.

 

[TESKATLIPOKA]

Here's a more direct comparison, the i7-1255U gets 1062 points at 20W in CB15 MT, vs 1025 for the 7840U at 12W, so the latter is 60% more efficient at roughly iso performance, a large gap still but not 2x.

Besides Alder Lake and derivates suffer at low power limits because the uncore uses ~4W by itself (AMD's mobile chips use ~1W) if that can be improved, it'll help move the needle a bit for Intel.

That's not what's interesting.
Check out how 2P+8E performs against i7-1260P or i7-12700H.
At 15-20W It wins against more cores.
At 28W 12700H finally wins, but is only 14% faster and at 35W It is 19.5% faster, that's not a big difference considering the core count.

 

[Saylick]

MLID talks about MTL in this new video, and it's hysterical: youtube link

He says the usual stuff like MTL is bad, Intel is stupid, AMD is better, Ryzen rocks, etc., etc. But after a while he starts to talk about MTL's efficiency. And after watching the segment, it's clear he's actually an total dimwit who doesn't understand tech that much. He's convinced MTL is gonna be only 8% more efficient than RPL in normal usage scenario (his own words, but carefully includes a lots of if and buts in between to cover his behind just in case).

I think it's centered on his outmoded belief that the CPU tile is the one thats going to drive MTL when in reality it's gonna be the SoC tile thats gonna be active most of the time for most of the users in the world.

Anyways, next thursday is going to a very interesting day.

Compared to RPL, I fully expect MTL to be far more efficient, especially in low power or idle situations. How MTL stacks up compared to PHX? I'm not entirely convinced MTL will be more efficient but it should be far more competitive.

MTL's biggest claim to fame in the efficiency department appears to stem from its ability to shut off the compute die and just rely on two lower power E cores on the SOC tile, which itself is fabricated using TSMC N6.

Meanwhile, Rembrandt already had fine-grained clock and power gating on a per core basis which carried over to PHX, so even though it may not be a complete shutting off of idle cores, PHX as a whole is built on a newer fabrication process (TSMC N4).

We'll just have to wait for reviews to know for sure.

EDIT: Corrected a mistake regarding the node used for MTL's SOC tile. Also added a slide for the power islands for PHX.

 

[SiliconFly]

...1062 points at 20W... ...1025 points at 12W... ...is 60% more efficient at roughly iso performance...

Thats not the right way to compare. Power curves tend to be exponential.

 

[jpiniero]

and just rely on two lower power E cores on the SOC tile, which itself is fabricated using Intel 7.

TSMC N6

 

[Saylick]

TSMC N6

My mistake. I'll correct my previous post. Thanks for the catch.

 

[Thunder 57]

MLID talks about MTL in this new video, and it's hysterical: youtube link

He says the usual stuff like MTL is bad, Intel is stupid, AMD is better, Ryzen rocks, etc., etc. But after a while he starts to talk about MTL's efficiency. And after watching the segment, it's clear he's actually stupid and doesn't understand tech that much. He's convinced MTL is gonna be only 8% more efficient than RPL in normal usage scenario (his own words, but carefully includes a lots of if and buts in between to cover his behind just in case).

I think it's centered on his outmoded belief that the CPU tile is the one thats going to drive MTL when in reality it's gonna be the SoC tile thats gonna be active most of the time for most of the users in the world.

Anyways, next thursday is going to a very interesting day.

I go out of my way to not watch "leakers" and give them money indirectly and an audience.

 

[HoveringStyle]

Thats not the right way to compare. Power curves tend to be exponential.

I know, and relative efficiency does increase for the 1255U at 15W but it scores 896 and I was trying to keep the "iso performance" comparison made by Abwx.

That's not what's interesting.
Check out how 2P+8E performs against i7-1260P or i7-12700H.
At 15-20W It wins against more cores.
At 28W 12700H finally wins, but is only 14% faster and at 35W It is 19.5% faster, that's not a big difference considering the core count.

Yes! and it also shows us some possible insight as to why the P series was axed, there wasn't a single point in which 4+8 was the best performing configuration, even at its default 28W.

 

[TESKATLIPOKA]

Yes! and it also shows us some possible insight as to why the P series was axed, there wasn't a single point in which 4+8 was the best performing configuration, even at its default 28W.

P-series was axed? It's more like It was renamed than axed. That TDP is still used with MTL chips.

 

[coercitiv]

P-series was axed? It's more like It was renamed than axed. That TDP is still used with MTL chips.

Not only that, P cores are still more power efficient in most consumer oriented scenarios.

 

[TESKATLIPOKA]

Not only that, P cores are still more power efficient in most consumer oriented scenarios.

That depends on workload and frequency(TDP).

On these picture you can see some very interesting things.
1.) Gracemont for the same performance needs 68% higher clockspeed in a vectorized workload
2.) For an integer workload It actually needs only 24% higher clock for the same performance
I am just not sure If HT was enabled or not, probably not.

So for laptops with <=35W It makes sense to use them not just because of area efficiency, but also some additional power efficiency.
I already showed before, using more P-cores at low TDPs is totally pointless, they provide 0 benefit.

Gracemont is good until 3GHz, after that kaboom.

Of course, you still need P-cores, but E-cores are not a bad thing, just that scheduling is a problem.
If MTL allows a separate voltage regulation for E-cores, that would be a good thing.

 

[DrMrLordX]

If MTL allows a separate voltage regulation for E-cores, that would be a good thing.

Raptor Lake allows a separate voltage plane for ecores already. Not that all Raptor Lake platforms take advantage of it.

 

[mikk]

MLID talks about MTL in this new video, and it's hysterical: youtube link

He says the usual stuff like MTL is bad, Intel is stupid, AMD is better, Ryzen rocks, etc., etc. But after a while he starts to talk about MTL's efficiency. And after watching the segment, it's clear he's actually stupid and doesn't understand tech that much. He's convinced MTL is gonna be only 8% more efficient than RPL in normal usage scenario (his own words, but carefully includes a lots of if and buts in between to cover his behind just in case).

Last time he said MTL is like Whiskey Lake and now it's better than Raptor Lake....Ok. Why is the power diagram starting at 50W? MTL is mainly a 28W model from what we know. And why is the gap at 90W bigger than at 50W, that doesn't make sense to me. But sure he hates everything from Intel and hypes everything from AMD lately, not a good source I guess. Did he hype the exciting Phoenix refresh?

 

[TESKATLIPOKA]

Raptor Lake allows a separate voltage plane for ecores already. Not that all Raptor Lake platforms take advantage of it.

Are you sure? I thought It was not implemented in the end.
BTW, which platforms use It? Mobile ones?

 

[DrMrLordX]

Are you sure? I thought It was not implemented in the end.
BTW, which platforms use It? Mobile ones?

Apparently DLVR is enabled on mobile Raptor Lake but not desktop.