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.)

Abwx · Oct 2, 2024

DavidC1 said:
So what? He's measuring motherboard power, so either AMD's system is a) wasting more system power thus can't allocated much to SoC or b) their sensors are bonked and it's using close to 20W rather than 15W.

You dont get it, it s due to the 258V boost being activated when set at 15W up to 22W while the 370 wont exceed 15W if set at 15W, Asus S16 use 6W on the main when idle and with a 16 OLED, so theories of 30W plateform power when the chip is at 15W is just plain ignorance.

https://www.notebookcheck.net/Asus-Zenbook-S-16-laptop-review-The-first-Copilot-laptop-with-AMD-Zen-5-inside-a-1-3-cm-thick-case.868219.0.html

João Bortolace · Oct 2, 2024

CouncilorIrissa said:
LNL die shots.

https://twitter.com/x/status/1841497643178148185

Can you send here? Twitter is dead in brazil

controlflow · Oct 2, 2024

adroc_thurston said:
Yes costs are a factor you know.

Yes it doesn't come for free, but those fancy PMICs, MoP and generally high level of integration is translating into real world power differences. LNL makes a lot of sense for premium low power thin and light.

Magio · Oct 2, 2024

DavidC1 said:
You can see on the AMD's side, Steam Deck is most efficient with system capped at 15W, so Valve made a good choice.

Yep, the Deck is clearly held back by its aging architecture and process node, but you can clearly see it was built for performance at 15W (and even less) unlike AMD's chips since which fall apart at that kind of power draw despite their superior nodes and archs.

9949asd · Oct 2, 2024

Abwx said:
They go nowhere, this review is just dubbious, if the 370 plateform would need 30W to feed the Soc with 12W then the 370 coudnt score17k at 30W in R23 while the 288V is stuck at 10k at this same power.

The fanboy is the one who is shawn 22W in a screen and is still claming that it s 15W,

Look better your own screenshots, 10% higher fps at 50% more power :

Oh the you wants to compare multicore performance now? Tell me, a 4+4 8c8t compare to 4+8 12c24t, and you proud of it?

DavidC1 · Oct 2, 2024

Magio said:
Yep, the Deck is clearly held back by its aging architecture and process node, but you can clearly see it was built for performance at 15W (and even less) unlike AMD's chips since which fall apart at that kind of power draw despite their superior nodes and archs.

Deck also uses AMD, but that aside I know what you mean.

9949asd said:
Oh the fanboy wants to compare multicore performance now? Tell me, a 4+4 8c8t compare to 4+8 12c24t, and you proud of it?

I would suggest that you would be careful using that word, otherwise I know a ban hammer might be coming, or at least a warning. I have my own opinions regarding AT mods, but just telling you.

Magio · Oct 2, 2024

DavidC1 said:
Deck also uses AMD, but that aside I know what you mean.

I know I was pointing out that the newer AMD designs aren't as well optimized for low power performance as the Deck. It's not even just 15W, Deck can even be made to deliver adequate performance (for less intensive games) at 10W or lower. All that was clearly not a priority for AMD when making Z1 Extreme or Strix Point.

Off topic but I hope Z2 Extreme will be better than Strix at low power, and especially hope that Deck 2 gets a custom chip with input from Valve again.

DavidC1 · Oct 2, 2024

Magio said:
I know I was pointing out that the newer AMD designs aren't as well optimized for low power performance as the Deck. It's not even just 15W, Deck can even be made to deliver adequate performance (for less intensive games) at 10W or lower. All that was clearly not a priority for AMD when making Z1 Extreme or Strix Point.

Off topic but I hope Z2 Extreme will be better than Strix at low power, and especially hope that Deck 2 gets a custom chip with input from Valve again.

I have a feeling AMD is chasing Intel for high clock speeds and paying the price for it. It means the modern AMD chips are prioritized for higher clocks and sacrifice low clock efficiency.

AMDK11 · Oct 2, 2024

poke01 said:
desktop x86 CPUs absolutely. You see the reason why the M chips have such high ST is because Apple wanted to make an all purpose core that can used in mobile and desktop without sacrificing ST for mobile. If you go with that foundation for building chips you get the M series.

This is IPC, i.e. comparison at the same clock speed.

The x86 design cores from Intel and AMD provide high clock speeds, which on the one hand simplifies the design, but also makes the core logic less dense than it could be. High clock speeds mean many compromises in the core microarchitecture.

Apple designs cores for much lower clock speeds, so it can have a pipeline with fewer stages, lower cache latencies, and greater capacity. Apple cores are the power of huge amounts of resources: L1-I 192KB, L1-D 128KB, 10-Wide decoding, ROB 900+ 8xALU and very low latency.

That's it in a nutshell and simplified way.

H433x0n · Oct 2, 2024

Abwx said:
They go nowhere, this review is just dubbious, if the 370 plateform would need 30W to feed the Soc with 12W then the 370 coudnt score17k at 30W in R23 while the 288V is stuck at 10k at this same power.

The fanboy is the one who is shawn 22W in a screen and is still claming that it s 15W,

Look better your own screenshots, 10% higher fps at 50% more power :

This isn't "rigging" the results. The Phawx has noted something similar as mentioned here. I've queued it up to begin where he talks about it at ~17:10 in the video. The power reporting is less accurate on the AMD system, he typically compares an AMD CPU at 15W to an Intel CPU at ~20W and finds that they both will pull similar power. This would suggest that a LNL system that shows 22W (remember 2W for the memory modules) would be equivalent to the total package power draw as an AMD chip at 15W.

I know you're going to totally dismiss this but figured it's worth mentioning anyway.

DavidC1 · Oct 2, 2024

AMDK11 said:
Apple designs cores for much lower clock speeds, so it can have a pipeline with fewer stages, lower cache latencies, and greater capacity. Apple cores are the power of huge amounts of resources: L1-I 192KB, L1-D 128KB, 10-Wide decoding, ROB 900+ 8xALU and very low latency.

That's it in a nutshell and simplified way.

And Intel/AMD needs to follow that direction, because I suspect large L1 caches are also contributing to efficiency in their designs because it keeps lot of data from going out into slower, higher power cache levels and memory.

The basis of Apple's design is also having stellar design team, because being able to have 4GHz clocks at such low power at just 9 pipeline stages and humongous L1 cache with 3 cycle latency is amazing. The E core team can still do quite a bit better, which is catch up with Apple's. It is even more impressive when you consider Mx chips have been stagnating for a while with smart engineers all into "AIeeee" hype and moving to startups.

DavidC1 · Oct 2, 2024

H433x0n said:
This isn't "rigging" the results. The Phawx has noted something similar as mentioned here. I've queued it up to begin where he talks about it at ~17:10 in the video. The power reporting is less accurate on the AMD system, he typically compares an AMD SoC at 15W TDP to an Intel CPU at ~20W and finds that they both will pull similar power. This would suggest that a LNL system that shows 22W (remember 2W for the memory modules) would be equivalent to the total package power draw as an AMD chip at 15W.

This was also apparent for AMD's GPUs(at least the Polaris generation I don't know how the newer generations behave), where "board power" meant without the GDDR memory and the competitor Nvidia could accurately measure real board power.

poke01 · Oct 2, 2024

DavidC1 said:
And Intel/AMD needs to follow that direction, because I suspect large L1 caches are also contributing to efficiency in their designs because it keeps lot of data from going out into slower, higher power cache levels and memory

Qualcomm follows the same cache hierarchy as M series and it fails in efficiency. Core architecture matters too I guess

adroc_thurston · Oct 2, 2024

controlflow said:
but those fancy PMICs, MoP and generally high level of integration is translating into real world power differences.

Obviously but the price ain't nice.

controlflow said:
LNL makes a lot of sense for premium low power thin and light.

It's a one-off and for a good reason.

DavidC1 said:
I have a feeling AMD is chasing Intel for high clock speeds and paying the price for it

Everyone is chasing clocks. You're like a year away from aa64 parts getting close to or breaching 5G mark.

poke01 said:
Core architecture matters too I guess

and physical implementation

DavidC1 · Oct 2, 2024

poke01 said:
Qualcomm follows the same cache hierarchy as M series and it fails in efficiency. Core architecture matters too I guess

Of course. And there are matters which are impossible to see from a high level. You could give two design teams exactly the same requirements, heck even the same base uarch and one will simply end up being better than the other.

Some people are just better at some things than others.

9949asd · Oct 2, 2024

DavidC1 said:
Deck also uses AMD, but that aside I know what you mean.

I would suggest that you would be careful using that word, otherwise I know a ban hammer might be coming, or at least a warning. I have my own opinions regarding AT mods, but just telling you.

Well, some one is blind for the test results. This is not the first time in the thread. Just few days ago, the same thing showing 370 using 60+w and lunar lake using 40w. But fanboys keep claiming the soc power are the same on 28w🤣

adroc_thurston · Oct 2, 2024

DavidC1 said:
You could give two design teams

Well it's the same design team lmao.

9949asd said:
But fanboys keep claiming the soc power are the same on 28w

Platform power != SOC power.

DavidC1 · Oct 2, 2024

9949asd said:
But fanboys keep claiming the soc power are the same on 28w🤣

If you can't control yourself than I have to suggest you have a bit of problem yourself. You can feel uncomfortable about something but still be more civil.

As they say, if you want to stay, you gotta "play the game".

9949asd · Oct 2, 2024

DavidC1 said:
If you can't control yourself than I have to suggest you have a bit of problem yourself. You can feel uncomfortable about something but still be more civil.

As they say, if you want to stay, you gotta "play the game".

Excuse me? Am I in a AMD thread ?

AMDK11 · Oct 2, 2024

poke01 said:
Qualcomm follows the same cache hierarchy as M series and it fails in efficiency. Core architecture matters too I guess

Core architecture is the key that matters most (IPC). Intel claims that future generations will no longer focus on high clock speeds, but on IPC. It will be interesting to see where this goes.

DavidC1 · Oct 2, 2024

H433x0n said:
This would suggest that a LNL system that shows 22W (remember 2W for the memory modules) would be equivalent to the total package power draw as an AMD chip at 15W.

By the way, it isn't entirely accurate to quote 2W. I bet this is maximum, similar to TDP numbers. Unless you run some sort of a memory power virus, you'll be hard pressed to even use 1W.

I can tell you on my Kabylake-Y Yoga that 1W is very rarely often reached. This is even more true on Lunarlake where the memory controller and the PHY itself is optimized for even lower power and has MLC to reduce access further.

Intel is clearly aware of this, thus for the most part the "2W" number is used to give more power budget for the rest of the SoC such as the CPU and GPU.

Abwx · Oct 2, 2024

H433x0n said:
This isn't "rigging" the results. The Phawx has noted something similar as mentioned here. I've queued it up to begin where he talks about it at ~17:10 in the video. The power reporting is less accurate on the AMD system, he typically compares an AMD CPU at 15W to an Intel CPU at ~20W and finds that they both will pull similar power. This would suggest that a LNL system that shows 22W (remember 2W for the memory modules) would be equivalent to the total package power draw as an AMD chip at 15W.

I know you're going to totally dismiss this but figured it's worth mentioning anyway.

That s wrong, for the S16 Computerbase measured 28W in CB in HVinfo and NBC measured a little more than 30W for CB at the main, so AMD s telemetry is very accurate and it so for many years, beside you re talking of the Z1ex, dunno what it has to do with the 370.

All numbers for 288V and 370 can be found here :

Intel Core Ultra 9 288V „Lunar Lake“ im Test

Es ist so weit: Nach einem Jahr Salamischeibenmarketing ist Intel Lunar Lake ab sofort verfügbar. Core Ultra 200V überzeugt im Test.

www.computerbase.de

DavidC1 · Oct 2, 2024

AMDK11 said:
Core architecture is the key that matters most (IPC). Intel claims that future generations will no longer focus on high clock speeds, but on IPC. It will be interesting to see where this goes.

Therefore history doesn't repeat but rhymes. It is mini Netburst to Core all over again. Raptorlake degradation is Pentium III 1.13GHz but 23 years later. The motivation is quite similar. Both pushed clocks too high to better go against competition, which in both cases is AMD.

KompuKare · Oct 2, 2024

DavidC1 said:
Transistor counts don't really matter because they give wildly differing ones. Performance, performance per watt, peak clocks, performance at certain area, those matter.

Of course some armchair cost calculations would be interesting too but there are far too many hidden things.

It used to be that Intel's costs were pretty opaque but while they too are now on TSMC their new CPUs do use fancy new packaging technology (which I maintain is Intel's best investment in years) so even if we could estimate yields and guess TSMC's wafer costs - well we have no way to guess how much packaging costs are.

Costs only one factor. As Apple demonstrate if a costly SOC can command a high price due to great performance and perf/watt whereas a slightly cheaper SOC may not, then going cheap may be false economy.

Not expecting any price war, but I would still like to know who could afford to start one; I suspect Intel's production costs are higher and while their volumes make their fixed costs less of a factor it makes little sense for them start any price war. The other x86 vendor is no longer a budget brand so doubt they'd start anything.

Then again I may just love the idea of taking the area, going a yield calculator and a bit of playing with a spreadsheet later come up with some guesses

DavidC1 · Oct 2, 2024

KompuKare said:
Costs only one factor. As Apple demonstrate if a costly SOC can command a high price due to great performance and perf/watt whereas a slightly cheaper SOC may not, then going cheap may be false economy.

Silicon is called 21st century Goldmine for a reason.

The basic materials are really cheap. It's more expensive now(mostly because of processing such as EUV and R&D) hence why computer prices are rising, when it has been falling drastically. I remember Dell PC ads featuring $7000 desktops for the average consumer. Or how Pentium 4 was $1499 cdn.

The dual core Sandy Bridge generation for example was about 100mm2 and those were sold as cheap and readily as stamps would. It's like $5 for die and $5 for packaging, which makes die size vs price arguments silly. And they would turn it around and sell the top parts for $150(the amount Intel makes, not MSRP) or more. Or how 600mm2 server chips cost maybe $150 to make and is sold for many thousands of dollars.

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

Senior member

Attachments

Lifer

Member

Member

Member

Member

Golden Member

Member

Golden Member

Senior member

Golden Member

Golden Member

Golden Member

Platinum Member

Diamond Member

Golden Member

Member

Diamond Member

Golden Member

Member

Senior member

Golden Member

Lifer

Golden Member

Golden Member

Golden Member