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.)

 

[FlameTail]

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.

Slower in MT? Sure.

But slower in ST?

 

[SiliconFly]

Slower in MT? Sure.

But slower in ST?

Especially is ST.

It'd be a surprise if it even catches up with MTL in ST. I think (sustained) boost clocks will be significantly lower than mainstream cpus just to keep it within power & thermal envelope.

 

[Markfw]

Is it just me, or is there a lot of confusion on how well Meteor lake does in all aspects, power, performance, efficiency, etc... ??????

 

[igor_kavinski]

View attachment 94648

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...

Max 2.8 GHz boost means it is definitely targeting fanless design like M1 Macbooks.

It may also perform similarly with very low max TDP.

 

[igor_kavinski]

Is it just me, or is there a lot of confusion on how well Meteor lake does in all aspects, power, performance, efficiency, etc... ??????

It's Intel's 2nd Tiled beta CPU after Lakefield. Third time's the charm with Lunar Lake. However, I only expect it to beat maybe the fastest U series 9W Raptor Lake CPU.

 

[Tigerick]

Max 2.8 GHz boost means it is definitely targeting fanless design like M1 Macbooks.

It may also perform similarly with very low max TDP.

I am not so sure, even though LNL is targeting M-series, the two-chiplet design and x86 CPU core might make difference in clock speed. Let me check with my source ...

 

[FlameTail]

Let me check with my source ...

You have sources?

Max 2.8 GHz boost means it is definitely targeting fanless design like M1 Macbooks.

It may also perform similarly with very low max TDP.

So LNL's performance is going to be around M1 level, while it's competition is the M3?

Sounds like an L for Intel.

 

[Tigerick]

How do you get 7.7 Tflops? If Xe3 uses SIMD16 ALUs like Xe2 it's double than that unless they lower the number of vector engines per Xe core.

Your question puzzles me as well, turn up BMC dGPU's design has reduced numbers of EU by half with doubling ALU. That's means we are still dealing with same amount of ALU, thus 7.7TF should remain the same.

 

[Tigerick]

OK, here comes some leaks not from me but credential leaker about ARL & PTL:

• ARL-H (6P+8E) comes with iGPU (Build by newer process, N4P) 8 Xe1 LPG+ cores clocking at 2.3GHz
• ARL-HX (8P+16E, H and HX's tCPU are different dies) comes with 4 Xe cores
• SoC has been updated to support WiFi7, still on N6
• PTL-H's tCPU integrates with 4P+8E+4LPe without HT, SoC and MC
• Here comes interesting part, PTL's tGPU comes with 12 Xe3 LPG core clocking at 2.5GHz. Yeah it is GT3 with 7.7 TF, slightly higher than Strix Point. And they are built on Intel 3+ process. Finally, IFS build tGPU.
• ADM so far is dead.

Regarding PTL-H's, there is new update on tGPU tile. I was surprised by IFS able to make tGPU with 12 Xe cores, turn up that part is made by TSMC's N3E process...

IFS did manage to make 4 Xe cores with Intel 3-E process. Yeah, IFS still have long way to go..

That's why sometimes the leaks need updating too

 

[mikk]

View attachment 94648

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...

Your are not sure lol, this is a 0000 CPU which is never final. It should answer your question.

Regarding PTL-H's, there is new update on tGPU tile. I was surprised by IFS able to make tGPU with 12 Xe cores, turn up that part is made by TSMC's N3E process...

Do you have a source for your 12 Xe claim? I haven't heard from it.

 

[deasd]

MeteorLake based MSI Claw ~40% more expensive than Rog Ally

https://twitter.com/x/status/1763562802348749142

 

[igor_kavinski]

MeteorLake based MSI Claw ~40% more expensive than Rog Ally

Good luck to it CLAWing back marketshare for Intel

 

[H433x0n]

New results for MTL, that is pretty close to being apples/apples. The laptops being compared for all of these CPUs are ROG G14 & G16.

Interesting to see how well Hawk Point performs. It performs well enough to hold off MTL from getting a clean win, which was surprising to me.

 

[SiliconFly]

New results for MTL, that is pretty close to being apples/apples. The laptops being compared for all of these CPUs are ROG G14 & G16.

Interesting to see how well Hawk Point performs. It performs well enough to hold off MTL from getting a clean win, which was surprising to me.

That looks interesting. Ultra 9 beats Ryzen 9 by a wide margin!

Is it a one off result? Cos, it's in stark contrast with older results where Ryzen was kicking Ultra's behind.

 

[Hitman928]

New results for MTL, that is pretty close to being apples/apples. The laptops being compared for all of these CPUs are ROG G14 & G16.

Interesting to see how well Hawk Point performs. It performs well enough to hold off MTL from getting a clean win, which was surprising to me.

Hmm, seems to be a significant difference between the 185h and 155h here. Is the power limit here PL1 or PL2? The JT video on the Asus model with the 185h showed no real difference between it and the 155h when limited to the same power and more closely matches the 155h shown here at 28 w.

Edit: I think JT isn’t a strict limit either, actually, until you get up around 30 W as the Asus has a much lower PL2 than most others.

 

[igor_kavinski]

New results for MTL, that is pretty close to being apples/apples. The laptops being compared for all of these CPUs are ROG G14 & G16.

Interesting to see how well Hawk Point performs. It performs well enough to hold off MTL from getting a clean win, which was surprising to me.

Umm...that just looks like Intel's desktop-turned-mobile silicon dominating Hawk Point beyond its intended TDP.

But yeah, the lower TDP performance of Hawk Point is very nice.

 

[Abwx]

New results for MTL, that is pretty close to being apples/apples. The laptops being compared for all of these CPUs are ROG G14 & G16.

Interesting to see how well Hawk Point performs. It performs well enough to hold off MTL from getting a clean win, which was surprising to me.

These curves are boggus, the 155H and 185H perfs increase about linearly with power from 10W to 30W, there s no transistors that scale like this, at the very best the curve is a root square law when it comes to perf/power and of course a square law when that s power/perf.

This can be possible only if there s a turbo whose power is proportional to the base power at each curve point, in wich case the perf/power is artificialy inflated by benching at say 25W for a 15W base power FI.

 

[Hitman928]

Umm...that just looks like Intel's desktop-turned-mobile silicon dominating Hawk Point beyond its intended TDP.

But yeah, the lower TDP performance of Hawk Point is very nice.

The 8945hs is marketed as having a 45W default TDP. If the 185h really was able to match the 8945hs at this power level, it would be a big deal for Intel. Problem is that I'm pretty sure the power levels shown are PL1 settings and not a hard power limit being enforced, so the extra power the CPUs are allowed to use from the PL2 window is coming into play.

Edit: To illustrate, NBC has benchmark results for the 185h in an Asus laptop (review not yet posted but you can still view the benchmark results). The 185h in their review sample scores 12,688 points in CB r23 which is roughly what the above screenshot is showing for the 185h at 30W. However, the 185h in the NBC review has a PL2 of 64W and the CB r23 test isn't long enough to not have that be a major influence. JT's results also showed the 185h at 11,179 with a PL1 of 28 W and PL2 enabled. So I don't think there's any way you could get the 185h to score as shown above without PL2 being enabled and the CPU boosting to above the shown power levels for at least some part of the run.

 

[H433x0n]

The 8945hs is marketed as having a 45W default TDP. If the 185h really was able to match the 8945hs at this power level, it would be a big deal for Intel. Problem is that I'm pretty sure the power levels shown are PL1 settings and not a hard power limit being enforced, so the extra power the CPUs are allowed to use from the PL2 window is coming into play.

Found the video that accompanies the graphs.

 

[Hitman928]

Found the video that accompanies the graphs.

Thanks for find that. In their table, they say they are setting PL1 = PL2 for their power testing, but their power consumption graph with their score for repeated tests doesn't seem to agree:

This is the 185h over multiple runs:



Corresponding power consumption:




So with a base power of 70W and boost power of 95 W for roughly 30% of each run (looking at the later runs), the 185h scores ~18350.

Then, in the table and on their performance over power chart, they show the 185h scoring the same amount with PL1=PL2=65W.



There's no possible way this is accurate unless the PL2 isn't being enforced correctly and it is still boosting as before.

Edit: Same applies for the 8945hs in the video.

 

[Abwx]

MTL perf/Watt is overestimated by about 20-23% on thoses truncated graphs, it is reported as 65W a score that was actually performed at a 80W average, same for the score at 75W wich is actually performed on the first run at a 92W average.

Indeed why should the 185H be that far apart of a 155H wich is exactly the same chip using the same core count, process and so on, if not by the magical effect of a higher turboing and an even more magical curves doctoring.

 

[SiliconFly]

These curves are boggus, the 155H and 185H perfs increase about linearly with power from 10W to 30W, there s no transistors that scale like this, at the very best the curve is a root square law when it comes to perf/power and of course a square law when that s power/perf.

This can be possible only if there s a turbo whose power is proportional to the base power at each curve point, in wich case the perf/power is artificialy inflated by benching at say 25W for a 15W base power FI.

Those curves in no way represent "transistors". MTL has a mix of logic that execute instructions at different power levels based on various factors. And they're pretty much capable of scaling performance from extreme low power levels to moderate to high due to the lpe cores to e cores to p cores respectively. Just because MTL can perform better than Ryzen, doesn't mean it's all bogus.

 

[Abwx]

Those curves in no way represent "transistors". MTL has a mix of logic that execute instructions at different power levels based on various factors. And they're pretty much capable of scaling performance from extreme low power levels to moderate to high due to the lpe cores to e cores to p cores respectively. Just because MTL can perform better than Ryzen, doesn't mean it's all bogus.

It perform better if it use way more power, that s not unheard off by Intel s standards, if absolute perf is not enough then crank the watts up and rate the whole thing at a ridiculously lower TDP than what is actually consumed, guess that you re aware of DT CPUs rated at 125W and using 253W on the benches, that s exactly the same for mobile parts.

 

[DavidC1]

The whole thing about miracle firmware seems applicable only to a very select amount of devices. Ultrabookreview hasn't updated their article either.

https://www.notebookcheck.net/Asus-Zenbook-Duo-OLED-2024-convertible-review-Multitasking-champion-with-Core-Ultra-9.807413.0.html

The 185H is as disappointing as the 155H.

We still don't even that have kind of data for Meteor Lake!
...
I still have no idea of how MTL performs other than similar to RPL but perhaps more efficient somewhere in the curve?

It's still not good enough to compete.

Also the thing LP-e cores seem to be good at doing is sleeping and wasting even the little silicon space it takes.

 

[adroc_thurston]

just to keep it within power & thermal envelope.

LNL is a 30W thingy.
So not really?

Ultra 9 beats Ryzen 9 by a wide margin!

at 90w.