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.)

 

[igor_kavinski]

For some reason MSI also decided to set this laptop up with dual SSDs in Raid 0, seems like an odd choice for this class of laptop. Seems potentially suboptimal for a low power device.

It's a 16 inch laptop. More targeted towards people who will sit at their desks/coffee shops for longer periods of time doing heavy duty stuff than the mobile warriors carrying 13.3/14-inch laptops. I wouldn't mind the extra speed.

 

[mikk]

I do wonder if the tested firmware is not working correctly in terms of shutting down the unneeded tiles, but we'll apparently have to wait another few weeks at minimum to see if things improve with an updated firware. I'm honestly not too hopeful at this point, but there is a real possibility of improvement if Intel rushed the release to try and keep their 2023 promise. If that's the case, it probably would have been smarter to "launch" the CPU like a month ago, tell everyone it's shipping now to partners, and then announce laptop availability will be in Q1 of 2024. Basically what AMD did.

All of of the reviews with Cinebench R23 scores are more in-line with the non updated pcode scores from Golden Pig. The scheduling fix must be new. This might be unrelated to the LP-E behavior but shows how unoptimized it runs at the moment.

 

[igor_kavinski]

Has any site taken infrared images of the CPU while it's being pushed 100% (all tiles being used full blast)? Would be interesting to see which tile emits the most heat.

 

[Hitman928]

Here's the Hothardware chart normalized for battery life with and without adjusting the non-MTL laptops for 91% charge (not sure which is accurate). I only did the top 5 as I didn't want to spend a ton of time looking up all the battery sizes but as you can see, the MTL laptop goes from 1st to 4th out of the top 5 when normalizing for the size of the battery in each unit. This, of course, doesn't account for other variables between the laptops.

Original Data


My chart which includes normalizing to the MSI Prestige 16 AI EVO battery size.



Edit: fixed MSI laptop not properly normalized with the 91% calculation.

 

[Hitman928]

And what will they do after abandoning LPE?

Lunar Lake appears to be a 2 tile design with all of the processing (NPU, GPU, CPU) happening on the compute tile and no processing cores on the SOC tile.

 

[Philste]

Problem is, Intel is rumored to be abandoning the LPE cores after ARL.

Look at the LNL Layout again. It has 4 LPE-Cores. They may be back on the CPU Tile to get the newest process as well, but they don't have access to the L3. LNL has an SLC for them. So LNL is: 4 P-Cores with L3 and 4 Off-Ring LPE-Cores that may have access to the SLC.

 

[FlameTail]

Look at the LNL Layout again. It has 4 LPE-Cores. They may be back on the CPU Tile to get the newest process as well, but they don't have access to the L3. LNL has an SLC for them. So LNL is: 4 P-Cores with L3 and 4 Off-Ring LPE-Cores that may have access to the SLC.

How supremely bizarre

 

[Geddagod]

Look at the LNL Layout again. It has 4 LPE-Cores. They may be back on the CPU Tile to get the newest process as well, but they don't have access to the L3. LNL has an SLC for them. So LNL is: 4 P-Cores with L3 and 4 Off-Ring LPE-Cores that may have access to the SLC.

source?

Lunar Lake appears to be a 2 tile design with all of the processing (NPU, GPU, CPU) happening on the compute tile and no processing cores on the SOC tile.

pogchamp


Also, there are rumors that with PTL, Intel is just moving to combine the CPU and SOC tile on the same node. This would pretty much eliminate the need for separate SOC cores, while also not having to deal with the weird memory bandwidth and latency shenanigans that come from separating the SOC and compute tiles.

 

[Hitman928]

Look at the LNL Layout again. It has 4 LPE-Cores. They may be back on the CPU Tile to get the newest process as well, but they don't have access to the L3. LNL has an SLC for them.

I haven't seen any LNL die shots, where did you see the layout?

So LNL is: 4 P-Cores with L3 and 4 Off-Ring LPE-Cores that may have access to the SLC.

Are you saying that LNL has no regular E-cores and is just P-cores with 4 LPE-Cores? Doesn't seem right. . .

 

[Hitman928]

So MTL is doing really bad on Linux.

Intel Core Ultra 7 155H CPU Fails Miserably Against The AMD Ryzen 7 7840U In Linux Benchmarks

Intel's Core Ultra 7 155H "Meteor Lake" CPU gets crushed by AMD's Ryzen 7 7840U in Linux tests, falling behind in performance & efficiency.

wccftech.com

The graphics side has a much better showing in their tests, beating out the Radeon 780m. Their collection of games tested leaves a lot to be desired, though.

Edit: The Ryzen laptop also had much slower RAM which will obviously make a difference with GPU tests.

Intel Meteor Lake Arc Graphics: A Fantastic Upgrade, Battles AMD RDNA3 Integrated Graphics Review - Phoronix

www.phoronix.com

 

[Khato]

My chart which includes normalizing to the MSI Prestige 16 AI EVO battery size.

FYI, you went the wrong way with the 91% charge calculation. That would decrease their relative advantage compared to the MSI Prestige 16 AI EVO, not increase it. eg, with the HP Dragonfly Pro the calculation is (903/64.6)*99.9*.91 = 1270. Also of note is that the MSI and Dell Latitude are 2560x1600 displays rather than 1920x1200.

 

[Hitman928]

FYI, you went the wrong way with the 91% charge calculation. That would decrease their relative advantage compared to the MSI Prestige 16 AI EVO, not increase it. eg, with the HP Dragonfly Pro the calculation is (903/64.6)*99.9*.91 = 1270. Also of note is that the MSI and Dell Latitude are 2560x1600 displays rather than 1920x1200.

No, the 91% charge calculation is normalizing them to 100% charge, assuming they had a 91% charge in the Hot Hardware test. With that said, I forgot to normalize the MSI result to 100% as well, I will fix that.

 

[mikk]

Ultrabookreview updated the firmware in their Asus Zenbook review to the latest Bios 203 and the score went from 11126 to 12132 at 28W in his loop test compared to Bios 201. Maybe it comes with the updated pcode?

I reran and updated the Cinebench scores again, with the laptop fresh from a cold boot up. In Cinebench R23, the CPU starts at 50W and ends up around 35W for the best-effort run. For the 10 min loop, it starts at 50W and then gradually drops and stabilizes at around 28W (after about 5-6 minutes it gets to 28W).

Bios 201
CineBench R23: CPU 12048 pts (best run), CPU 11126 pts (10 min loop test), CPU Single Core 1751 pts (best run);

Bios 203
CineBench R23: CPU 13873 pts (best run), CPU 12132 pts (10 min loop test), CPU Single Core 1761 pts (best run);


In the Notebookcheck review the Acer at 28W scored 10573 points which is even worse than the Asus on bios 201.

 

[igor_kavinski]

Bios 203
CineBench R23: CPU 13873 pts (best run)

This is good. With beefier cooling, it would be decent. Let's hope Minisforum releases a Core 9 185H mini-ITX mobo soon.

 

[Geddagod]

I haven't seen any LNL die shots, where did you see the layout?

He might be referencing to the thermal image of LNL that maybe is real, but that doesn't show much. Rough guestimations on die size and layout at best, but if anything it appears as if the E-cores are right below the -P-core cluster, and is prob on the same interconnect.

Are you saying that LNL has no regular E-cores and is just P-cores with 4 LPE-Cores? Doesn't seem right. . .

Lol

 

[Geddagod]

Ultrabookreview updated the firmware in their Asus Zenbook review to the latest Bios 203 and the score went from 11126 to 12132 at 28W in his loop test compared to Bios 201. Maybe it comes with the updated pcode?



Bios 201
CineBench R23: CPU 12048 pts (best run), CPU 11126 pts (10 min loop test), CPU Single Core 1751 pts (best run);

Bios 203
CineBench R23: CPU 13873 pts (best run), CPU 12132 pts (10 min loop test), CPU Single Core 1761 pts (best run);


In the Notebookcheck review the Acer at 28W scored 10573 points which is even worse than the Asus on bios 201.

the sustained results improvement at 28 watts suspiciously match this graph:

 

[Khato]

No, the 91% charge calculation is normalizing them to 100% charge, assuming they had a 91% charge in the Hot Hardware test. With that said, I forgot to normalize the MSI result to 100% as well, I will fix that.

Default behavior is to charge to 100%. They made a specific note in that review that MSI configured different than default behavior resulting in a start of test charge level of 91%. Which is actually a great thing that more manufacturers should be doing as keeping lithium ion batteries at 100% charge level when plugged for long periods of time is worse than 100 -> 0 -> 100 charge cycling. But they don't... Why? Simplest explanation is planned obsolescence.

 

[mikk]

the sustained results improvement at 28 watts suspiciously match this graph:
View attachment 90763

Roughly 12K for 28W should be accurate for the updated pcode, I guess it's included in Bios 203. I'm curious how it will affect the game benchmarks which he wants to update as well. Maybe it won't change but this needs to be tested. They still have to fix the low power issue below 25W.

 

[Hitman928]

Default behavior is to charge to 100%. They made a specific note in that review that MSI configured different than default behavior resulting in a start of test charge level of 91%. Which is actually a great thing that more manufacturers should be doing as keeping lithium ion batteries at 100% charge level when plugged for long periods of time is worse than 100 -> 0 -> 100 charge cycling. But they don't... Why? Simplest explanation is planned obsolescence.

Here is the quote

The battery was only 91% charged when we started the test due to Windows' battery protection mechanisms.

 

[Khato]

Here is the quote

Correct. And if you look up what Microsoft has to say about Window's battery protection - https://support.microsoft.com/en-us...-windows-2ac1b4ba-6027-410a-b80e-f6767d867002 - it's entirely dependent on the manufacturer to implement. Hence both assumptions that it's enabled by default on all laptops and would result in the same 91% starting state of charge are incorrect.

 

[Abwx]

the sustained results improvement at 28 watts suspiciously match this graph:
View attachment 90763

That curve is bs, it s just PL2 that is increased to boost the score.

In the new curve we can see that from 20W to 40W the score increase from 9500 to roughly 14500, that s 61% higher perf for 2x the power and that s just impossible by the very laws of semiconductors physics.

61% higher perf require 3x the power with a very good process that scale close to a square law wich is the theorical best case for mosfets, so this curve is just made up, anyone with some knowledge of electronics will spot such a discrepancy.

 

[mikk]

That curve is bs, it s just PL2 that is increased to boost the score.

In the new curve we can see that from 20W to 40W the score increase from 9500 to roughly 14500, that s 61% higher perf for 2x the power and that s just impossible by the very laws of semiconductors physics.

61% higher perf require 3x the power with a very good process that scale close to a square law wich is the theorical best case for mosfets, so this curve is just made up, anyone with some knowledge of electronics will spot such a discrepancy.

There was a scheduling issue in the old pcode, you are wrong. You can see this also in the updated ultrabookreview Cinebench scores where it scored 12.1K after 10 Minutes at 28W.

 

[Khato]

That curve is bs, it s just PL2 that is increased to boost the score.

In the new curve we can see that from 20W to 40W the score increase from 9500 to roughly 14500, that s 61% higher perf for 2x the power and that s just impossible by the very laws of semiconductors physics.

61% higher perf require 3x the power with a very good process that scale close to a square law wich is the theorical best case for mosfets, so this curve is just made up, anyone with some knowledge of electronics will spot such a discrepancy.

No, it's quite possible and doesn't violate any laws of semiconductor physics.

The 7840HS result in that graph goes from ~8k at 15W to ~13k at 30W. That's also roughly 61% increase in performance for 2x the power. And those data points match up with other reviews.

With respect to semiconductor physics, keep in mind that power is linear with frequency and squared with voltage. So it's expected to have closer to linear scaling at lower power levels where the percentage increase in voltage is smaller for a given percentage increase in frequency.

 

[Hitman928]

Correct. And if you look up what Microsoft has to say about Window's battery protection - https://support.microsoft.com/en-us...-windows-2ac1b4ba-6027-410a-b80e-f6767d867002 - it's entirely dependent on the manufacturer to implement. Hence both assumptions that it's enabled by default on all laptops and would result in the same 91% starting state of charge are incorrect.

I never said that all of the laptops tested were at 91% charge. I said it was unclear, which, unless we know if each laptop is using smart charging and how it's configured, it is still unclear. That's why I did both so the info is there both ways. It's possible some of the others were even less charged than the MSI laptop, I have no idea. If someone wants to spend the time tracking down each model's configuration, I'll gladly adjust appropriately. Otherwise, the data is there and the comparison will most likely fall between the MSI only being normalized from 91% to all of them being normalized from 91%. The only thing that really changes is how close the MSI gets to the top 3.

 

[H433x0n]

That curve is bs, it s just PL2 that is increased to boost the score.

In the new curve we can see that from 20W to 40W the score increase from 9500 to roughly 14500, that s 61% higher perf for 2x the power and that s just impossible by the very laws of semiconductors physics.

61% higher perf require 3x the power with a very good process that scale close to a square law wich is the theorical best case for mosfets, so this curve is just made up, anyone with some knowledge of electronics will spot such a discrepancy.

UltraBookReview updated pcode and got a result that's nearly the same.

The curve on the old pcode went from 7900 -> 13500, which is roughly 70% increase. If anything, the slope becomes less extreme with the updated pcode.