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.)

 

[controlflow]

How is refreshing every 15 secs even realistic as a light load?

His claim was literally "So it appear that when idling MTL is not that efficient". I think this benchmark pretty clearly refutes that. You can argue about whether or not this is a realistic benchmark but certainly it is a reasonable proxy for idle or very light load power efficiency.

 

[controlflow]

If MTL was as efficient than in this graph NBC would have measured better autonomy, even if only 30-40% that what they got.

Guess that to get those times they did set a convenenient corner case where only the LPE cores were active with a purely HTML page displayed if not a blank screen...

Anyway the idle values measured by NBC do not match those surprisingly favourable numbers, most likely that they are more or less made up..

Edit : Here the most favourable case i was talking about, and even in this exemple improvement in respect of RPL is 33%, not 90%, when that really happen because just a tiny more activity will wake up the biggest cores :

Intel Core Ultra alias Meteor Lake: Details und Specs

Heute startet endlich Intels neue CPU-Generation Meteor Lake in Form von elf U- und H-Modellen bei allen namhaften Partnern.

www.computerbase.de

How is only using LP-E cores a "convenenient corner case"? That's the entire point of those cores and the design of MTL.
The idea of the architecture is to turn off certain tiles when they are not being used. Several of the reviews have shown that there are issues with scheduling and turning off the CPU tile when they aren't needed. The big disparities in test results are more likely the result of immature FW and drivers rather than fabrication of data as you suggested.

The numbers from Hardware Canucks don't line up with what you want to see so "they are more or less made up"? This doesn't sound like a very objective take to me.

 

[Thunder 57]

How is only using LP-E cores a "convenenient corner case"? That's the entire point of those cores and the design of MTL.
The idea of the architecture is to turn off certain tiles when they are not being used. Several of the reviews have shown that there are issues with scheduling and turning off the CPU tile when they aren't needed. The big disparities in test results are more likely the result of immature FW and drivers rather than fabrication of data as you suggested.

The numbers from Hardware Canucks don't line up with what you want to see so they are "they are more or less made up"? This doesn't sound like a very objective take to me.

Try doing anything on an OS with two LP-E cores. Lunar Lake seems to be getting rid of them. Hardly seems like a victory for MTL. If it was as great as Intel claims we would've had plenty of reviews on the 14th rather than a soft launch. Have you seen Intel's own slide that indicates IPC regression on the P cores? Color me unimpressed. I do want to see some real reviews though.

 

[gdansk]

If it was as great as Intel claims we would've had plenty of reviews on the 14th rather than a soft launch

Soft launch is because Intel said Intel 4 in 2023. Great or not it had to ship.

 

[controlflow]

Try doing anything on an OS with two LP-E cores. Lunar Lake seems to be getting rid of them. Hardly seems like a victory for MTL. If it was as great as Intel claims we would've had plenty of reviews on the 14th rather than a soft launch. Have you seen Intel's own slide that indicates IPC regression on the P cores? Color me unimpressed. I do want to see some real reviews though.

Where did I say the LP E cores would be useful for doing real work? I specifically responded to the claim by the other user that MTL has high idle power usage. This is objectively not true, at least when the scheduling is working as intended.

I think MTL largely looks like a mixed bag. The RWC P core looks disappointing and is badly in need of a rework. The Crestmont E cores look like a decent improvement in both IPC and efficiency. The GPU and the steadily improving graphics drivers are a big win. The NPU looks largely unsupported by most SW for now and doesn't bring much to the end user yet. It will be at least a few more weeks and some number of updates until we have the full picture on MTL as a whole.

 

[controlflow]

The first 165H review?

Meteor Lake Core Ultra Review: Intel's All New Tile CPU Tested

Intel just launched its Meteor Lake Core Ultra mobile processors. Here's a benchmark deep dive, including AI and battery life.

hothardware.com

Battery life test here looks good:

"Yes, it can be that good. Running the PCMark 10 Battery Video Loop test, we scored the longest laptop battery life for any machine we have ever tested. In case you're bad at math, that is 20 hours and 1 minute. The test was still running as we were preparing to post this review. That is enough battery to watch the entire Lord of the Rings film trilogy on battery power...twice. And you want the kicker? The battery was only 91% charged when we started the test due to Windows' battery protection mechanisms."



However it has a 99.9wh battery and only an IPS LCD.

 

[Saylick]

The first 165H review?

Meteor Lake Core Ultra Review: Intel's All New Tile CPU Tested

Intel just launched its Meteor Lake Core Ultra mobile processors. Here's a benchmark deep dive, including AI and battery life.

hothardware.com

Battery life test here looks good:

"Yes, it can be that good. Running the PCMark 10 Battery Video Loop test, we scored the longest laptop battery life for any machine we have ever tested. In case you're bad at math, that is 20 hours and 1 minute. The test was still running as we were preparing to post this review. That is enough battery to watch the entire Lord of the Rings film trilogy on battery power...twice. And you want the kicker? The battery was only 91% charged when we started the test due to Windows' battery protection mechanisms."

View attachment 90744

However it has a 99.9wh battery and only an IPS LCD.

Nice!

They had a little more to say about the battery life:

We didn't test gaming battery life on this machine because it is not a gaming system, but you can bet that it will be considerably lower thanks to the SoC having to keep the GPU tile powered up. You see, part of the reason that the battery life for video playback is so long is that the Core Ultra CPU is able to power down both the graphics tile and the compute tile (with the CPU cores) while simply playing back a video like this. The LP E-cores and the video block both reside on the SoC tile, so most of the CPU is completely shut down during a video playback if nothing else is going on.

 

[Hitman928]

The first 165H review?

Meteor Lake Core Ultra Review: Intel's All New Tile CPU Tested

Intel just launched its Meteor Lake Core Ultra mobile processors. Here's a benchmark deep dive, including AI and battery life.

hothardware.com

Battery life test here looks good:

"Yes, it can be that good. Running the PCMark 10 Battery Video Loop test, we scored the longest laptop battery life for any machine we have ever tested. In case you're bad at math, that is 20 hours and 1 minute. The test was still running as we were preparing to post this review. That is enough battery to watch the entire Lord of the Rings film trilogy on battery power...twice. And you want the kicker? The battery was only 91% charged when we started the test due to Windows' battery protection mechanisms."

View attachment 90744

However it has a 99.9wh battery and only an IPS LCD.

Yeah, if you normalize for battery size, it’s not nearly as impressive and loses the top spot to at least the next two in line. Still a very solid showing but not anything that special compared to current competition.

 

[Thunder 57]

Soft launch is because Intel said Intel 4 in 2023. Great or not it had to ship.

Oh so like Cannon Lake had shipped by the end of the year in 2017 but it actually didn't? That was a turd sandwhich nobody wanted. Intel isn't doing to well on their "5 nodes in 4 years" promise.

Where did I say the LP E cores would be useful for doing real work? I specifically responded to the claim by the other user that MTL has high idle power usage. This is objectively not true, at least when the scheduling is working as intended.

I think MTL largely looks like a mixed bag. The RWC P core looks disappointing and is badly in need of a rework. The Crestmont E cores look like a decent improvement in both IPC and efficiency. The GPU and the steadily improving graphics drivers are a big win. The NPU looks largely unsupported by most SW for now and doesn't bring much to the end user yet. It will be at least a few more weeks and some number of updates until we have the full picture on MTL as a whole.

I've heard time and time again the CPU tile can turn off and the 2 LP-E cores on the SOC can do things and save buttloads of power. I'm not buying it. The GPU looks much improved but how can they turn off that tile and still have a usable laptop?

Let's face it, if Intel had a "Centrino Moment" they would be shouting from the rooftops. Intstead we get few reviews after the launch date. It's a joke. I'd explain why Intel has gone downhill but I would get a vacation for saying so. Rearview mirror, heh.

 

[SiliconFly]

Mobile wise, not sure. Depends on power

8+32 on mobile doesn't sound plausible as it may exceed the power envelope.

 

[controlflow]

I've heard time and time again the CPU tile can turn off and the 2 LP-E cores on the SOC can do things and save buttloads of power. I'm not buying it. The GPU looks much improved but how can they turn off that tile and still have a usable laptop?

You are misunderstanding this. Look at the graph again. They got about 20 hours of video playback using a 99.9wh battery charged to 91%. (~91wh).
During this playback, the CPU and GPU tiles would be largely turned off. How would it still be a usable laptop? Because the LP-E cores handle the light background processing needs and the display and video decode is handled by the SOC tile, NOT the GPU tile. The Meteor Lake SOC tile contains the display and media engine blocks. The GPU is actually not needed at all to decode video or even drive the display. Everything is handled by the SOC tile alone for the most part.

During this video play if a heavier background process comes up, the CPU tile may be switched back on from time to time but most of the time a light task like video playback wouldn't need anything beyond what is in the SOC tile (the LP-E cores and the display/media engines).

 

[SiliconFly]

You don't load new pages and just refresh the ones that are open and cached?

In general, or more specifically, in the real world, an average user doesn't load more than a few hundred pages a day.

So, running stress benchmarks that continuously loads multiple pages simultaneously does not reflect MTL real world power usage. The former is perfect but much closer.

 

[SiliconFly]

It's a mix of both in real life. Does the page being cached help Intel more than AMD? One wonders.

Bingo

 

[SiliconFly]

How is only using LP-E cores a "convenenient corner case"? That's the entire point of those cores and the design of MTL.
The idea of the architecture is to turn off certain tiles when they are not being used. Several of the reviews have shown that there are issues with scheduling and turning off the CPU tile when they aren't needed. The big disparities in test results are more likely the result of immature FW and drivers rather than fabrication of data as you suggested.

The numbers from Hardware Canucks don't line up with what you want to see so they are "they are more or less made up"? This doesn't sound like a very objective take to me.

You hit the mark.

Calling LP E cores as "convenenient corner case" is the most foolish explanation I've come across. MTL's design philosophy is focused squarely on those LP E cores like you mentioned.

 

[controlflow]

Yeah, if you normalize for battery size, it’s not nearly as impressive and loses the top spot to at least the next two in line. Still a very solid showing but not anything that special compared to current competition.

Yeah, they did say they started with 91% battery charge FWIW. 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.

 

[TESKATLIPOKA]

8+32 on mobile doesn't sound plausible as it may exceed the power envelope.

If they wanted, then they could do It, but It wouldn't make much sense.
It would need to have >100W PL1 or clocks would massively drop, in which case It's better to just use 8+16 with higher clocks, although at worse efficiency.

Here is an example of 13900HX 8P+16E.
Power Limit 2: 130W; Power Limit 1: 55W

From 4311pts down to 2649pts. -58% TDP results in -38.5% performance.
8P+8E i7-13700HX manages comparable performance at 70W, so saving 15W by using 2 extra E-clusters isn't worth It. Link

I think It would be similar with ARL 8P+8E vs 8P+16E vs 8P+32E. The last one would be the most efficient, but to be also significantly more performant It would need a very high power limit, which is achievable only in the largest laptops.

 

[mikk]

From Raichu tile size measurement:

CPU tile: 8.72 mm x 7.99 mm = 69.67 mm2
IO tile: 9.20 mm x 2.98 mm = 27.42 mm2
SOC tile: 10.85 mm x 9.23 mm = 100.15 mm2
iGPU tile: 10.22 mm x 4.33 mm=44.25 mm2

Close to what Andreas Schilling said, CPU and Soc tile a little bigger from Raichu.

 

[João Bortolace]

Die shot of Soc and GPU title

https://twitter.com/x/status/1737443039034134788

 

[Hitman928]

In general, or more specifically, in the real world, an average user doesn't load more than a few hundred pages a day.

So, running stress benchmarks that continuously loads multiple pages simultaneously does not reflect MTL real world power usage. The former is perfect but much closer.

I guess people can judge by their own use cases, but I highly doubt most people's web browsing experience is closer to hitting ctrl+r on the same webpage every 15 seconds and doing literally nothing else versus the more typical web browsing battery test which runs a script loading a loop of web pages (not simultaneously) with some scrolling and link clicking on each page spaced a reasonable amount of time apart, which may also include watching an embedded video.

For example:

We turn the Wi-Fi on and Bluetooth off and run an automatic script that simulates viewing and scrolling through various web pages. These include RTINGS.com, Wikipedia, and NPR, as well as an embedded 6-minute YouTube video that takes up the entire browser window. We run the test until the laptop dies or goes into emergency shutoff mode. Once we power the laptop back on, we record the elapsed time until shutoff that was recorded by the script.
This test affects the total battery score the most because we expect it to reflect most people's workloads. Alongside general web browsing, we also expect this result to closely reflect workloads that include using a single application at a time to run light productivity tasks. These include word processing and spreadsheets, as well as most idle tasks like reading articles.

or here:

Wi-Fi mode: the possible battery life while surfing the Internet via Wi-Fi with medium brightness (~150 cd/m²) and power-saving options ("balanced" mode) switched on. We measure the runtime by letting the device run an automatic script (HTML 5, JavaScript, no Flash - update 03.05.2015 v1.3), which picks a mix of websites and switches between them every 30 seconds.

What Hardware Canucks tests as their "web browsing" battery life test is what Notebookcheck would call their reader/idle test which I think is a much more accurate label.

 

[FlameTail]

Die shot of Soc and GPU title

https://twitter.com/x/status/1737443039034134788

View attachment 90753View attachment 90754

Now label it!

 

[Hitman928]

I also just watched the Geekerwan video, here is what he says about the LPE cores:

The two ultra-low power small cores. . . will not be used in daily applications. They can only be used in standby idle state, when watching videos locally [not streaming], or in the software specially programmed for them. . .

I think Intel should really allow more applications for [the] ultra-small cores. After all, what kind of performance do I need to type in Word? If you don't let me use the ultra-small core at this time, when should I use it?

So the only time the LPE tile advantage can come into play is when not interacting with the laptop at all. Even just browsing the web, typing in
Word, or checking email will keep the compute tile powered up. This seems to be confirmed in the battery life tests we have so far and falls in line with my expectations. The only realistic scenario you should see some significant benefit is from locally saved video playback, but even that doesn't seem to be a working scenario for power savings according to the battery tests we have.

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.

 

[Hitman928]

Yeah, they did say they started with 91% battery charge FWIW. 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.

They say that the Windows battery protection won’t let them charge to more than 91% so it’s unclear to me whether the other laptops are also at 91% or 100%. However, even if you assume the MTL laptop was at 91% to start and the others were at 100%, it still loses to the next two in line once you normalize for battery life and adjust for the lower starting charge, it’s just in a closer third place.

 

[Hulk]

The LPE core development is going to be very much like the adoption of the E cores when Alder Lake first hit shelves. It'll be hit and miss more often than not in the beginning and but will get better with time.

For all the hate Intel is getting they have been bold with the adoption of the hybrid approach on the desktop (they bet the farm on it) and now with the tiles and the LPE on the SoC. I know many people will claim this is a "fail" and a "joke" but that's some time of strange cheerleading for a team. I don't understand it actually. Silicon designed by thousands of people with billions of transistors and the end result is "fail" and/or a "joke" yet the CPU seems to be working?

On the other side AMD has been bold as well. Bringing 8, 12, and 16 core parts to market using chiplets and basically blowing the lid off the multicore processor market and then quickly following up with Zen 2, Zen 3, Zen 4, and now Zen 5 just around the corner. It's an amazing series of releases.

Let's try and stay on the tech

 

[Hitman928]

The LPE core development is going to be very much like the adoption of the E cores when Alder Lake first hit shelves. It'll be hit and miss more often than not in the beginning and but will get better with time.

For all the hate Intel is getting they have been bold with the adoption of the hybrid approach on the desktop (they bet the farm on it) and now with the tiles and the LPE on the SoC. I know many people will claim this is a "fail" and a "joke" but that's some time of strange cheerleading for a team. I don't understand it actually. Silicon designed by thousands of people with billions of transistors and the end result is "fail" and/or a "joke" yet the CPU seems to be working?

On the other side AMD has been bold as well. Bringing 8, 12, and 16 core parts to market using chiplets and basically blowing the lid off the multicore processor market and then quickly following up with Zen 2, Zen 3, Zen 4, and now Zen 5 just around the corner. It's an amazing series of releases.

Let's try and stay on the tech

Problem is, Intel is rumored to be abandoning the LPE cores after ARL. If true, I don’t know how much improvement we’ll see with them if there isn’t a long term plan to use them. Only thing we can do is wait and see I guess.

 

[FlameTail]

Problem is, Intel is rumored to be abandoning the LPE cores after ARL. If true, I don’t know how much improvement we’ll see with them if there isn’t a long term plan to use them. Only thing we can do is wait and see I guess.

And what will they do after abandoning LPE?