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	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 (20A)	Arrow Lake (N3B)	Lunar Lake	Panther Lake
Platform	Mobile H/U Only	Desktop Only	Desktop & Mobile H&HX	Mobile U Only	Mobile H
Process Node	Intel 4	Intel 20A	TSMC N3B	TSMC N3B	Intel 18A
Date	Q4 2023	Q1 2025 ?	Desktop-Q4-2024 H&HX-Q1-2025	Q4 2024	Q1 2026 ?
Full Die	6P + 8P	6P + 8E ?	8P + 16E	4P + 4E	4P + 8E
LLC	24 MB	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.)

 

[dullard]

??!? Think about what you are saying. You can't get 10 hours on a 70WHr battery with 30W power average. It's 7W average.

30 W when not in idle.

 

[DavidC1]

30 W when not in idle.

I get what you are saying, but using numbers like that doesn't make you sound very knowledgeable about this.

Battery life on light load is dominated by idle power, and the ability of the platform to reach those idle states. Very little to do with TDP.

Apple got this formula down long time ago. Intel made lot of BS about "this power savings" and "that power savings" with no real result(except few like with Pentium M and Haswell). I've read countless claims like oh this new transistor offers 0.05V reduction or how it can power down between keystrokes and reviews would show pretty much same as the predecessor. Intel talks too much. They should shut up and just deliver.

Apple figured out that you could have a core as big and wide as they'd like(simplifying things here), but all they needed to do is have really really low platform idle power, and the CPU to be able to reach it.

Dell's claims are basically almost complete idle. Look back on that article you linked.

an astonishing 19 hours and 41 minutes based on our web surfing results with its screen set to 150 nits of screen brightness, though Dell promised it could get up to 27 hours when playing 1080p video.

So according to their own tests, 27 = 19.7, so why would they believe the new results? If you take the same ratio then Lunarlake would get 22.

 

[dullard]

So, please help me out. I'll try to boil your argument down:

Apple got this formula down long time ago.

All garbage.

Then, Intel beats Apple in battery life:

 

[DavidC1]

So, please help me out. I'll try to boil your argument down:


Then, Intel beats Apple in battery life:
View attachment 106902

The same test shows MBA having not a big advantage over Meteorlake. I'd wait until more tests.

Yes and Lunarlake finally got them to use the same method to get to their levels. But you know, that's only going to last one generation, because they'll give that up on Pantherlake.

If the rumors are saying 35% reduction in SoC power, I don't see how they'll achieve that over Lunarlake. It's probably over Meteorlake.

The garbage statement is about people saying 10 hours is some fancy never-before seen figure when we got that with a 50WHr battery back with Broadwell in 2015. For battery life test, it really sucks. And M3/Snapdragon sucking is additional proof.

 

[Khato]

Yes and Lunarlake finally got them to use the same method to get to their levels. But you know, that's only going to last one generation, because they'll give that up on Pantherlake.

What aspects of PTL are going to make it vastly inferior to LNL with respect to power consumption exactly? They aren't continuing the on package memory which will indeed be unfortunate, but beyond that?

 

[OriAr]

Oof if these are the yields for a 100mm2 chip on Intel 3

https://twitter.com/x/status/1831598153692545468

https://twitter.com/x/status/1831621821894160887

I am gonna call BS on that as there is no way Intel would have ramped up SRF and GNR if yields were that bad.
Raichu already said d0 for Intel 3 is somewhere in the 0.1-0.15 range.

 

[SiliconFly]

... Haswell... ... Pentium M ...

You're missing the big picture. When you're comparing LNL with older Intel CPUs, you're comparing just the CPU power draw alone. Whereas, the rest in general are talking about the power efficiency & battery life of the entire laptop I think.

Lunar Lake CPU alone can't bring in that much efficiency. It's the bundled LPDDR & other tightly integrated system components (not just in the SoC but also possibly in the motherboard too) & the accelerators & the system drivers too might have power efficiency specific optimizations that overall help the Lunar Lake system achieve higher efficiency.

So, comparing CPU states/wattage alone to arrive at battery life is incorrect.

 

[cannedlake240]

but beyond that?

Splits off the GPU tile(and likely media engine with it), so that's going to have some power D2D penalty. There's also a rumor that it reverts back from LNL style PMICs. Maybe a more efficient process node and LP-E core could make up for that, idk... Doesn't help that Intel nodes are typically worse than TSMC at this

 

[digitaldreamer]

30 W when not in idle.
View attachment 106899

Correct me if I'm wrong: Wouldn't a 30W draw only last 2.3 hours on a 70Wh battery?

 

[SiliconFly]

What aspects of PTL are going to make it vastly inferior to LNL with respect to power consumption exactly? They aren't continuing the on package memory which will indeed be unfortunate, but beyond that?

It's come to my understanding that DDR is very power inefficient. Has high power draw when active and even on idle and sleep. On the other hand, LPDDR it seems uses way less power during idle and incredibly less power during sleep.

Many here including me said before that our Intel laptops have battery drain even during sleep which is true cos DDR remains active and uses way too much power during sleep (one of the key reasons I always shutdown my laptop to reduce battery cycle and extend overall battery life).

The new Lunar Lake laptop use LPDDR and hence when they sleep their power usage is so negligible it appears to us as thought the system doesn't use any power at all during sleep.

PTL on the other hand (I'm assuming) ditches LPDDR for standard DDR I think which drastically reduces power efficiency. And thats not a good thing imho.

(I'm not much interested in power usage and hence don't follow up on it much. Some of my assumptions here might be wrong. So sorry if I'm.)

Correct me if I'm wrong: Wouldn't a 30W draw only last 2.3 hours on a 70Wh battery?

Yep. A sustained 30W power draw will do exactly that. But you're still counting only the CPU power draw. Under full load, power draw of other components should also come into play.

 

[poke01]

There is no way lunar is at 30 watts during the web browser test

 

[DavidC1]

What aspects of PTL are going to make it vastly inferior to LNL with respect to power consumption exactly? They aren't continuing the on package memory which will indeed be unfortunate, but beyond that?

They can go On Package memory with PTL-U if they want, but that won't do anything.

Lunarlake has a purpose optimized LPDDR PHY for lower power when in previous generations it just had it for supporting it. There's always potential, but it needs to be released/used before it can be kinetic.

Lunarlake also has an SLC cache because proximity determines power efficiency. Lunarlake also doesn't need to deal with three types of cores and majority of workloads will run on the E core, which is actually efficient.

GPU tile is separate on Pantherlake, which will increase both interconnect power use(won't be significant but still) and latency. Latency is the key to both performance and low power. IO on a separate tile doesn't matter as much because the bandwidth demands are low.

GPU on the other hand is going to have to go through the interconnect to communicate both with the CPU and with the memory controller. A bus is the last thing that can be powered off, because it's like a road. If you have just one car travelling, you need it. So now you have a high demand device on a slower, higher power lane.

Ultimately the problem is ideology, which is that Intel is run by beancounters and they don't view needing a separate purpose optimized SoC for low power.

Let me put it this way. If Pantherlake is well executed it can be an improvement over Meteorlake. But it could have been the best if they put all Lunarlake had. Let's quote Bionic_Squash again. Lunarlake does not have a direct successor.

The new Lunar Lake laptop use LPDDR and hence when they sleep their power usage is so negligible it appears to us as thought the system doesn't use any power at all during sleep.

What? No. Every laptop since Sandy Bridge or something uses LPDDR of some sort. But that's the thing. It's just for support. Lunarlake has a purpose optimized LPDDR interface, it's not just "Oh you can stick LPDDR now". So you get that benefit plus whatever little LPDDR itself offers.

 

[cannedlake240]

problem is ideology

They want be an AI chip company now. Apparently that's what they are investing in, next gen GPU ip and stuff. No cash left for the low power niche

 

[DavidC1]

Lunar Lake CPU alone can't bring in that much efficiency. It's the bundled LPDDR & other tightly integrated system components (not just in the SoC but also possibly in the motherboard too) & the accelerators & the system drivers too might have power efficiency specific optimizations that overall help the Lunar Lake system achieve higher efficiency.

Please people, do a little bit of research, read a bit of history before you say something.

They have been doing what you've been saying since at least Haswell. I can tell you that my chip can go to 0.5W on HWInfo with the screen-on. It won't happen if I have a mouse connected. It won't happen if sound is on. I bought a Haswell device just for the fun of optimizing it. It wouldn't reach below 1.5W for CPU power. I found out it's due to the crappy Qualcomm WiFi. I switched it out for an Intel WiFi and it dropped to 0.9W.

The CPU since Haswell is like a driver for a car. It has a leash to all the other components, and if one component misbehaves, the CPU itself can't power down. That's what the EVO certification is about too. A reduction in CPU power is more than just CPU. It's almost everything, even a little bit of screen.*

From the Lunarlake GPU presentation:
-Display buffering for race to halt on memory access
-Skip fetching and transmitting repeated frames
-Skip waking cores on repeated frames
-Queueing frame into display to reduce core wakeups
-Early transport

The above is the summary of just display-related power savings in Lunarlake.

 

[dullard]

Correct me if I'm wrong: Wouldn't a 30W draw only last 2.3 hours on a 70Wh battery?

30 W when under load. Much lower (but unreported) in idle. The load power could be set as low as 8 W for the CPU. I haven't seen how low it idles yet.

Since you don't know the idle time nor the idle power, you can't just take 70 Whr / 30 W and get 2.3 hr. Doing so would be setting it at max CPU power constantly. Very few laptop use cases do that.

That is why @DavidC1 is completely mixed up in this thread. He is so against Lunar Lake that he can't think clearly right now. Simple math like that isn't sufficient. Plus, there is his laughable idea that a 4 P + 4 E Panther Lake-U is not a successor for a 4 P + 4 E Lunar Lake-U. He thinks Lunar Lake is a one-off. Let alone his idea that his 2017 Yoga laptop is comparable to a 2024 Lunar Lake laptop. The Lunar Lake laptop would run circles around his 2017 Yoga. Low power consumption gives great battery life, but we actually need performance too. Heck, I could build a laptop, put a useless CPU in it with low idle power and show massive battery life--but if it takes hours to do anything, no one would want it.

But, suppose a random laptop with a random CPU does work at 30 W and then say idles down to 5 W. Suppose it gets the work done quickly enough that it spends 95% of the time in idle mode. Then, that CPU would average 30 W * 5% + 5 W * 95% = 6.25 W. With a 70 Whr battery, it would get (70 Whr) / (6.25 W) = 11.2 hours of run time at that 30 W load power setting.

 

[OriAr]

Lunar Lake power draws while doing various tasks according to Jaykihn:

https://twitter.com/x/status/1827630742765191254

 

[cannedlake240]

Panther Lake-U is not a successor for a 4 P + 4 E Lunar Lake-U. He thinks Lunar Lake is a one-off.

This is true though. PTL-U is not a direct successor to LNL the way M2 is to M1. LNL is Intel's attempt at an M1, ARM soc competitor. If the latter were the case Intel would be doubling down on driving maximum efficiency at all levels. PTL however is more of a regular Intel mobile SoC iteration that integrates only some of the improvements found in LNL.
Plus it's targeting the more mainstream -U segment with 4Xe3 iGPU, while LNL is a premium tier design using the best igpu Intel has to offer.

 

[SiliconFly]

Every laptop since Sandy Bridge or something uses LPDDR of some sort ...

I don't think so. None of the regular Intel laptops like i3 or i5 or i7 use LPDDR. At least until last year. I had dozens of laptop RAM sticks lying around (leftovers from years of upgrading my old laptops) and all of them were regular laptop form factor DDR sticks. Mostly DDR4. My current i9 laptop uses DDR5.

 

[dullard]

This is true though. PTL-U is not a direct successor to LNL the way M2 is to M1. LNL is Intel's attempt at an M1, ARM soc competitor. If the latter were the case Intel would be doubling down on driving maximum efficiency at all levels. PTL however is more of a regular Intel mobile SoC iteration that integrates only some of the improvements found in LNL.
Plus it's targeting the more mainstream -U segment with 4Xe3 iGPU, while LNL is a premium tier design using the best igpu Intel has to offer.

How would you predict these two chips will compare as a whole?

1. Lunar Lake: 17 W, 4 P, 4E, 8 Xe2 cores (60 TOPS on the iGPU)
2. Panther Lake: 15 W, 4P, 4E, 4 Xe3 cores (double the iGPU performance of Lunar Lake: 120 TOPS on the GPU). Supposedly 35% more energy efficient than Lunar Lake. http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...akes-discussion-threads.2606448/post-41189786

Panther Lake-U: up to four high-performance cores, up to four ultra-low-power cores, and up to four Xe graphics clusters. These CPUs will have a processor base power of 15W, which is in-line with contemporary thin-and-light notebook processors.

https://www.tomshardware.com/pc-com...e-configurations-for-laptops-after-lunar-lake

 

[SiliconFly]

... Plus, there is his laughable idea that a 4 P + 4 E Panther Lake-U is not a successor for a 4 P + 4 E Lunar Lake-U. He thinks Lunar Lake is a one-off. ...

Even I think PTL-U might be very close to LNL if the OEMs pair it with proper LPDDR.

Let alone his idea that his 2017 Yoga laptop is comparable to a 2024 Lunar Lake laptop. The Lunar Lake laptop would run circles around his 2017 Yoga. Low power consumption gives great battery life, but we actually need performance too. Heck, I could build a laptop, put a useless CPU in it with low idle power and show massive battery life--but if it takes hours to do anything, no one would want it.

True.

MTL ... 8 Xe2 cores (60 TOPS on the iGPU) ...
PTL ... 4 Xe3 cores (double the iGPU performance of Lunar Lake ...

Assuming PTL's GPU is as strong as LNL, then we can say with very high confidence that PTL-U will be LNL's perfect replacement.

... Lunarlake has a purpose optimized LPDDR PHY for lower power ...

In theory PTL-U can have a simple dual PHY that supports both DDR & LPDDR as required.

 

[mikk]

How would you predict these two chips will compare as a whole?

1. Lunar Lake: 17 W, 4 P, 4E, 8 Xe2 cores (60 TOPS on the iGPU)
2. Panther Lake: 15 W, 4P, 4E, 4 Xe3 cores (double the iGPU performance of Lunar Lake: 120 TOPS on the GPU). Supposedly 35% more energy efficient than Lunar Lake. http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...akes-discussion-threads.2606448/post-41189786

https://www.tomshardware.com/pc-com...e-configurations-for-laptops-after-lunar-lake

4 Xe3 2x faster than 8 Xe2? I would tone down your expectations lol. 120 TOPS certainly comes from the bigger 12 Xe3.


PTL-U is more a traditional U chip for cheaper devices. There is no on package RAM and it has a smaller iGPU. From the Dell roadmap next after LNL-MX with 20W TDP comes NVL, maybe LNL-MX gets a real successor with a chip from the Nova Lake generation. However PTL also should get a big battery life increase (over MTL) just from the 4xLPE Darkmont in the compute tile, this is the big thing on Lunar Lake. Also the refreshed Cove+Mont cores might be even more opimized for efficiency.

 

[Khato]

Thanks for the opinions regarding why PTL isn't going to measure up to LNL on power efficiency. No question that the integrated graphics being subject to a D2D interface isn't ideal from an efficiency perspective. Will be interesting to see how opinions shift as correct implementation details make their way out.

 

[SiliconFly]

Just came across this LPDDR info...

The Gist:
- LPDDR is more expensive
- LPDDR performance is very close to DDR
- LPDDR is hyper power efficient
- LPDDR uses very very less power during idle/sleep
- LPDDR needs to be soldered
- LPDDR is usually not found in regular Intel laptops

"LPDDR SDRAM is designed to be able to scale down the refresh rate of the memory to a very low clock rate when idle, allowing for extremely low power draw while any part of the RAM is not actively being written to or read from (but where the RAM still retains all information, since it is not turned off entirely), while standard DDR SDRAM cannot do this.

This comes with a number of tradeoffs; it requires stricter bin tolerances and is therefore more expensive to do, this type of downclocking doesn't play well with socketed RAM form-factors, so all LPDDR SDRAM must be soldered, and lastly, LPDDR SDRAM designs doesn't seem to scale as well for extremely high-density memory (think 128GB DIMMs for servers and the like).

You can work around these restrictions to some extent. The first can simply be overcome if you're willing to throw money at the problem. The second may not be an issue depending on your target customer or device, especially since you can still offer factory customizability. For the third, it's worth noting that soldered RAM is very space-efficient (this applies for both DDR and LPDDR), so it's actually a viable strategy to just throw more LPDDR SDRAM at the chip if you're willing to make the memory bus wide enough. You'd be paying a premium for both the expensive memory and the expensive memory bus though, so it really only makes sense to do so if you need low idle power draw.

It's also worth noting that LPDDR SDRAM has half the channel bandwidth of DDR SDRAM, but this does not matter much in practice, since all laptop designs that use LPDDR SDRAM feature a memory bus with at least twice as many channels to compensate.

Other than that, there's no performance penalty worth noting for LPDDR RAM. LPDDR RAM has gotten a bad reputation over the last few years especially, since Intel tied LPDDR4 support to a new 10nm memory controller which they weren't able to ship because 10nm was a disaster, and AMD didn't bother supporting LPDDR4 before Zen 2, so the PC laptop market was stuck on using either LPDDR3 or DDR4 for four years after the DDR3 to DDR4 transition."

Also, if PTL-U defaults to standard DDR, then it won't be a true LNL successor.

 

[majord]

Lunar goodness

the run times he's getting on all these laptops for netflix are so bad for such huge batteries I don't know what to make of it..

What is his testing conditions? max brightness+ max performance mode?!

 

[poke01]

the run times he's getting on all these laptops for netflix are so bad for such huge batteries I don't know what to make of it..

What is his testing conditions? max brightness+ max performance mode?!

I would wait for third party reviews as this is a Lenovo+Intel sponser review but it’s looking good.

Dave 2D often doesn’t tell the whole story ie how’s the battery under load and how is sleep etc