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.)

 

[coercitiv]

That depends on workload and frequency(TDP).

Sure it depends, but most consumer workloads do not have excellent thread count scaling and they run at high clocks. E cores shine in specialized workloads or as companion cores, taking care of background tasks at low clocks. On the opposite side of the spectrum, my NAS definitely likes having 4 E cores over 1 P core, that's for sure.

 

[TESKATLIPOKA]

Apparently DLVR is enabled on mobile Raptor Lake but not desktop.

Ok, It's possible.
Turbo boost saw 400MHz for P-core and 300MHz for E-core uplift, but base stayed more or less the same for 15-28W models, at most higher by 100MHz.

 

[HoveringStyle]

P-series was axed? It's more like It was renamed than axed. That TDP is still used with MTL chips.

I think axed is appropriate, that TDP was originally used in U-series chips, P-series was an attempt at having a cut down H-series die replace them but that particular P+E configuration didn't end up hitting any performance/watt sweet spots against either 2+8 or 6+8. MTL H is 28W yes, but there no 4+8 "7"s anymore, they use the full H-series die.

 

[TESKATLIPOKA]

I think axed is appropriate, that TDP was originally used in U-series chips, P-series was an attempt at having a cut down H-series die replace them but that particular P+E configuration didn't end up hitting any performance/watt sweet spots against either 2+8 or 6+8. MTL H is 28W yes, but there no 4+8 "7"s anymore, they use the full H-series die.

P-series use 28W as default(base).
U-series use 15W as default, Alder Lake also had some 9W models.

2P+8E was used only for the U-series.
4-6P+8E was used for both P and H series.
Not sure if HX series with less than 8P-cores use the mobile chip or a cutdown desktop chip, maybe both.
The above-mentioned things are true for both Alder and Raptor Lake.

You can be sure, that there will be cutdown models with 4P+8E.
MTL-U already has such a model named Ultra 7 155U.

 

[gdansk]

Is it fair to say they are 15W parts when they'll spend 2 minutes at 55W?
I don't think it was always that bad. A few minutes at 30W could be fine.

Maybe that is something MTL can supply: more reasonable power limits.

 

[HoveringStyle]

P-series use 28W as default(base).
U-series use 15W as default, Alder Lake also had some 9W models.

2P+8E was used only for the U-series.
4-6P+8E was used for both P and H series.
Not sure if HX series with less than 8P-cores use the mobile chip or a cutdown desktop chip, maybe both.
The above-mentioned things are true for both Alder and Raptor Lake.

You can be sure, that there will be cutdown models with 4P+8E.
MTL-U already has such a model named Ultra 7 155U.

P-series also used the 2+8 die, for the i3, but the point I'm trying to make is that there is no longer any 4+8 chip at 28W being sold as "top of the line", they will be Ultra 5 and that might be explained by the 6+8 configuration just being more performant than 4+8 at practically every relevant power limit, from 15W to 28W and beyond,

The Ultra 7 155U exists but it targets 15W, it's the U-series equivalent to the rare i7-1280P/1370P (28W but 6+8).

 

[tamz_msc]

Is it fair to say they are 15W parts when they'll spend 2 minutes at 55W?
I don't think it was always that bad. A few minutes at 30W could be fine.

Maybe that is something MTL can supply: more reasonable power limits.

PL2 being 55 watts doesn't affect the average power consumption much, if at all, when using a laptop normally.

Race to idle means that compared to old days when the CPU would consume dozens, if not more, watts, modern CPUs are designed to operate in bursts.

It is the reason why I strongly dislike Cinebench when used as a metric for efficiency on laptops in particular.

 

[gdansk]

PL2 being 55 watts doesn't affect the average power consumption much, if at all, when using a laptop normally.

Race to idle means that compared to old days when the CPU would consume dozens, if not more, watts, modern CPUs are designed to operate in bursts.

It is the reason why I strongly dislike Cinebench when used as a metric for efficiency on laptops in particular.

Totally disagree. It's part of the reason many, many laptops have intolerable fan noise. Sure it turns down after 2 minutes but the 30W cooling solution is still saturated. I set an actual 15W TDP and it stays quiet.

But with the default power limits? I browse a poorly written site in Chrome and the fans go wild.
This is a problem Apple has worked around by not having ridiculous power limits for the CPU which allows them to safely delay turning on fans until absolutely necessary.

 

[FlameTail]

Totally disagree. It's part of the reason many, many laptops have intolerable fan noise.
You browse a poorly written site in Chrome? Vroom go the fans.
A problem Apple has worked around by not having ridiculous power limits for the CPU and delayed fans.

Yeah. This is why I am not a fan of some people supporting the idea of the Snapdragon X Elite boosting past it's efficiency range.

Doing that, any thermal/efficiency benefits they gain will simply go out of the window.

Intel and AMD rely roo much on boost clocks for performance, whereas Apple Silicon delivers the max performance at base.

 

[tamz_msc]

Totally disagree. It's part of the reason many, many laptops have intolerable fan noise. Sure it turns down after 2 minutes but the 30W cooling solution is still saturated. I set an actual 15W TDP and it stays quiet.

But with the default power limits? I browse a poorly written site in Chrome and the fans go wild.
This is a problem Apple has worked around by not having ridiculous power limits for the CPU which allows them to safely delay turning on fans until absolutely necessary.

I have the data to back what I'm saying. Fans spinning up is tangentially related to this issue with power consumption.

 

[tamz_msc]

So this is why Apple is good at what it does.

Addendum: Clock Ramp on ADL, Zen 4, M1, and More

We recently released an article on how quickly CPUs increased their clocks from idle, and received criticism in that it didn’t include CPUs newer than Kaby Lake. That was because I just didn&…

chipsandcheese.com

A completely different approach where clock ramp to fmax is 20x slower compared to modern x86 CPUs.

That's the secret to power efficiency and low noise.

Apple marketing will not let you know this on their own.

 

[Tup3x]

Yeah. This is why I am not a fan of some people supporting the idea of the Snapdragon X Elite boosting past it's efficiency range.

Doing that, any thermal/efficiency benefits they gain will simply go out of the window.

Intel and AMD rely roo much on boost clocks for performance, whereas Apple Silicon delivers the max performance at base.

You missed the point. It can scale in different form factors where better cooling is possible. It doesn't mean that thin laptop will suddenly burst to 80w for few seconds.

 

[PJVol]

So this is why Apple is good at what it does.

Addendum: Clock Ramp on ADL, Zen 4, M1, and More

We recently released an article on how quickly CPUs increased their clocks from idle, and received criticism in that it didn’t include CPUs newer than Kaby Lake. That was because I just didn&…

chipsandcheese.com

A completely different approach where clock ramp to fmax is 20x slower compared to modern x86 CPUs.

Where have you seen an approach that could be considered somewhat good? I see the article shows quite the opposite, where M1 max performs really bad in the testing scenario.

 

[tamz_msc]

Where have you seen an approach that could be considered somewhat good? I see the article shows quite the opposite, where M1 max performs really bad in the testing scenario.

Apple silicon doesn't ramp as fast because their wider cores are designed to get more work done at lower clock speeds, thereby saving power.

 

[TESKATLIPOKA]

P-series also used the 2+8 die, for the i3, but the point I'm trying to make is that there is no longer any 4+8 chip at 28W being sold as "top of the line", they will be Ultra 5 and that might be explained by the 6+8 configuration just being more performant than 4+8 at practically every relevant power limit, from 15W to 28W and beyond,

I missed that one model, but It was only Alder Lake, Raptor didn't have such a model.

To me, It doesn't matter that 4P+8E will be only Ultra 5 at 28W for example.
Not sure how It will look like with MTL, but with Alder Lake only at 28W and above was the 6P+8E model more performant and at 20w both of them performed as 2P+8E.

The Ultra 7 155U exists but it targets 15W, it's the U-series equivalent to the rare i7-1280P/1370P (28W but 6+8).

You think Intel won't release this configuration for 28W?
I think that will depend on how many faulty chips they will have, or If this single model is enough to sell all of them.

I hope NBC will compare MTL with different core configuration at different TDPs, for me It will be very interesting comparison.

 

[TESKATLIPOKA]

Apple silicon doesn't ramp as fast because their wider cores are designed to get more work done at lower clock speeds, thereby saving power.

It has nothing or very little to do with how performant the core is.
Apple's M1 ramps from 1Ghz to 3.2GHz after 100ms.
AMD's 6650U ramps from 1.4GHz to 4.5GHz after 1ms.

6650U is already running at 4.5GHz, while M1 is still at 1GHz.
You can't say that It is because M1 provides comparable performance to 6650U at only 1GHz, because that's not true.

The reason why M1 ramp up so slowly is that Its main focus is efficiency and saving power, after all It's a mobile chip.
But for AMD or Intel this is not the priority, but responsiveness, that's why they ramp up so fast.

It will be interesting to see If MTL will improve responsiveness or actually worsen It to save power.

 

[FlameTail]

MLID doesn't have good news about MTL in his latest video.

 

[TESKATLIPOKA]

View attachment 89998View attachment 89999

MLID doesn't have good news about MTL in his latest video.

Bad news where? I don't care about TOPs and that IGP performance is good.

 

[FlameTail]

Bad news where? I don't care about TOPs and that IGP performance is good.

Yeah the TOPS is the bad news.

I was expecting MTL to have 45+ TOPS NPU.

I am surprised it's not the case.

 

[FlameTail]

So Qualcomm has the leadership in AI performance at 45 TOPS.

Although not for long, as Strix Point will debut in later 2024.

 

[TESKATLIPOKA]

Yeah the TOPS is the bad news.

I was expecting MTL to have 45+ TOPS NPU.

I am surprised it's not the case.

And those TOPs will help me in what way exactly?
Will It provide higher FPS?
Will It make my system faster?

 

[mikk]

Hawk Point is the same silicon as Phoenix, it's just a NPU clock speed bump up. Someone should test the efficiency and see how it changed, performance isn't everything. Afaik MTL NPU is clocked at a modest 1.6 Ghz which results in ~13 TOPs. To be fair maybe this refers to 185H and 10 TOPs is valid for lower models, we don't know yet. Funny how he proudly mentioned Hawk Point is shipping whereas he didn't say the same about MTL. AMD can paper announce something and he is hyping it up with already shipping.

 

[DrMrLordX]

AMD can paper announce something and he is hyping it up with already shipping.

Given that there's no major changes in silicon, it's not likely to be a paper launch. They're already producing Phoenix.

 

[Glo.]

And those TOPs will help me in what way exactly?
Will It provide higher FPS?
Will It make my system faster?

If you will ask your system to order a cheese pizza, system with 50 TOPs will order it 0.00005s faster than the one with 45 TOPs. Assuming you have enough RAM, obviously.

 

[Thibsie]

And those TOPs will help me in what way exactly?
Will It provide higher FPS?
Will It make my system faster?

It will recognize faces from your photos a tad little faster.
Not happy ? My god....