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.)

 

[Abwx]

Why do you say you don’t know the number? I’m asking you to use your method to calculate the efficiency difference.

For this some numbers are required, FI a Cinebench R23 MT score and reliable power measurements, NBC did a comparison, you can check the numbers if you can find the link and normalize them according to the score difference.

If a CPU does 1.25x a better score at same perf it will use 0.64x the power to match the comparison, so its perf/watt is actually 56% better than the comparison.

If one prefer pure perf rather than effciency he can boost the CPU at the same power than the comparison, but in this case this is a speed comparison at a given power and here perf/watt become a second order metric.

 

[Hitman928]

For this some numbers are required, FI a Cinebench R23 MT score and reliable power measurements, NBC did a comparison, you can check the numbers if you can find the link and normalize them according to the score difference.

If a CPU does 1.25x a better score it will use 0.64x the power to match the comparison, so its perf/watt is actually 56% better than the comparison.

If one prefer pure perf rather than effciency he can boost the CPU at the same power than the comparison, but in this case this is a speed comparison at a given power and here perf/watt become a second order metric.

I'm not interested in Cinebench, that changes the work load. The work load is Spec_int, so how much more efficient is the M1 in this work load?

 

[Abwx]

I'm not interested in Cinebench, that changes the work load. The work load is Spec_int, so how much more efficient is the M1 in this work load?

There s surely numbers, so you can find them if you re that interested in the subject.

Beside i was comparing the 7840U to the 155H, so if you re bringing another CPU in the comparison you it would be a minimum to provide numbers rather than asking questions about it, dunno why you re asking me to do your homework...

 

[Hitman928]

There s surely numbers, so you can find them if you re that interested in the subject.

Beside i was comparing the 7840U to the 155H, so if you re bringing another CPU in the comparison you it would be a minimum to provide numbers rather than asking questions about it, dunno why you re asking me to do your homework...

M1 is in the same graph you are using to compare the 7840u and 155h. . .

 

[Abwx]

M1 is in the same graph you are using to compare the 7840u and 155h. . .

So you could do the math but still asking me to do it for you..?.

As i said, asking the others to do your homework...

 

[Hitman928]

So you could do the math but still asking me to do it for you..?.

As i said, asking the others to do your homework...

I'm asking you because if I use your method as I understand it, the M1 is infinitely more power efficient. . .

 

[Abwx]

I'm asking you because if I use your method as I understand it, the M1 is infinitely more power efficient. . .

We could as well normalize the perfs within a comparable range...

The M1 has a single measurement but it s enough to evaluate its perf/watt, it score 9.2 at about 11W, with a score of 8.2 it would use 8W, that s two watt less than the 7840U at 8.2, so that s 25% better perf/Watt for the M1 at same 8.2 throughput.

That being said there s something that is not right with this curve unless that s a single core test.

 

[Hitman928]

We could as well normalize the perfs within a comparable range...

The M1 has a single measurement but it s enough to evaluate its perf/watt, it score 9.2 at about 11W, with a score of 8.2 it would use 8W, that s two watt less than the 7840U at 8.2, so that s 25% better perf/Watt for the M1 at same 8.2 throughput.

That being said there s something that is not right with this curve unless that s a single core test.

You can't just make up data. The problem is your method falls apart under certain scenarios. For example, any scenario where 1 CPU can reach a performance the other can't no matter how much power you push into it. In that scenario, the CPU that reaches a higher performance will be infinitely more power efficient at that performance, which of course makes no sense. No one calculates power efficiency in a general way the way you are doing it. What some people do (again, usually more on the enterprise side) is calculate energy used per task, but you are not doing that. Some times, in limited scenarios it makes sense to set a performance target and calculate efficiency around that target (say GPUs with a frame rate target) but that is not how things are done generally because it always comes with stipulations and you would still calculate performance per watt as (fps target) / watts.

You are free to use your efficiency calculation if you wish, but it is not how it is done as an industry standard and you shouldn't expect anyone else to accept your way of calculating efficiency.

 

[cebri1]

We could as well normalize the perfs within a comparable range...

The M1 has a single measurement but it s enough to evaluate its perf/watt, it score 9.2 at about 11W, with a score of 8.2 it would use 8W, that s two watt less than the 7840U at 8.2, so that s 25% better perf/Watt for the M1 at same 8.2 throughput.

That being said there s something that is not right with this curve unless that s a single core test.

Nope applying your same method, M1 is AT LEAST 45% more efficient than the 7840U.

 

[Abwx]

No one calculates power efficiency in a general way the way you are doing it. What some people do (again, usually more on the enterprise side) is calculate energy used per task, but you are not doing that.

And what am i doing..?..

When the 7840U score 8.2 in Spec it means that it did a specified task in a time X, if i set the M1 such that it score also 8.2 it means that it do the job in the same time than the 7840U.

When set this way the 7840U use 10W say for 10s, meanwhile the M1 use 8W during the same 10s to execute the same amount of work, overall it used 25% less energy to do the task, if that s not energy per task what is it.?...

To not bias the result such a comparison must be done at same throughput, wich i did.

 

[H433x0n]

You can't just make up data. The problem is your method falls apart under certain scenarios. For example, any scenario where 1 CPU can reach a performance the other can't no matter how much power you push into it. In that scenario, the CPU that reaches a higher performance will be infinitely more power efficient at that performance, which of course makes no sense. No one calculates power efficiency in a general way the way you are doing it. What some people do (again, usually more on the enterprise side) is calculate energy used per task, but you are not doing that. Some times, in limited scenarios it makes sense to set a performance target and calculate efficiency around that target (say GPUs with a frame rate target) but that is not how things are done generally because it always comes with stipulations and you would still calculate performance per watt as (fps target) / watts.

You are free to use your efficiency calculation if you wish, but it is not how it is done as an industry standard and you shouldn't expect anyone else to accept your way of calculating efficiency.

Multiple people have had this same conversation with him (including me). It’s too tantalizing for him to measure it as watt/perf since it exaggerates the difference of whichever CPU is more efficient (which in this case benefits AMD).

I’m surprised he’s accepting that graph at face value tbh. It shows a big perf/watt boost that arguably even exceeds GoldenPig’s data with new microcode. The implications of that are Phoenix no longer being dramatically more efficient and outperformed post 30-35W.

 

[Abwx]

Multiple people have had this same conversation with him (including me). It’s too tantalizing for him to measure it as watt/perf since it exaggerates the difference of whichever CPU is more efficient (which in this case benefits AMD).

I’m surprised he’s accepting that graph at face value tbh. It shows a big perf/watt boost that arguably even exceeds GoldenPig’s data with new microcode. The implications of that are Phoenix no longer being dramatically more efficient and outperformed post 30-35W.

But if things were to be compared at same power and not same throughput why did Intel compare their 45W boosted 155H to a 7840U that is used in the 25-30W range in all regular designs..?...

They could had set their CPU in the 25-30W range as well, in wich case your comparison at same power would make sense, but it happens that in real world the comparison is forcibly at same throughputs or so since they want to be competitive perfs wise, so you re talking essentialy of a theorical metric that does not exist in real laptops...

 

[H433x0n]

They could had set their CPU in the 25-30W range as well, in wich case your comparison at same power would make sense, but it happens that in real world the comparison is forcibly at same throughputs or so since they want to be competitive perfs wise, so you re talking essentialy of a theorical metric that does not exist in real laptops...

So what you're saying is that despite this new data showing a +20% perf/watt increase you still don't believe GoldenPig's data?

 

[Hitman928]

And what am i doing..?..

When the 7840U score 8.2 in Spec it means that it did a specified task in a time X, if i set the M1 such that it score also 8.2 it means that it do the job in the same time than the 7840U.

When set this way the 7840U use 10W say for 10s, meanwhile the M1 use 8W during the same 10s to execute the same amount of work, overall it used 25% less energy to do the task, if that s not energy per task what is it.?...

To not bias the result such a comparison must be done at same throughput, wich i did.

Others don’t adjust the power consumption to equalize the times of completion. Your method doesn’t make sense as a general way for comparing efficiency between processors and in fact can’t be applied generally because it falls apart in some scenarios. Additionally, it is not a useful comparison for the vast majority of consumer use cases.

You can calculate it that way if you want to, but your not going to be able to convince anyone else here that it’s the proper way to measure efficiency, especially when everyone else is following the industry standard way of doing it.

 

[Hitman928]

So what you're saying is that despite this new data showing a +20% perf/watt increase you still don't believe GoldenPig's data?

The new data is clearly being influenced by the boosted power level. Golden Pig’s data is encouraging but I’d like to see confirmation from others who I know are power limiting the CPU to the various power levels in the graph.

Edit: The new data I’m referring to is the latest Cinebench numbers.

 

[H433x0n]

Incase we have any photoshop enthusiasts would love to see MTL plotted on this graph. For some reason the OP didn’t include it.

 

[DrMrLordX]

I wish the 7840U was available in more mobile devices. It's a phenom.

In theory other Phoenix SoCs should behave identically given the same RAM. The only difference would be binning and turbo profiles.

 

[adroc_thurston]

The only difference would be binning and turbo profiles.

And fabric modes.

 

[FlameTail]

Incase we have any photoshop enthusiasts would love to see MTL plotted on this graph. For some reason the OP didn’t include it.

View attachment 91266

This is single core, I think?

M1 Max consuming half the power while delivering the same performance as it's competitors. Talk about sheer efficiency... and that's only the M1 Max. We now have M2 Max and M3 Max.

 

[H433x0n]

This is another graph that shows Crestmont actually being a true ecore. No idea why he didn’t have Crestmont and Gracemont in the same graph though.



Edit: Not sure how I feel about this this guy making the graphs. Here's a result of 7840U when using Windows OS, it's closer than the reference point he's been using against MTL running Windows OS. He's also not providing the comparison points for Gracemont v Crestmont and Crestmont v Zen 4C.

 

[Abwx]

So what you're saying is that despite this new data showing a +20% perf/watt increase you still don't believe GoldenPig's data?

I dont think that there s a 20% better perf/Watt, that doesnt make sense, what is likely is that it s the power profile that has been changed, the CPU boost a longer time and this increase the score by 20% at the "28W" point.

 

[TESKATLIPOKA]

Yes, because if you dont take account of the time those numbers do not make sense, as said RPL clocked low enough would have better perf/watt than both MTL and 7840U, but it would use much more time to complete a same task.

Here is that graph, show me where on that graph would RPL have better efficiency than MTL or PHX?

There is no such point where RPL would be more efficient.

You can find such a point for MTL vs PHX, where MTL would be more efficient
Example: PHX: 8.6 at 15W vs MTL: 7.4 at 10W
But this would still be a nonsense, because you don't compare It at the same ISO power or same ISO performance.

 

[H433x0n]

I dont think that there s a 20% better perf/Watt, that doesnt make sense, what is likely is that it s the power profile that has been changed, the CPU boost a longer time and this increase the score by 20% at the "28W" point.

There’s a graph in this thread showing MTL getting +20% perf/watt at 10W over RPL. Crestmont is almost double the efficiency of Gracemont at 5W. You don’t think this translates to workloads other than Specint?

 

[Abwx]

Here is that graph, show me where on that graph would RPL have better efficiency than MTL or PHX?
View attachment 91285
There is no such point where RPL would be more efficient.

You can find such a point for MTL vs PHX, where MTL would be more efficient
Example: PHX: 8.6 at 15W vs MTL: 7.4 at 10W
But this would still be a nonsense, because you don't compare It at the same ISO power or same ISO performance.

At 7.5W RPL does 0.65 pts per watt, at 16W the 155H does 0.51 pts per watt, the former is more efficient in this case, that s what i said in another post, if you increase the execution time of a CPU you increase its perf/watt.

In this case it s not a comparison at same throughput, wich is the bias i was talking about, comparing CPUs at different throughputs is not indicative of their respective perf/watt.

 

[TESKATLIPOKA]

This is another graph that shows Crestmont actually being a true ecore. No idea why he didn’t have Crestmont and Gracemont in the same graph though.

View attachment 91278

What is this weird graph?
There is:
155H P-core
155H LP P-core
155H E-core
155H LP E-core
There is no such thing like 155H LP P-core.
Also why E-core is more efficient than LP-E core? If It's because of SoC being made on a worse process, then I have to question If that LP E-core in really save power.