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]

Ok so going by your logic Apple silicon MacBook pros with 99 Wh batteries are inefficient as well.

Their battery in the 14" is 70Wh, but at the same time the laptop cost 3900$,
so not comparable with HP s 360.

And deceptive TDPs? Please show me where AMD lists the short duration power limits in their product description webpages.

Lo, short duration you said....?.
like the 14900K wich is stated as 125W while at stock setting, that is out of the box, it s not even limited to 253W but is working at full 350W tilt all the time.

It goes the same for mobile, there are short durations that actually last forever, it is years that they state FI 28W while the devices are benched at 54-62W for "short" durations, so that s not only TDP but also perfs claims that are deceptive.

Lenovo Yoga with Phoenix has a bigger battery than that. Poor efficiency according to you.

2-3Wh bigger...
If anything this point to MTL not being better than PHX power efficency wise, if they use TSMC s N4P/N3 they can eventually match AMD s efficency assuming that their hybrid CPU uarch is as efficient than Zen 4..

For the time the equivalent current product, the 13700H wich is 6 + 8 has half the perf/watt of a 7840U at low power, wich means that the 13700H@13W has same scores and efficency as a 7840U@7-8W, and that s not even a direct competitor since 6P + 8E amount to 10P.

 

[tamz_msc]

Their battery in the 14" is 70Wh, but at the same time the laptop cost 3900$,
so not comparable with HP s 360.



Lo, short duration you said....?.
like the 14900K wich is stated as 125W while at stock setting, that is out of the box, it s not even limited to 253W but is working at full 350W tilt all the time.

It goes the same for mobile, there are short duration that actually last forever...



2-3Wh bigger...
If anything this point to MTL not being better than PHX power efficency wise, if they use TSMC s N4P/N3 they can eventually match AMD s efficency assuming that their hybrid CPU uarch is as efficient than Zen 4..

If price determined efficiency, then everyone would be driving Rimacs instead of Teslas.

And congratulations, you've once again proved that you know nothing.

 

[SiliconFly]

Meteor Lake – GPU Architecture

Finally some clarity. He says MTL's Xe Vector Engines directly support DP4a used by XeSS which is the reason they didn't include dedicated XMX units. Reduced die area saves cost. Without XMX, XeSS performance drop should be minimal (probably in single digit).

 

[SiliconFly]

He's also confirmed that the main power savings in MTL is brought about by keeping the CPU tile & GPU tile powered down most of the time. Remember, most of the people in the world just use browsers, mail apps or some office/productivity apps most of the time.

That seems to be their intended target. MTL's power efficiency should be out of the world in these cases, which is majority of the average users in the world. But not so much in gaming, video editing, etc. And most importantly, it woundn't perform well in power efficiency benchmarks that stresses all tiles continuously. Such a benchmark wouldn't reflect MTL's real world usage scenario.

 

[BorisTheBlade82]

He's also confirmed that the main power savings in MTL is brought about by keeping the CPU tile & GPU tile powered down most of the time. Remember, most of the people in the world just use browsers, mail apps or some office/productivity apps most of the time.

That seems to be their intended target. MTL's power efficiency should be out of the world in these cases, which is majority of the average users in the world. But not so much in gaming, video editing, etc. And most importantly, it woundn't perform well in power efficiency benchmarks that stresses all tiles continuously. Such a benchmark wouldn't reflect MTL's real world usage scenario.

Yep, a couple of years ago, AMD stated that Intel admittedly had the best CPUs at "doing nothing". Looks like MTL is another one in that line.

 

[Hulk]

Yep, a couple of years ago, AMD stated that Intel admittedly had the best CPUs at "doing nothing". Looks like MTL is another one in that line.

Perhaps. But if shutting down the graphics and cpu tile actually allow web browsing, office work, and watching videos at extremely low power usage then those activities will have to be redefined as "doing nothing."

 

[H433x0n]

Yep, a couple of years ago, AMD stated that Intel admittedly had the best CPUs at "doing nothing". Looks like MTL is another one in that line.

Did they really say that?

 

[inf64]

Looks like MTL is a real dud. No x86 performance info at all, and it will be lucky if it beats Phoenix in non-synthetic workloads with maybe a bit better battery life. Intel will also have to compete with Hawk Point and soon with Zen 5 derivatives. ARL better be +30% CPU and +2x GPU uplift or they are DOA.

 

[jpiniero]

Looks like MTL is a real dud. No x86 performance info at all, and it will be lucky if it beats Phoenix in non-synthetic workloads with maybe a bit better battery life. Intel will also have to compete with Hawk Point and soon with Zen 5 derivatives. ARL better be +30% CPU and +2x GPU uplift or they are DOA.

Hasn't really mattered up to now since the AMD competition is really Cezanne. Maybe that will change?

 

[FlameTail]

So Lunar Lake is rumoured to have a 160-bit memory bus.

That's weird.

Very weird.

 

[FlameTail]

I mean the bus width itself isn't unusual. After all, each LPDDR5X channel has a width of 16 bits.

So 160-bit wide bus means 10 channel.

That's not the issue. The issue is how are memory packages going to be paired with this thing?

As far as I know, LPDDR memory packages come in sizes of 32, 64, 128 bits per package.

So they'll have to use 5× 32 bit packages, which doesn't sound very cost effective.

Unless something like an 80 bit package exists. Then 2× 80 bit packages can do the job.

 

[mikk]

So Lunar Lake is rumoured to have a 160-bit memory bus.

That's weird.

Very weird.

Where do you see a 160 bit rumor?

 

[Hulk]

Looks like MTL is a real dud. No x86 performance info at all, and it will be lucky if it beats Phoenix in non-synthetic workloads with maybe a bit better battery life. Intel will also have to compete with Hawk Point and soon with Zen 5 derivatives. ARL better be +30% CPU and +2x GPU uplift or they are DOA.

+30% CPU and 2x GPU for ARL never gonna happen. Better dump your Intel stock now everybody.
Then again it wouldn't be the first time somebody on this forum pronounced Intel/AMD DOA.

 

[FlameTail]

Intel Lunar Lake-MX SoC with On-Package LPDDR5X Memory Detailed

With the reality of high performance Arm processors from Apple and Qualcomm threatening Intel's market share in the client computing space, Intel is working on learner more PCB-efficient client SoCs that can take the fight to them, while holding onto the foundations of x86. The first such...

www.techpowerup.com

Hmm. Perhaps they got it wrong?

 

[tamz_msc]

Intel Lunar Lake-MX SoC with On-Package LPDDR5X Memory Detailed

With the reality of high performance Arm processors from Apple and Qualcomm threatening Intel's market share in the client computing space, Intel is working on learner more PCB-efficient client SoCs that can take the fight to them, while holding onto the foundations of x86. The first such...

www.techpowerup.com

Hmm. Perhaps they got it wrong?

It's most likely a typo. It should be 16-bit instead of 160-bit.

 

[H433x0n]

Looks like MTL is a real dud. No x86 performance info at all, and it will be lucky if it beats Phoenix in non-synthetic workloads with maybe a bit better battery life. Intel will also have to compete with Hawk Point and soon with Zen 5 derivatives. ARL better be +30% CPU and +2x GPU uplift or they are DOA.

Let’s say in a hypothetical scenario MTL marginally outperforms Phoenix and matches Hawk Point in all aspects (1T, nT & efficiency) how would it be DOA? If anything you could make the inverse argument since the rest of the intangibles are all in Intel’s favor (volume & OEM support).

 

[Abwx]

Let’s say in a hypothetical scenario MTL marginally outperforms Phoenix and matches Hawk Point in all aspects (1T, nT & efficiency) how would it be DOA? If anything you could make the inverse argument since the rest of the intangibles are all in Intel’s favor (volume & OEM support).

6P + 8E is comparable to 10P, so if they re not competitive with an equivalent 25% core count advantage it would be quite a disaster.

So far a current 6P + 8E 13700H require 45W to match a 7840U@25W, if we take account that the uncore use more proportionately in a 25W set CPU than in a 45W one then 2x the perf/watt improvement wont be too much to be competitive again.

That s compared to a 8C/16T, if AMD release a 12C/24T they would get back the advantage, and largely, even with 10C/20T variants and at same process than their current mobile SKUs.

 

[H433x0n]

6P + 8E is comparable to 10P, so if they re not competitive with an equivalent 25% core count advantage it would be quite a disaster.

It’s not, the 6P+8E configuration takes equivalent die space as 8 Redwood Cove cores. That’s the point of the e-cores - maximize area efficiency to make up for Intel’s current oversized P cores.

So far a current 6P + 8E 13700H require 45W to match a 7840U@25W, if we take account that the uncore use more proportionately in a 25W set CPU than in a 45W one then 2x the perf/watt improvement wont be too much to be competitive again.

That s compared to a 8C/16T, if AMD release a 12C/24T they would get back the advantage, and largely, even with 10C/20T variants and at same process than their current mobile SKUs.

It was a hypothetical, neither of us know how it’ll perform.

 

[Abwx]

It’s not, the 6P+8E configuration takes equivalent die space as 8 Redwood Cove cores. That’s the point of the e-cores - maximize area efficiency to make up for Intel’s current oversized P cores.

Were talking CPU perfs, not optimisation of the waffer surfaces to increase manufacturing margins.

As for oversized P cores they are such because they are manufactured with a node equivalent to TSMC s N7, if AMD had released Zen 4 with the same process as Zen 3 they wouldnt boast area advantage over Intel as they did in their marketing slides, eventually 10-15% but not much more.

It was a hypothetical, neither of us know how it’ll perform.

The fact that Intel didnt release numbers that close to launch is somewhat suspicious, but whatever the numbers we already know that they need 2x better CPU perf/watt compared to a 13700H to match a 7840U in CPU usage.

For their GPU if they increase the EU count by 1.33x then at same frequency and process this would increase power by 1.33x as well, while inceasing frequency by 1.5x will increase power by about 2.5x, so that s 1.33 x 2.5 = 3.32x the power for 2x the perfs.

Even with a process that would have 2x the efficency of their Intel 7 this would require to increase the GPU power enveloppe by 1.66x to meet the 2x GPU perf target.

 

[tamz_msc]

So far a current 6P + 8E 13700H require 45W to match a 7840U@25W

Neither the 13700H nor the 7840U actually consume that much running the most stupid benchmark imaginable for laptops, that is Cinebench. To be more precise, the average power consumption during the first run of any CB MT test is higher than those numbers, by a factor of 2x in case of the latter.

 

[Abwx]

Neither the 13700H nor the 7840U actually consume that much running the most stupid benchmark imaginable for laptops, that is Cinebench. To be more precise, the average power consumption during the first run of any CB MT test is higher than those numbers, by a factor of 2x in case of the latter.

There s enough data at NBC, there s tons of tests, among others one laptop where the 13700H is at a stable PL1 = PL2 = 45W, that s a good reference, it score 1800 pts in CB R15, same score as a 7840U stuck at 25W.

 

[tamz_msc]

There s enough data at NBC, there s tons of tests, among others one laptop where the 13700H is at a stable PL1 = PL2 = 45W, that s a good reference, it score 1800 pts in CB R15, same score as a 7840U stuck at 25W.

Yeah NBC doesn't provide enough data to conclude anything. And CB R15? That takes maybe 5-10s to run the MT test? LOL. Show me any 7840U giving you the same perf as a 13700H with 45W PL2 while being "stuck as 25W".

You're clueless about how laptops work, and are extrapolating based on desktop behavior, which is wrong, misleading, and biased.

 

[TESKATLIPOKA]

Neither the 13700H nor the 7840U actually consume that much running the most stupid benchmark imaginable for laptops, that is Cinebench. To be more precise, the average power consumption during the first run of any CB MT test is higher than those numbers, by a factor of 2x in case of the latter.

If the reviewer has a brain, then he will not test only a single run of CB R23, but will wait for the short burst to end and use only that score.

Hopefully, this is only with sustained TDP.
27% difference at 40W TDP.
At lower TDP the difference is probably bigger.
13700H needs 65W to be on par with 40W Phoenix.

MTL needs a good 30% higher clocks at 40W to be comparable to Phoenix.

edit: or less, If there is some IPC increase.

 

[Abwx]

Yeah NBC doesn't provide enough data to conclude anything. And CB R15? That takes maybe 5-10s to run the MT test? LOL. Show me any 7840U giving you the same perf as a 13700H with 45W PL2 while being "stuck as 25W".

You're clueless about how laptops work, and are extrapolating based on desktop behavior, which is wrong, misleading, and biased.

Lol, you negate the numbers once they dont suit your intel bias, live with it, numbers can be checked, it s just that you are disgruntled by your favourite brand lagging AMD.

NBC put all runs including the first one, we can see the perf at peak power and then at throttled power, here we have a 45W 13700H wich is not throttled; it score 1800 pts from the first to the last run :

https://www.notebookcheck.net/Schenker-Vision-16-2023-Review-16-inch-lightweight-with-many-ports.738862.0.html

And a 7840U set at 30/24W, at 30W it score 1930 pts and at 24W the score decrease to 1800 pts :

https://www.notebookcheck.net/Lenovo-ThinkPad-T14s-G4-review-Business-laptop-is-better-with-AMD-Zen4.763581.0.html

So much for clulessness, numbers are numbers, and they say that you dont even have the slightest idea about Intel s huge disadvantage currently, that s why you are overestimating MTL s eventual numbers...

 

[TESKATLIPOKA]

Yeah NBC doesn't provide enough data to conclude anything. And CB R15? That takes maybe 5-10s to run the MT test? LOL. Show me any 7840U giving you the same perf as a 13700H with 45W PL2 while being "stuck as 25W".

NBC uses a 25 cycle loop for Cinebench R15. That's good for getting rid of the short burst, what is nonsensical is that they use the first result in their chart.