Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads

Tigerick · Aug 22, 2022

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.)

Hitman928 · Aug 4, 2024

TwistedAndy said:
I use 125H with 4.4GHz because it has a lower PL1 (28W). Even in this case, a single P-core uses nearly half of the PL1 limit. That was the initial point.

Obviously, power limits and the power configuration depend on the OEM and can even be dynamically adjusted by the system.

Returning to Lunar Lake, we don't know how much power it can consume under full ST load. I expect to see something around 10W for 5.0GHz.

There really isn't correlation between PL1 value and ST power consumption on MTL. 155H also has a specified PL1 of 28W. Additionally, the PL1 value is adjustable by the laptop maker, so it's not relevant unless the ST boost uses that much (which the 155H just about does) and you want to sustain that power level past the PL2 duration. What is relevant is how much power it actually uses to reach boost frequencies as that will tell you if it can sustain that frequency or not given the PL1 values. Even if it can't, it will still boost past the PL1 value, it just won't be able to sustain it once the PL2 duration time ends.

I think LNL going from > 30 W to reach 5.1 GHz to 10 W is extremely unlikely, even with the better node. Most likely, LNL won't be able to sustain 5 GHz or above at 17 W. Reducing the tile count to 2 and being on a better node will certainly give it a chance, but I still think it's unlikely. I think that either it will have to lower clocks once the PL2 runs out (except the top SKU which has a PL1 of 30 W), or it will need to be configured for higher PL1.

DavidC1 · Aug 4, 2024

TwistedAndy said:
Obviously, power limits and the power configuration depend on the OEM and can even be dynamically adjusted by the system.

Why does that matter in this argument? If they lower the power limit then Meteorlake won't hit 5.1GHz and will perform worse. It shows it needs 30W to just have 1 core active.

I doubt the process itself will do much. According to Intel data it's anywhere from little over 10% to close to 20% gain in performance per clock over Redwood Cove. Since the top Lunarlake clocks the same 5.1GHz, that means the P core will be 10-20% faster in ST but use similar >25W as Meteorlake does.

Of course, using the same data you can clock it down to 4.5GHz for the same performance and probably use 20W, but that rarely works out.

poke01 said:
Here is the difference tho, an M4 using 16-20 watts at 5.1GHz will still run circles around a Zen 5/lion cove mobile core.

The 4.4GHz M4 runs circles around Zen 5/LNC.

DavidC1 · Aug 4, 2024

Doug S said:
I'd also argue area doesn't matter NEARLY as much as people here seem to want to believe.

Area matters in terms of the quality of the engineering of the core, which is the most important part of the CPU.

Area also shows roughly the amount of transistors(thus power) it took to create the said core. Apple beats the competition to a pulp while the core area is smaller. It has both exceptional perf/watt and perf/mm2, and even absolute performance.

Philste said:
And that's why I'm disappointed with Strix. Only good looking results is at Phoronix, in rest of the Tests it's about 20% faster than Phoenix at same TDP, despite having 50% more Cores/Threads. 7900X is 32% faster than 7700X at same TDP.

Yea, because Strix has E cores, which are slower. And despite the argument that it's superior because of the same core arrangement, in the big picture the differences are really small. Reviews show scheduler issues, and tests show Zen 5 is actually 6% faster than Zen 5c.

ErichZannMusician · Aug 4, 2024

This is why I take Linux benchmarking results with heaps of salt:

Tiny Linux Patch Up To 32% Faster, Up To 18% Less Energy For Intel Xeon Emerald Rapids - Phoronix

www.phoronix.com

Updated Intel Meteor Lake Tuning For Linux Shows Huge Performance/Power Improvements - Phoronix

www.phoronix.com

No offense to Michael Larabel.

TwistedAndy · Aug 4, 2024

DavidC1 said:
Why does that matter in this argument?

Because the CPU power characteristics depend on a lot of settings and microcode optimization.

My favorite examples are Dell Precision 7670 and 7770, which suffered severe performance issues because of the incorrect IA AC/DC Loadline values. This issue was fixed 6-7 months after release.

Another example is the incorrect EPP values for Meteor Lake and some other platforms: Updated Intel Meteor Lake Tuning For Linux Shows Huge Performance/Power Improvements or Tiny Linux Patch Up To 32% Faster, Up To 18% Less Energy For Intel Xeon Emerald Rapids.

Also, there are great and very useful features like Intel DTT and IPF (no), which often manipulate power limits in a very strange way.

I won't mention various issues with voltage offsets, which literally move the V/f curve up or down, making all our speculations pointless.

Hitman928 said:
I think LNL going from > 30 W to reach 5.1 GHz to 10 W is extremely unlikely, even with the better node. Most likely, LNL won't be able to sustain 5 GHz or above at 17 W. Reducing the tile count to 2 and being on a better node will certainly give it a chance, but I still think it's unlikely. I think that either it will have to lower clocks once the PL2 runs out (except the top SKU which has a PL1 of 30 W), or it will need to be configured for higher PL1.

I hope we will get more details on Hot Chips in a few weeks.

Doug S · Aug 4, 2024

DavidC1 said:
Area matters in terms of the quality of the engineering of the core, which is the most important part of the CPU.

Area also shows roughly the amount of transistors(thus power) it took to create the said core. Apple beats the competition to a pulp while the core area is smaller. It has both exceptional perf/watt and perf/mm2, and even absolute performance.

The "quality of engineering"? You can tell that based on overall performance and performance/watt. I'd say the quality of engineering is pretty poor if someone managed to make a core that does poorly on both those metrics that actually matter, but somehow manages to do well on the totally irrelevant metric performance/mm^2. That's like winning best personality in a beauty contest. Its nice, but its not what you were shooting for.

poke01 · Aug 4, 2024

TwistedAndy said:
The Lion Cove core in Arrow Lake will show a similar performance to the Apple M4 P-core.

I’m sure it will but at vastly higher frequencies and power consumption.

TwistedAndy said:
Currently, the P-core in Apple M4 is 14.4% faster than Raptor Cove in mobile 14900HX in Geekbench 5:

This doesn’t mean much, you are comparing a fanless chip to a big giant laptop and even then the 14900HX loses. Do you have any idea the amount of watts that 14900hx used to reach 5.8GHz and to deliver that ST score?

Look, lunar lake will be a great product and it will live up to what Intel claimed. Like you said no point bickering and best to wait for benchmarks so let’s leave it there.

DavidC1 · Aug 4, 2024

TwistedAndy said:
Because the CPU power characteristics depend on a lot of settings and microcode optimization.

What do you think those settings and ucode do? Especially in ST. You won't get a 30W core to magically go down to 15W with BIOS settings. At 4.4GHz or so where Meteorlake needs to get the power level you wish for, the Apple core will outperform it by 60%, while consuming almost 40% less power.

Doug S said:
The "quality of engineering"? You can tell that based on overall performance and performance/watt. I'd say the quality of engineering is pretty poor if someone managed to make a core that does poorly on both those metrics that actually matter, but somehow manages to do well on the totally irrelevant metric performance/mm^2. That's like winning best personality in a beauty contest. Its nice, but its not what you were shooting for.

Please tell me in what area the Apple cores are inferior in?

It absolutely curb stomps both P cores, to the point where it makes Conroe vs Netburst an embarassment. At least in the previous years it was able to keep ST lead with near 6GHz clocks. M4 proves that's not even true anymore.

Both of the x86 E cores are superior in the perf/W metric to their P core counterparts too. Need I also remind you that AMD is significantly better than Intel in that regard? Smaller core and happens to be much better in perf/W. Sunny/Golden/Lion are all disappointments. Conroe was also a smaller core to Netburst too.

While there are other factors that determine the size of the core, at such levels of difference, yes it indeed determines how good it is engineered.

Doug S · Aug 4, 2024

DavidC1 said:
Please tell me in what area the Apple cores are inferior in?

It absolutely curb stomps both P cores, to the point where it makes Conroe vs Netburst an embarassment. At least in the previous years it was able to keep ST lead with near 6GHz clocks. M4 proves that's not even true anymore.

Both of the x86 E cores are superior in the perf/W metric to their P core counterparts too. Need I also remind you that AMD is significantly better than Intel in that regard? Smaller core and happens to be much better in perf/W. Sunny/Golden/Lion are all disappointments. Conroe was also a smaller core to Netburst too.

While there are other factors that determine the size of the core, at such levels of difference, yes it indeed determines how good it is engineered.

Where did I say Apple's cores were inferior? I wasn't talking about a specific architecture at all in the post you quoted. If you think I did, you're inferring something out of thin air that I didn't state or imply.

Yes M4 dominates in both absolute perf and perf/watt, so I don't see what would be the point of tossing in a perf/mm^2 comparison. That's pretty meaningless, especially if you're only considering the size of the cores themselves and not including higher levels of cache, the memory controllers, or the intercore and interchip communication (ring bus, IOD or AS Ultra's "10,000 I/Os")

And yes it is pretty obvious that everyone's E cores beat their respective P cores in perf/watt. Otherwise what reason would they have for existing? As you say AMD is better in perf/watt so what difference does it make if it is a smaller core? Yeah when you want hundreds of cores its nice to fit more of them in a given silicon area, but in the age of chiplets that doesn't really matter or limit you the way it did when server chips were on a single piece of silicon limited by yield (or failing that, the reticle)

TwistedAndy · Aug 5, 2024

poke01 said:
This doesn’t mean much, you are comparing a fanless chip to a big giant laptop and even then the 14900HX loses. Do you have any idea the amount of watts that 14900hx used to reach 5.8GHz and to deliver that ST score?

Unfortunately, I don't have accurate data for 14th-gen ST power consumption. We need to keep in mind that 14900HX is three major nodes behind (10nm vs. 3nm), not to mention sub-nodes like N3B and N3E. Arrow Lake is expected to be much more representative here.

DavidC1 said:
What do you think those settings and ucode do? Especially in ST. You won't get a 30W core to magically go down to 15W with BIOS settings. At 4.4GHz or so where Meteorlake needs to get the power level you wish for, the Apple core will outperform it by 60%, while consuming almost 40% less power.

Those settings (AC/DC Loadline, voltage offsets, and the microcode defaults) adjust the position of V/f curve. As a result, a 12950HX CPU can, for example, run at low 2.6/2.0GHz frequencies and consume ~75W of power under MT load. That's exactly what happened with Dell Precision 7670 and 7770 (link).

It's worth mentioning that in late 2022, Intel released the microcode update, slightly increasing the voltages and introducing useless features like "Undervolt Protection" to "improve the system security and stability".

As a result, some of the 13th and 14th-gen SKUs are now having stability issues because of the voltages being too high. What a surprise!

poke01 · Aug 5, 2024

https://twitter.com/x/status/1820346860990578817

Those Skymont cores are doing some work for MT. This is the 2nd last SKU of the LNL family. Not bad. As expected the ST is mediocre.

511 · Aug 5, 2024

Not only that david Huan on twitter said Intel's marketing slide can be taken at face value with the caveats of the testing

https://twitter.com/x/status/1820042682397237273

FlameTail · Aug 5, 2024

poke01 said:
This is the 2nd last SKU of the LNL family. Not bad. As expected the ST is mediocre.

Don't worry. It will keep company with the 3.4 GHz X1E78100

majord · Aug 5, 2024

511 said:
Not only that david Huan on twitter said Intel's marketing slide can be taken at face value with the caveats of the testing

https://twitter.com/x/status/1820042682397237273

what are we talking here, the 14% IPC > RPC and 2% for Skymont? was there more perf data on the slides? (other than core efficiency stuff)

majord · Aug 5, 2024

511 said:
Around 14-15W i guess

According to slides it should use more than MTL @ top freq?

Nothingness · Aug 5, 2024

511 said:
Not only that david Huan on twitter said Intel's marketing slide can be taken at face value with the caveats of the testing

https://twitter.com/x/status/1820042682397237273

But don't forget to read the fine prints on their slides.

511 · Aug 5, 2024

Nothingness said:
But don't forget to read the fine prints on their slides.

I know that's why i said caveats either way marketing being true feels weird 😅

Magio · Aug 5, 2024

511 said:
Around 14-15W i guess

I can't imagine it being much higher than that indeed, these benchmarks were made at 17W PL1/PL2 and 30W PL1/PL2 and, while the 30W results are kinda borked, the 17W seem in line with expectations in ST perf (roughly 10% behind M3):

https://twitter.com/x/status/1805718733207298541

Since LNL at 17W includes 2W for memory, this should mean a Lion Cove thread with a 15W power limit can do that score. It may even consume a bit less, but in theory it shouldn't pull more.

Hitman928 · Aug 5, 2024

poke01 said:
https://twitter.com/x/status/1820346860990578817

Those Skymont cores are doing some work for MT. This is the 2nd last SKU of the LNL family. Not bad. As expected the ST is mediocre.

ST score is solid for the clock speed. If LNL can really reach 5.1 GHz, it should match, or even slightly beat, STX ST performance in GB6.

Philste · Aug 5, 2024

Hitman928 said:
If LNL can really reach 5.1 GHz, it should match, or even slightly beat, STX ST performance in GB6.

There already is an 2900 result in Geekbench. So yeah, it's slightly above Strix Point.

CouncilorIrissa · Aug 5, 2024

Philste said:
There already is an 2900 result in Geekbench. So yeah, it's slightly above Strix Point.

It's also impressive in the context that that 2900 run shows some variance in clock speed, whereas the only STX runs breaking 2.9k on Windows are the ones that sustain 5.1 (which is very few of them on P16, an practically none on the Zenbook).

DavidC1 · Aug 5, 2024

511 said:
Not only that david Huan on twitter said Intel's marketing slide can be taken at face value with the caveats of the testing

https://twitter.com/x/status/1820042682397237273

At least Intel is winning on something.

Magio said:
I can't imagine it being much higher than that indeed, these benchmarks were made at 17W PL1/PL2 and 30W PL1/PL2 and, while the 30W results are kinda borked, the 17W seem in line with expectations in ST perf (roughly 10% behind M3):

The early leaks need to be taken with a grain of salt because especially in that score, as you said the MT results are wonky. True it could be that it's excellent at 17W, but the just released 228V score outperforms the 30W in MT and not by a small amount.

SiliconFly · Aug 5, 2024

DavidC1 said:
The early leaks need to be taken with a grain of salt because especially in that score, as you said the MT results are wonky. True it could be that it's excellent at 17W, but the just released 228V score outperforms the 30W in MT and not by a small amount.

Tbh, haven't read much about 228V and its performance. But isn't it one of those low-end parts that people usually don't care about?

511 said:
... said Intel's marketing slide can be taken at face value ...

DavidC1 said:
At least Intel is winning on something.

Intel *always* wins in power point slides!

Magio · Aug 6, 2024

DavidC1 said:
The early leaks need to be taken with a grain of salt because especially in that score, as you said the MT results are wonky. True it could be that it's excellent at 17W, but the just released 228V score outperforms the 30W in MT and not by a small amount.

The 228V score we have is in Geekbench 6, this is Geekbench 5.4.

I do agree we shouldn't take those early scores entirely at face value, the 30W is clearly weird and MT in general seems a bit disappointing. But my point is purely that a single Lion Cove thread seems capable of 2000+ on Geekbench 5 at a 15W power limit, which should mean that in LNL there's not much reason to push that thread beyond 15W.

511 · Aug 6, 2024

Magio said:
The 228V score we have is in Geekbench 6, this is Geekbench 5.4.

I do agree we shouldn't take those early scores entirely at face value, the 30W is clearly weird and MT in general seems a bit disappointing. But my point is purely that a single Lion Cove thread seems capable of 2000+ on Geekbench 5 at a 15W power limit, which should mean that in LNL there's not much reason to push that thread beyond 15W.

There is only one issue with Lunar lake it is not cheap it is even more expensive than Strix Point

Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads

Senior member

Attachments

Diamond Member

Senior member

Senior member

Member

Member

Platinum Member

Platinum Member

Senior member

Platinum Member

Member

Platinum Member

Senior member

Diamond Member

Senior member

Senior member

Diamond Member

Senior member

Member

Diamond Member

Senior member

Senior member

Senior member

Golden Member

Member

Senior member