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.)

 

[Wolverine2349]

It in fact does not exist. In the lab where I work we have several codenames for products that start, fail and the lesson learned are merged with the next one. A R&D exercise without a concrete instance is not a product and therefore does not exist. I cannot get sad about a potential something that never had any details but a youtuber or leaker telling me something great was in the works but canceled, now smear Intel, they were to able to launch this rumored product they cannot execute...

To me that logic is nonsense!

I look at launched products and evaluate for my needs:
Sandybridge -> epic
Haswell -> epic
broadwell -> alright
Skylake -> epic
Skylake again -> alright
Skaylake again again -> tired
Sunny Cove -> are we back
Cypress Cove -> WTF...
Willow Cove (Tiger Lake) -> epic
Golden Cove (Alder Lake) -> epic
Redwood Cove -> ???
Golden Cove again -> tired

Why do you say Golden Cove again tired? What does that mean? I mean Lion Cove is not Golden Cove its a 14% IPC uplift over even the 5% better IPC than GLC RPC is it not? So Lion Cove 19% IPC uplift over GLC?

 

[jdubs03]

Ehh that 2513 pts at 5.698Ghz is worse than a good bit of 9950X submissions in ST. Most of the MT submissions are like 26000-27000. So slight advantage for the 285K.

And when I say worse, I mean points/Ghz. But their clock speed boosts are quite similar around 5.7GHz, so easy comparison.

 

[inf64]

CapFrameX posted a new 285K GB5 benckleaks result:

https://twitter.com/x/status/1826186324946149408

It’s consistent with the previous GB6 285K benchmark leak.

He measures ARL-S PPC gain at a whopping 18%.

This puts ARL clearly ahead of competition (even after accounting for the mythical future update that competition may or may not receive and/or will or will not give it a extreme performance bump).

GB6 is more relevant and the 285K and 9950X are basically neck and neck

ASUS System Product Name vs ASUS System Product Name - Geekbench

 

[SiliconFly]

Ehh that 2513 pts at 5.698Ghz is worse than a good bit of 9950X submissions in ST. Most of the MT submissions are like 26000-27000. So slight advantage for the 285K.

And when I say worse, I mean points/Ghz. But their clock speed boosts are quite similar around 5.7GHz, so easy comparison.

9950X is 3.5% slower than 285K in GB6 ST. Not equal. Not faster. But slower than 285K.

285K has the performance crown at the moment (considering the leaked scores).

 

[reb0rn]

Shame if intel would have AVX10 they could have a nice lead...

 

[Abwx]

9950X is 3.5% slower than 285K in GB6. Not equal. Not faster. But slower than 285K.

285K has the performance crown at the moment (considering the leaked scores).

There s a recent submission with a 265K at 5.5GHz that score 3080 in GB 6, can we use it as basis for ARL perf the same way as you re using a vast amount of 9950X averaged scores against only 3 285k scores average..?.

And btw, about your ARL 3450 score, pick the 9950X one you want in this long list, there s 14 pages, how much pages for the 285K ? :

Geekbench Search - Geekbench

 

[DokiDoki]

GB6 is more relevant and the 285K and 9950X are basically neck and neck.

One day GB5 is more relevant because it's "multi-core scaling is better than GB6".

The next day GB6 is more relevant because it has a bunch of AVX-512 accelerated workloads which makes my team look better than it actually is.

You can pick-and-choose whatever suits you but it won't change reality. Lion Cove is a better core than Zen 5 in pure performance.

If you insist that those tasks where Zen 5 takes the lead in Geekbench is important, then also you should be looking at ARL because it has an NPU and iGPU which can actually help with those tasks.

 

[inf64]

9950X is 3.5% slower than 285K in GB6 ST. Not equal. Not faster. But slower than 285K.

285K has the performance crown at the moment (considering the leaked scores).

I linked a comparison where 9950X is actually faster a bit. It's all under margin of error stuff, you do realize that?

 

[inf64]

One day GB5 is more relevant because it's "multi-core scaling is better than GB6".

The next day GB6 is more relevant because it has a bunch of AVX-512 accelerated workloads which makes my team look better than it actually is.

You can pick-and-choose whatever suits you but it won't change reality. Lion Cove is a better core than Zen 5 in pure performance.

If you insist that those tasks where Zen 5 takes the lead in Geekbench is important, then also you should be looking at ARL because it has an NPU and iGPU which can actually help with those tasks.

I am willing to predict that the overall ST results will be within margin of error between Zen 5 at 5.7Ghz and ARL at the same clock. I see that maybe in MT ARL could have a small advantage while in gaming it will likely lose to AMD X3D parts so nothing will change there.

 

[Abwx]

I am willing to predict that the overall ST results will be within margin of error between Zen 5 at 5.7Ghz and ARL at the same clock. I see that maybe in MT ARL could have a small advantage while in gaming it will likely lose to AMD X3D parts so nothing will change there.

14900K was faster than the 7950X in GB 6 MT, yet in Computerbase 13 MT tests it was faster only in CB R20/23, and still, at 330W power.

So far in the few numbers available in real tests, WebXPRT4, wich was designed by Intel, and Speedometer ARL doesnt even match the 14900K and is way behind the 9950X.

In CB R23 it score 43118 at allegedly 250W, but when we do the calculation using LNL 10200 pts score at 30W and iterating on TSMC s process 250W is an absolutely best case, and possibly that PL4 at 297W was engaged since they have a PL4 at this power for apps and a PL4 at 329W for the rest.

 

[DokiDoki]

I am willing to predict that the overall ST results will be within margin of error between Zen 5 at 5.7Ghz and ARL at the same clock. I see that maybe in MT ARL could have a small advantage while in gaming it will likely lose to AMD X3D parts so nothing will change there.

X3D already loses to Raptor Lake in frame-time consistency. ARL will simply widen the gap.

 

[SiliconFly]

14900K was faster than the 7950X in GB 6 MT, yet in Computerbase 13 MT tests it was faster only in CB R20/23, and still, at 330W power.

So far in the few numbers available in real tests, WebXPRT4, wich was designed by Intel, and Speedometer ARL doesnt even match the 14900K and is way behind the 9950X.

In CB R23 it score 43118 at allegedly 250W, but when we do the calculation using LNL 10200 pts score at 30W and iterating on TSMC s process 250W is an absolutely best case, and possibly that PL4 at 297W was engaged since they have a PL4 at this power for apps and PL4 at 329W for the rest.

Still hung up on the outdated & power inefficient 14900K? A new era has begun with Arrow Lake. And it’s faster than competition. The playfield has changed dramatically. It’s actually a little more than an even playfield now. Exciting times ahead.

X3D already loses to Raptor Lake in frame-time consistency. ARL will simply widen the gap.

I too believe ARL will easily match or outmatch the competition’s top X3D part.

 

[Abwx]

Still hung up on the outdated & power inefficient 14900K? A new era has begun with Arrow Lake. And it’s faster than competition. The playfield has changed dramatically. It’s actually a little more than a even playfield now. Exciting times ahead.

Problem is not the 14900K as such, it s that in the 2 ST benchs already known they are below not only this CPU but also below the 9950X, we are no more talking of GB here but of actual browsing perf, so far the IPC in those tests is below Zen 5, i posted them in my previous post and here they are again.

Guess that currently Intel is devising about the exact power they ll use since at 250W they will have a win only in CB R20/R23 for MT, and a marginal one.

They may well set 297W as PL2, or use an extended 297W PL4 since that s their apps performance profile for ARL, otherwise they ll lose in too much benches to be convincing, so we ll surely be good for another round of exagerated TDPs.

And likely that AMD limiting their CPU at 200W has somewhat disrupted their plans as they surely expected 230W, wich would had rendered their own 250W/297W TDP acceptable.

 

[inf64]

X3D already loses to Raptor Lake in frame-time consistency. ARL will simply widen the gap.

Maybe according to the troll CapeFrameX. All credible review websites put 7800X3D at the top of gaming charts.

 

[SiliconFly]

Problem is not the 14900K as such, it s that in the 2 ST benchs already known they are below not only this CPU but also below the 9950X, we are no more talking of GB here but of actual browsing perf, so far the IPC in those tests is below Zen 5, i posted them in my previous post and here they are again.

Guess that currently Intel is devising about the exact power they ll use since at 250W they will have a win only in CB R20/R23 for MT, and a marginal one.

They may well set 297W as PL2, or use an extended 297W PL4 since that s their apps performance profile for ARL, otherwise they ll lose in too much benches to be convincing, so we ll surely be good for another round of exagerated TDPs.

And likely that AMD limiting their CPU at 200W has somewhat disrupted their plans as they surely expected 230W, wich would had rendered their own 250W/297W TDP acceptable.

ARL has a two full node jump and is expected to be as efficient as competition. But the funny thing is, it doesn’t matter what power the cpu runs at. All that matters is what the stock power usage settings the manufacturer has set.

If the competition is underperforming compared to intel, then they have a problem. Not intel’s problem.

People slot in 500W gpus in their desktops. I wouldn’t mind some extra power usage, if any. Least bothered. Looks like ARL for the win this generation.

 

[dullard]

Guess that currently Intel is devising about the exact power they ll use since at 250W they will have a win only in CB R20/R23 for MT, and a marginal one.

They may well set 297W as PL2, or use an extended 297W PL4 since that s their apps performance profile for ARL, otherwise they ll lose in too much benches to be convincing, so we ll surely be good for another round of exagerated TDPs.

Intel, Intel's customers (Dell, HP, Lenovo, etc), and suppliers such as motherboards and power supplies have a massive ecosystem based around certain power level configurations. The thought that Intel will still be contemplating power levels with less than 2 months until launch is laughable. The fundamental design decisions have already been locked in stone for quite a while. Do you actually expect that is Dell just going to dump a million power supplies (and coolers / airflow designs) or ASUS just going to widen circuit traces and add beefier power connectors now?

Intel Power Supply recommendations: https://edc.intel.com/content/www/u...y-design-guide/2.1a/processor-configurations/

The rumors are of a reduced PL2 and reduced PL4 (not to jump PL2 up to 297 W). https://wccftech.com/intel-arrow-la...w-pl1-177w-pl2-333w-pl4-24-core-unlocked-cpu/

 

[DokiDoki]

Maybe according to the troll CapeFrameX. All credible review websites put 7800X3D at the top of gaming charts.

X3D is only for putting out the longest bars in charts. In actual gameplay, using settings that you would normally use - that is native res with a high enough frame rate target - the inconsistency is noticeable.

AMDip is real.

I will test it out myself when Zen5 X3D launches. But for that I will need to wait for Big Boy Blackwell because the monitor I currently have is too OP for the 4090.

Dell UltraSharp 40 Curved Thunderbolt™ Hub Monitor - U4025QW | Dell India

Shop the UltraSharp 40 Curved Thunderbolt Hub Monitor - U4025QW featuring a stunning 5K display with a 120Hz refresh rate.

www.dell.com

 

[Abwx]

ARL has a two full node jump and is expected to be as efficient as competition. But the funny thing is, it doesn’t matter what power the cpu runs at. All that matters is what the stock power usage settings the manufacturer has set.

By ditching SMT they ll have to compensate with frequency by say 15%, wich increase power by 25 % at same perf compared to to a SMT equipped P cores, that eat in the power efficency jump by quite some margin, otherwise they would have done better than 250W in the Cinebench leak, likely 200W.

Intel, Intel's customers (Dell, HP, Lenovo, etc), and suppliers such as motherboards and power supplies have a massive ecosystem based around certain power level configurations. The thought that Intel will still be contemplating power levels with less than 2 months until launch is laughable. The fundamental design decisions have already been locked in stone for quite a while. Do you actually expect that is Dell just going to dump a million power supplies or ASUS just going to widen circuit traces and add beefier power connectors now?

Intel Power Supply recommendations: https://edc.intel.com/content/www/u...y-design-guide/2.1a/processor-configurations/

The rumors are of a reduced PL2 and reduced PL4 (not to jump PL2 up to 297 W). https://wccftech.com/intel-arrow-la...w-pl1-177w-pl2-333w-pl4-24-core-unlocked-cpu/

Big OEMs sell mainly mid and low range CPUs in their PCs, the higher end chips are not even a marginality in their DT sales, so whetever PSUs they have in inventories will be good for next gens.

Hnece Intel can set whatever power suit their competitive need since those CPUs cater only the retail market where margin are much better, wich justify the costs
to release such products, as for Intel s plans here their projections :

 

[Abwx]

X3D is only for putting out the longest bars in charts. In actual gameplay, using settings that you would normally use - that is native res with a high enough frame rate target - the inconsistency is noticeable.

AMDip is real.

I will test it out myself when Zen5 X3D launches. But for that I will need to wait for Big Boy Blackwell because the monitor I currently have is too OP for the 4090.

Dell UltraSharp 40 Curved Thunderbolt™ Hub Monitor - U4025QW | Dell India

Shop the UltraSharp 40 Curved Thunderbolt Hub Monitor - U4025QW featuring a stunning 5K display with a 120Hz refresh rate.

www.dell.com

So you dont have any AMD X3D CPU but are still stating that dips are real, guess that you brought us a pie of WCCFtech forums urban legends.

Check the minis here, both the 7950X3D and 7800X3D have the better minimum framerates with both the 4090 and the 7900XTX, that s all the point of an augmented cache if ever you didnt know.

Ryzen 9 9950X & 9900X: Gaming-Benchmarks: Spiele-Benchmarks auf zwei verschiedenen Grafikkarten

Ryzen 9 9950X & 9900X im Test: Spiele-Benchmarks auf zwei verschiedenen Grafikkarten / Gaming-Benchmarks mit einer GeForce RTX 4090

www.computerbase.de

 

[dullard]

Hnece Intel can set whatever power suit their competitive need since those CPUs cater only the retail market where margin are much better, wich justify the costs
to release such products, as for Intel s plans here their projections :

1) You have now said that (A) Intel is still deciding and (B) posted an image with fixed values. Which of your two posts are we to believe?

2) That image is just Jaykihn's estimations. Not an Intel sheet. We'll also put it in the rumor category.

3) None of Jaykihn's numbers have PL2 of 297 W.

 

[AcrosTinus]

Why do you say Golden Cove again tired? What does that mean? I mean Lion Cove is not Golden Cove its a 14% IPC uplift over even the 5% better IPC than GLC RPC is it not? So Lion Cove 19% IPC uplift over GLC?

Golden Cove again = Raptor Cove

 

[Abwx]

1) You have now said that (A) Intel is still deciding and (B) posted an image with fixed values. Which of your two posts are we to believe?

2) That image is just calculations. Not an Intel sheet. We'll also put it in the rumor category.

3) None of Jaykihn's numbers have PL2 of 297 W.

They have apps PL4 of 297W for the 8 + 16.

PL4 is not clear according to Igor s Lab, all it require is a short interrupt to get back to PL2 but this interrupt length is not really specified, it can be as low as a few dozen ms before the CPU can get back to the PL4 value, that s surely why the 253W PL2 specified 14900K/KS could get above 300W.

And btw in Computerbase 9950X review they did set the 14900K at PL1 = PL2 = 253W but in their Blender test it still managed to pull 275W for a long duration, so numbers are what they are, just indicative within a given error margin.

 

[Hitman928]

X3D is only for putting out the longest bars in charts. In actual gameplay, using settings that you would normally use - that is native res with a high enough frame rate target - the inconsistency is noticeable.

AMDip is real.

Trust me bro is not an argument.

 

[dullard]

They have apps PL4 of 297W for the 8 + 16.

PL4 is not clear according to Igor s Lab, all it require is a short interrupt to get back to PL2 but this interrupt length is not really specified, it can be as low as a few dozen ms before the CPU can get back to the PL4 value, that s surely why the 253W PL2 specified 14900K/KS could get above 300W.

And btw in Computerbase 9950X review they did set the 14900K at PL1 = PL2 = 253W but in their Blender test it still managed to pull 275W for a long duration, so numbers are what they are, just indicative within a given error margin.

1) You didn't answer question #1: are we to believe you that the numbers are changing or fixed?

2) I notice that you changed the goalpost from PL2 to PL4.

3) Lets look at that Computerbase 9950X review that you bring up. In the Blender test that you point out, the 14900K averaged 155.85 W with a peak of 274.79 W. So, to you that means intel is putting PL2 to 297 W? Really? https://www.computerbase.de/2024-08/amd-ryzen-9-9900x-9950x-test/5/

4) To prevent it from going above PL2 for very long, you have to set the optional PL3 value to be used. But, there is still a up to 10 ms spike allowed up to PL4. I notice that you call that 10 ms as "a long duration". https://edc.intel.com/content/www/u...heet-volume-1-of-2/001/package-power-control/

 

[Abwx]

3) Lets look at that Computerbase 9950X review that you bring up. In the Blender test that you point out, the 14900K averaged 155.85 W with a peak of 274.79 W. So, to you that means intel is putting PL2 to 297 W? Really? https://www.computerbase.de/2024-08/amd-ryzen-9-9900x-9950x-test/5/

The average is low because Blender has several cycles, at the start it is idling for 15s and then it goes at full throughput for 3 phases with some interrupts, the 275W are the phases of full throughput, that s not just short peaks as you seems to believe.

For the rest there will be reviews in less than two moths, and we ll surely have some leaks in the meantime, so we ll see how things pan out in the next weeks or so.