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.)

 

[poke01]

This is very impressive. Very. Intel is back people. Arrow Lake should score much higher in single.

Intel Corporation Lunar Lake Client Platform - Geekbench

Benchmark results for an Intel Corporation Lunar Lake Client Platform with an Intel Core Ultra 7 268V 2.20GHz processor.

browser.geekbench.com

 

[poke01]

Lunar Lake GB6 results. Never mind MT scores, but ST scores are ... nothing to be exciting about. I expected a lot more: I expected way over 3000.

https://browser.geekbench.com/v6/cpu/6678431 (ST: 2739)

https://browser.geekbench.com/v6/cpu/6678564 (ST: 2713)

That's slower than Ryzen 9 7945HX in ST mode. Darn. Could be engineering samples but clock speeds (add .gb6 to the URL) are perfectly fine.

I very much disagree. The 7945HX needs to boost to 5.3GHz to achieve 2700.

Micro Computer (HK) Tech Limited Uranus Series - Geekbench

Benchmark results for a Micro Computer (HK) Tech Limited Uranus Series with an AMD Ryzen 9 7945HX processor.

browser.geekbench.com

Lunar achieves 2700 using 4.8GHz, this is not poor relative to AMD.
I would also bet that Lunar’s lion cove is also not as power hungry in 1t.

What we see here M2 like performance and efficiency but for the x86 platform with eGPU support a much better software compatibility. This is good. The GPU should be good as well.

 

[whoshere]

@poke01

I've completely rewritten my message. Yeah, it's not that bad but it's not terribly exciting either. Alder Lake is more than 2 years old at this point, I expected more from a new uArch.

 

[poke01]

@poke01

I've completely rewritten my message. Yeah, it's not that bad but it's not terribly exciting either. Alder Lake is more than 2 years old at this point, I expected more from a new uArch.

At this point I’m glad Intel is focusing on efficiency but you are right in saying they need major leaps in IPC in future generations.

 

[itsmydamnation]

I was only answering your claim about SPEC. You didn't mention games in your previous post 😉

i wasn't disagreeing , just an in addition

 

[Hitman928]

Lunar Lake GB6 results running at roughly 4.9GHz:

https://browser.geekbench.com/v6/cpu/6678431 (ST: 2739)

https://browser.geekbench.com/v6/cpu/6678564 (ST: 2713)

An IPC increase is there but I expected more.

At 6GHz that would be 3,353 which is just 8% faster than 14900K Raptor Lake. Not really exciting but then it's a mobile platform, so Arrow Lake should fare better.

The clocks don't appear to be very stable so it's hard to say for sure, but it appears to be ~10% higher performance per clock than MTL in GB6.

 

[Hitman928]

I very much disagree. The 7945HX needs to boost to 5.3GHz to achieve 2700.

Micro Computer (HK) Tech Limited Uranus Series - Geekbench

Benchmark results for a Micro Computer (HK) Tech Limited Uranus Series with an AMD Ryzen 9 7945HX processor.

browser.geekbench.com

Lunar achieves 2700 using 4.8GHz, this is not poor relative to AMD.
I would also bet that Lunar’s lion cove is also not as power hungry in 1t.

What we see here M2 like performance and efficiency but for the x86 platform with eGPU support a much better software compatibility. This is good. The GPU should be good as well.

LNL won't be competing against Zen 4, it's Zen 5 that it will be compared with.

I don't think they'll be competing with M2 in efficiency either, at least not single and low threaded situations. LNL will still take way more power to get clocks up to ~5 GHz.

 

[DavidC1]

At 6GHz that would be 3,353 which is just 8% faster than 14900K Raptor Lake. Not really exciting but then it's a mobile platform, so Arrow Lake should fare better.

You can't compare an ultra power efficiency optimized platform with similar RAM, and Drive configurations with a essentially unlimited power on desktop. Generally the desktop chips are 5-7% faster than laptop chips of the same uarch.

I don't think they'll be competing with M2 in efficiency either, at least not single and low threaded situations. LNL will still take way more power to get clocks up to ~5 GHz.

Battery life, which is defined as bursty workloads is the bigger deal. Getting that will be a big thing, as currently they are very far behind.

 

[Ghostsonplanets]

Intel Core Ultra 7 268V "Lunar Lake" CPU appears on Geekbench - VideoCardz.com

Intel Core Ultra 200V in first synthetic test First Lunar Lake processor on Geekbench. Intel is set to launch its first Core Ultra 200 CPU series as early as September. The initial series will focus on low-power mobile devices and is expected to be certified as Copilot+PCs, owing to its powerful...

videocardz.com

Good scores for Lunar Lake. And that's the 17W SKU rather than the Core Ultra 9 30W.

 

[Gideon]

Sigh, I really wish Geekbench 6 would have kept the Crypto, FP and Int total scores. The perf per clock is so similar to Zen 5 mobile at similar GHz (which also isn't quite the desktop Zen 5 as it has halved it's FP structures and L3 cache), that separating these score would give a much more thorough overview.

I know this separation can be done manually but i'm too lazy. And the clocks (and reporting) are still all over the place as well. A bit too much to draw any conclusions.

Still I hope Arrow Lake has some surprises in store IPC-wise. Lunar Lake looks to be an excellent SKU overall but the IPC uplift from Raptor Cove (to me) looks a tad dissapointing

~4.8 Ghz 1360P vs ~4.8Ghz Lunar Lake

Intel Corporation Lunar Lake Client Platform vs SAMSUNG ELECTRONICS CO., LTD. 960QFG - Geekbench

Well actually 1360P clocked slightly slower:

Lunar Lake:
Mean: 4825.04 MHz
Median: 4874.0 MHz

Core i7-1360P:
Mean: 4723.42 MHz
Median: 4824.0 MHz

The ST difference is 12,2% but taking the mean clock speed into account it's only 9.8%

Then again, comparing to a similarily clocked Meteor Lake U, taking the "chiplet tax" into equation it's actually quite good:

Intel Corporation Lunar Lake Client Platform vs LENOVO 21ML0055US - Geekbench

 

[AcrosTinus]

Sigh, I really wish Geekbench 6 would have kept the Crypto, FP and Int total scores. The perf per clock is so similar to Zen 5 mobile at similar GHz (which also isn't quite the desktop Zen 5 as it has halved it's FP structures and L3 cache), that separating these score would give a much more thorough overview.

I know this separation can be done manually but i'm too lazy. And the clocks (and reporting) are still all over the place as well. A bit too much to draw any conclusions.

Still I hope Arrow Lake has some surprises in store IPC-wise. Lunar Lake looks to be an excellent SKU overall but the IPC uplift from Raptor Cove (to me) looks a tad dissapointing

~4.8 Ghz 1360P vs ~4.8Ghz Lunar Lake

Intel Corporation Lunar Lake Client Platform vs SAMSUNG ELECTRONICS CO., LTD. 960QFG - Geekbench

Well actually 1360P clocked slightly slower:

Lunar Lake:
Mean: 4825.04 MHz
Median: 4874.0 MHz

Core i7-1360P:
Mean: 4723.42 MHz
Median: 4824.0 MHz

The ST difference is 12,2% but taking the mean clock speed into account it's only 9.8%

Then again, comparing to a similarily clocked Meteor Lake U, taking the "chiplet tax" into equation it's actually quite good:

Intel Corporation Lunar Lake Client Platform vs LENOVO 21ML0055US - Geekbench

I don't trust geekbench one single bit. I have two systems A: 13700K and B:14900K, both score almost the same -> A: 3000ST and 20000MT and B:3100 and 21700MT. This cannot be this close...

 

[DavidC1]

Then again, comparing to a similarily clocked Meteor Lake U, taking the "chiplet tax" into equation it's actually quite good:

Intel Corporation Lunar Lake Client Platform vs LENOVO 21ML0055US - Geekbench

The faster 165U results are much closer than that. There's something wrong with that result.

With Geekbench you need to take top scores for that very reason.

Microsoft Corporation Surface Pro 10 for Business - Geekbench

Benchmark results for a Microsoft Corporation Surface Pro 10 for Business with an Intel Core Ultra 7 165U processor.

browser.geekbench.com

Geekbench did two fatal mistakes, one not separating the Int/FP results and the other taking out the ability to sort by lowest/highest scores.

User-submitted results like GB is the worst way to isolate pure uarch differences when we care about 1%.

 

[dullard]

Yea, we never heard anybody speculating how great it would be to remove HT until it came out that newer Intel designs would not have it. Now it is being proclaimed as the greatest thing since sliced bread. A classical case of rationalization, IMO.

It is nice to see how things have come full circle now. 2 decades ago the argument was all about how hyperthreading REDUCED performance on many programs. Then it slowly got better, to the point that you no longer had to turn off hyperthreading at it was on average roughly the same performance ( https://www.anandtech.com/show/1031/6 ). You would see some software improve and some get worse at that point in time. Then as software got more multithreaded, hyperthreading got better and better. But now as CPUs get more cores, the hyperthreading gets worse and worse.

I do like how you claim that no one speculated on removing HT, when it is easy to prove otherwise. From 2021:

30% boost from Hyperthreading is only in the very best case scenario. Lots of uses see closer to a 10% to 20% gain, some uses have performance losses with hyperthreading turned on (HPC for a notable example where benchmarks run weeks at a time, not seconds). Hyperthreading adds a lot of power and heat, but it gives more threads. So there is a balance between needing to throttle down frequencies vs having more threads running. On a whole, most software gets a small boost. But, only a few programs get 30% boost.

From 2018:

At least they did one thing right, which was to turn off hyperthreading which often destroys CFD performance.

From 2017:

The 7600 was quite an under-appreciated value in chips, especially for those of us that don't get much if any benefit from hyperthreading (meaning the i7 line isn't a good choice).

You could argue that I didn't clearly say "remove HT". I was arguing that HT isn't that great in many cases though.

 

[Hitman928]

Battery life, which is defined as bursty workloads is the bigger deal. Getting that will be a big thing, as currently they are very far behind.

Battery life is also very dependent on SOC/fabric power. I don't expect Intel to catch up with M2's efficiency in either of these aspects. I do think they'll do much better than their current offerings, but M2 is just so far ahead on these that I don't see LNL closing the gap completely.

 

[Henry swagger]

Intel Core Ultra 7 268V "Lunar Lake" CPU Benchmarks Leak: Up To 20% Faster Single-Thread Uplift Versus Meteor Lake

Benchmarks of Intel's upcoming Lunar Lake CPU, the Core Ultra 7 268V, have leaked out within Geekbench with a clock speed of up to 4.9 GHz.

wccftech.com

20% plus over meteor lake at 17 watts.. arrow lake hx should be even better 🥇

 

[Hitman928]

Intel Core Ultra 7 268V "Lunar Lake" CPU Benchmarks Leak: Up To 20% Faster Single-Thread Uplift Versus Meteor Lake

Benchmarks of Intel's upcoming Lunar Lake CPU, the Core Ultra 7 268V, have leaked out within Geekbench with a clock speed of up to 4.9 GHz.

wccftech.com

20% plus over meteor lake at 17 watts.. arrow lake hx should be even better 🥇

That's just WFFCtech doing WFFCtech things. Their 185H score is not accurate. That LNL result is only 3% higher than the 185H when it's holding its boost clock. The LNL is clocking lower, so we are again back to LNL being ~10% faster per clock than MTL (actually 7.5% in this case).

Dell Inc. Precision 5690 - Geekbench

Benchmark results for a Dell Inc. Precision 5690 with an Intel Core Ultra 9 185H processor.

browser.geekbench.com

 

[Abwx]

Just like Intel demonstrated, they have 115% single-thread performance at same power with their single-thread optimzed core as SMT-core running one thread.

That s not 15% better ST perf but 15% better perf/watt/area, good luck dealing with obscured metrics.

Page 375 - Discussion - Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads

Page 375 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[Gideon]

That's just WFFCtech doing WFFCtech things. Their 185H score is not accurate. That LNL result is only 3% higher than the 185H when it's holding its boost clock. The LNL is clocking lower, so we are again back to LNL being ~10% faster per clock than MTL (actually 7.5% in this case).

Dell Inc. Precision 5690 - Geekbench

Benchmark results for a Dell Inc. Precision 5690 with an Intel Core Ultra 9 185H processor.

browser.geekbench.com

That's a linux score (which runs better than windows and seems too high)

But I do agree the WFFC sample is too slow. This one is more representative:

Microsoft Corporation Surface Pro 10 for Business - Geekbench

Benchmark results for a Microsoft Corporation Surface Pro 10 for Business with an Intel Core Ultra 7 165U processor.

browser.geekbench.com

 

[Hitman928]

That's a linux score (which runs better than windows and seems too high)

But I do agree the WFFC sample is too slow. This one is more representative:

Microsoft Corporation Surface Pro 10 for Business - Geekbench

Benchmark results for a Microsoft Corporation Surface Pro 10 for Business with an Intel Core Ultra 7 165U processor.

browser.geekbench.com

That's a 165U, not a 185H. Intel CPUs typically don't benefit from Linux like AMD chips do. Maybe a slight bump. There are plenty of Windows examples scoring very nearly the same. I chose the one I did because it actually maintained its boost clock which made it easier to compare in terms of ppc. We can pick a Windows example though and calculate the average clocks and compare.

Acer Predator PHN14-51 - Geekbench

Benchmark results for an Acer Predator PHN14-51 with an Intel Core Ultra 9 185H processor.

browser.geekbench.com

This would put LNL at 9% faster clock for clock when taking the average reported clocks. Conclusion is still the same (caveats of the LNL still being pre-release, GB clock reporting errors, and variability in testing still apply).

 

[DavidC1]

You could argue that I didn't clearly say "remove HT". I was arguing that HT isn't that great in many cases though.

Not having it on client but on server makes lot more sense. With Skymont even more of the MT performance comes from the E cores, meaning taking out HT on just the 8 P cores is even a smaller impact.

And what happens in the future when the gap closes even further? Let's say the P cores lose the 10-12% per clock advantage?

 

[Nothingness]

i wasn't disagreeing , just an in addition

And one that I certainly agree with

 

[Wolverine2349]

Beating a X3D part is gonna be a bit difficult I think. But if ARL top part even comes close to a X3D, I'd consider it a big achievement. At best, we shouldn't expect more than a marginal win. For all practical purposes, they're gonna be more or less equal this year I think.

Beating any X3D part even 8 core one 9800X3D? What about games that scale to more than 8 cores well? Are there any such games?

Cause Intel has good core to core latency beyond 8. AMD beyond 8 dual CCX/CCD and bad latency penalty.

Or do no games have any benefit from more than 8 cores and will not for a long time thus 9800X3D will be all that is needed and best?

 

[SarahKerrigan]

Spec dot org has results for the E1050 96 core. Just do a search on google for spec results for ibm power processors and it's one of the first entries.

To give you some perspective:
IBM Power E1050 (2.95 - 3.90 GHz, 96 core, AIX) - [768 instances, 4 chips, SMT8, only 2 threads per core in use) SpecRate2017IntBase=1220 , Peak=1580
IBM Power E1080 (3.55 - 4 GHz, 120 core, AIX) - [120 cores, 8 chips, 960 instances] = SpecRate2017IntBase=1700 , Peak=2170
Kaytus KR1280E2 (AMD EPYC 9754) (256 cores/512 instances) - SpecRate2017IntBase=1930 , Peak=2100
Kaytus KR1280E2 (AMD EPYC 9654) (192 cores/384 instances) - SpecRate2017IntBase=1800 , Peak=1920
Dell PowerEdge R6625 (AMD EPYC 9754 128-Core Processor)- (256 cores/512 instances) = SpecRate2017IntBase=1820 , Peak=1980
Dell PowerEdge R6625 (AMD EPYC 9684X 96-Core Processor) - (192 cores/384 instances) = SpecRate2017IntBase=1790 , Peak=1870
ZTE R8500G5 Server System (1.90 GHz, Intel Xeon Platinum 8490H) - (240 cores/480 instances) = SpecRate2017IntBase=2000 , Peak=2050
HPE Compute Scale-up Server 3200 (1.90 GHz, Intel Xeon Platinum 8490H) - (960 cores/1920 instances) - SpecRate2017IntBase=7310 , Peak=[Not submitted]

IBM Power10 can be configured in a performance competitive manner, but, it doesn't look like they are competitive on a "per-core-throughput" basis.

Sure they are, if you buy IBM's definition of "core", which you shouldn't. (An "SMT8 core" is two cores.)

If one believes in "SMT8 cores" there's a solid advantage over everything else on that list, as best as I can tell. (That being said, the real app for large Power is scale-up HANA, at which it does... fine.)

As already mentioned, Power is not Z. Z single-thread performance is drastically higher.

 

[Gideon]

That's a 165U, not a 185H. Intel CPUs typically don't benefit from Linux like AMD chips do

Yeah, my bad you compared absolute perf (as did WFFC)

I chose 165U / 1360P particularily as they clock very close to the Lunar Lake sample (on average) and are lower TDP parts.

 

[coercitiv]

Cause Intel has good core to core latency beyond 8.

Who wants to take this one?