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.)

 

[dullard]

Though hard to believe MT would regress. I know Arrow Lake has no HT but isn't the e-cores Skymont supposed to be so much better than Gracemont that it should easily more than make up for it given that HT is not a real core. Like isn't Skymont supposed to have IPC and latency of Raptor Cove or maybe 2% better or is that just wishful thinking?

MT only regressed in that comparison because it was comparing a near top of the line Raptor Lake (8P-16E 13900k) to a mid-line Arrow Lake (8P-12E 265KF). The top of the line Arrow Lake would have higher frequencies and more E cores (8P-16E 285K). The move from 5.5 GHz to 5.7 GHz alone in the 285K would end up with higher MT scores even before adding in the 4 more E cores.

What needs to be stressed over and over again is that Arrow Lake is just a small performance boost over Raptor Lake. What you get instead is a significant power reduction, NPU, and double iGPU.

 

[ondma]

Not good to say the least.

That 265KF compared with my 13900K (Intel extreme profile and DDR5 @5600):

ST: 265KF 4.3% faster than 13900K
MT: 265KF 2.5% slower than 13900K

Gigabyte Technology Co., Ltd. Z890 AERO G vs Gigabyte Technology Co., Ltd. Z690 AORUS ELITE AX - Geekbench

Yea, but 265K is 8P/12E. 13900k is 8P/16E. You would not expect the 265K to match the 13900k in nT.
Edit: What Dullard said!!

 

[ondma]

MT only regressed in that comparison because it was comparing a near top of the line Raptor Lake (8P-16E 13900k) to a mid-line Arrow Lake (8P-12E 265KF). The top of the line Arrow Lake would have higher frequencies and more E cores (8P-16E 285K). The move from 5.5 GHz to 5.7 GHz alone in the 285K would end up with higher MT scores even before adding in the 4 more E cores.

What needs to be stressed over and over again is that Arrow Lake is just a small performance boost over Raptor Lake. What you get instead is a significant power reduction, NPU, and double iGPU.

Two out of 3 of which are pretty much meaningless on the desktop. Even the power reduction (and hopefully solving of the stability issues), is not really an advancement, just a partial remedy of a previous weak point. (I say "partial" because I think AMD will still be more power efficient.)
Too bad, really. With the less than stellar, to put it mildly, release of Zen 5, Intel had a chance to take a clear lead in consumer desktop, and even to perhaps beat the Zen 5 x3D parts in gaming. Sadly, it looks like they will just trade blows in most cases, and still trail x3D in gaming.

 

[dullard]

Two out of 3 of which are pretty much meaningless on the desktop. Even the power reduction (and hopefully solving of the stability issues), is not really an advancement, just a partial remedy of a previous weak point. (I say "partial" because I think AMD will still be more power efficient.)
Too bad, really. With the less than stellar, to put it mildly, release of Zen 5, Intel had a chance to take a clear lead in consumer desktop, and even to perhaps beat the Zen 5 x3D parts in gaming. Sadly, it looks like they will just trade blows in most cases, and still trail x3D in gaming.

I still think you are being too narrowly focused by ignoring the majority of users that use the iGPU. The NPU may or may not become anything. I happen to see lots of potential value in it--just depends on how the software side develops. And, yes power is eliminating a weakness. But, is that where are now, complaining that weaknesses are removed?

I don't think Intel ever claimed Arrow Lake was taking the performance crown away. That is supposed to happen with 18A. Going from behind to ahead in one generation was a stretch of the imagination. Arrow Lake Refresh might be good if it comes out with the often rumored (and then rumored to be cancelled) 32E core version (and since it should come out before Zen 6 which competes with Nova Lake).

 

[ondma]

I still think you are being too narrowly focused by ignoring the majority of users that use the iGPU. The NPU may or may not become anything. I happen to see lots of potential value in it--just depends on how the software side. And, yes power is eliminating a weakness. But, is that where are now, complaining that weaknesses are removed?

I don't think Intel ever claimed Arrow Lake was taking the performance crown away. That is supposed to happen with 18A. Going from behind to ahead in one generation was a stretch of the imagination. Arrow Lake Refresh might be good if it comes out with the often rumored (and then rumored to be cancelled) 32E core version (and since it should come out before Zen 6 which competes with Nova Lake).

Yea, most people use the igpu alone, but probably not on the top of the line desktop products. For general/office uses the igpu on RL was probably already "good enough". Improvement in igpu is great for low power mobile, like Lunar Lake, but for desktop, who cares. ARL-R will probably be a minimal improvement, since even Intel is calling it only a refresh. At one point, I thought the 8+32 rumored chip would be fantastic. It would be a nT monster, but not sure what the use case would be. I would actually be more interested in a 10 or 12 P, 0E configuration like the rumored Bartlett Lake, except on the latest node and architecture.

 

[Wolverine2349]

MT only regressed in that comparison because it was comparing a near top of the line Raptor Lake (8P-16E 13900k) to a mid-line Arrow Lake (8P-12E 265KF). The top of the line Arrow Lake would have higher frequencies and more E cores (8P-16E 285K). The move from 5.5 GHz to 5.7 GHz alone in the 285K would end up with higher MT scores even before adding in the 4 more E cores.

What needs to be stressed over and over again is that Arrow Lake is just a small performance boost over Raptor Lake. What you get instead is a significant power reduction, NPU, and double iGPU.

And stability I think so and it better be and hopefully and no degradation.

Though Skymont e-cores cores, will they perform with the same low latency Raptor Cove cores perform at clock normalized even though they have same IPC or 2% better supposedly?

 

[SiliconFly]

... it should easily more than make up for it given that HT is not a real core. ...

I don't think it can make up HT that easily due to two main reasons. One is loss of HT & two is clock regression. Overall HT improvement shouldn't be that great over RPL.

 

[MarkPost]

MT only regressed in that comparison because it was comparing a near top of the line Raptor Lake (8P-16E 13900k) to a mid-line Arrow Lake (8P-12E 265KF). The top of the line Arrow Lake would have higher frequencies and more E cores (8P-16E 285K). The move from 5.5 GHz to 5.7 GHz alone in the 285K would end up with higher MT scores even before adding in the 4 more E cores.

What needs to be stressed over and over again is that Arrow Lake is just a small performance boost over Raptor Lake. What you get instead is a significant power reduction, NPU, and double iGPU.

Yea, but 265K is 8P/12E. 13900k is 8P/16E. You would not expect the 265K to match the 13900k in nT.
Edit: What Dullard said!!

I've disabled four e-cores to equal core count, but yet 265KF would be a little slower in MT than a simulated "14700K":

Gigabyte Technology Co., Ltd. Z890 AERO G vs Gigabyte Technology Co., Ltd. Z690 AORUS ELITE AX - Geekbench

 

[dullard]

Though Skymont e-cores cores, will they perform with the same low latency Raptor Cove cores perform at clock normalized even though they have same IPC or 2% better supposedly?

I haven't seen much on latency yet. This is about all that I've seen: https://www.techpowerup.com/324033/...of-p-cores-and-e-cores-along-the-ringbus?cp=2

 

[dullard]

I've disabled four e-cores to equal core count, but yet 265KF would be a little slower in MT than a simulated "14700K":

Gigabyte Technology Co., Ltd. Z890 AERO G vs Gigabyte Technology Co., Ltd. Z690 AORUS ELITE AX - Geekbench

Did you also drop the frequency to 5.5 GHz when simulating?

 

[MarkPost]

Did you also drop the frequency to 5.5 GHz when simulating?

No, I'll try it

 

[MarkPost]

ok, done:

"processor_frequency": {
        "frequencies": [
            5481,
            5489,
            5491,
            5487,
            5484,
            5489,
            5491,
            5490,
            5491,
            5489,
            5490,
            5493,
            5493,
            5489,
            5492,
            5490,
            5491,
            5491,
            5474,
            5476,
            5481,
            5482,
            5483,
            5481,
            5483,
            5481,
            5483,
            5481,
            5484,
            5480,
            5483,
            5483,
            5480,
            5483,
            5483,
            5485,
            5483,
            5481,
            5483,
            5478,
            5485,
            5481,
            5485,
            5483,
            5483,
            5482,
            5480,
            5483,
            5485,
            5483,
            5485,
            5480,
            5485,
            5485,
            5485
        ]
    },

So, equal frecuency (5.5) and equal core count (8P - 12E):

ST: ARL 7.8% faster than RPL
MT: ARL 1.4% slower than RPL

Gigabyte Technology Co., Ltd. Z890 AERO G vs Gigabyte Technology Co., Ltd. Z690 AORUS ELITE AX - Geekbench

 

[dullard]

ok, done:

So, equal frecuency (5.5) and equal core count (8P - 12E):

ST: ARL 7.8% faster than RPL
MT: ARL 1.4% slower than RPL

Gigabyte Technology Co., Ltd. Z890 AERO G vs Gigabyte Technology Co., Ltd. Z690 AORUS ELITE AX - Geekbench

Thanks for doing all that.

I'm looking into where the performance drops the most. Looks like the worst Arrow Lake performance is "Background Blur". Ideally that would be done by the NPU. Next comes HTML5 rendering of 32 simultaneous pages. Any idea on that, lack of HT? Then comes photo filter (color and blur changes) which again ideally would be the NPU's task.

 

[MoistOintment]

Yea, most people use the igpu alone, but probably not on the top of the line desktop products. For general/office uses the igpu on RL was probably already "good enough". Improvement in igpu is great for low power mobile, like Lunar Lake, but for desktop, who cares. ARL-R will probably be a minimal improvement, since even Intel is calling it only a refresh. At one point, I thought the 8+32 rumored chip would be fantastic. It would be a nT monster, but not sure what the use case would be. I would actually be more interested in a 10 or 12 P, 0E configuration like the rumored Bartlett Lake, except on the latest node and architecture.

Issue is that Intel isn't designing ARL around how it'll look on the top of the line desktop part. ARL will be a pretty decent improvement in most of the product stack. The 285K vs 14900K is the least exciting generational change as that's where the max clock speed regression is the most pronounced.

 

[gdansk]

Issue is that Intel isn't designing ARL around how it'll look on the top of the line desktop part. ARL will be a pretty decent improvement in most of the product stack. The 285K vs 14900K is the least exciting generational change as that's where the max clock speed regression is the most pronounced.

Given the smaller performance delta between top most part and the lower ones, will Intel reduce the top MSRP or bring the midrange parts up?

 

[Hitman928]

Thanks for doing all that.

I'm looking into where the performance drops the most. Looks like the worst Arrow Lake performance is "Background Blur". Ideally that would be done by the NPU. Next comes HTML5 rendering of 32 simultaneous pages. Any idea on that, lack of HT? Then comes photo filter (color and blur changes) which again ideally would be the NPU's task.

Memory bandwidth and timings come into play a lot more on the MT test, though not as much as with GB5. Could also be a power limit difference limiting the frequency a little bit with ARL compared to the "13700K". The number of E-cores won't really matter for GB6 tests when you have 8 P-cores.

 

[ondma]

Issue is that Intel isn't designing ARL around how it'll look on the top of the line desktop part. ARL will be a pretty decent improvement in most of the product stack. The 285K vs 14900K is the least exciting generational change as that's where the max clock speed regression is the most pronounced.

What did they design it around then? The mid/low end may show less frequency regression (unless Intel purposely clocks them lower so they can "refresh" with higher clocks), but the lack of HT is going to hurt the nT performance of the low end of the product stack more than the high end, since the mid/low end products have fewer (or no) E cores.

 

[dullard]

What did they design it around then? The lack of HT is going to hurt the low end of the product stack more than the high end, since the mid/low end products have fewer (or no) E cores.

Which Arrow Lake-S chips have no E cores?

I don't know the final clock speeds of the cores, so it is hard to estimate. But wouldn't 6 P cores doing worse without HT be outweighed by the 8 E cores doing better?

 

[MoistOintment]

What did they design it around then? The lack of HT is going to hurt the low end of the product stack more than the high end, since the mid/low end products have fewer (or no) E cores.

The market shift towards dropping desktops has only accelerated since COVID. Last I checked, the desktop was about 1/3 of the market. If certain changes like dropping HT to improve PPA can help their laptop line but hurts their high end desktop, so be it.

Latest rumors are putting Core Ultra 3s at 4+4 configs. Most of the ARL laptops are gonna be 4+8's and 6+8's. Full load MT isn't a common task and even though ARL sees hardly any MT improvements at all vs RPL, it'll still be ahead of most of Zen 5 (comparing comparable parts).

And the clock speed regression's impact is Intel painting themselves into a corner with having to revise 10nm then Intel 7 over multiple generations to try and squeeze every less Mhz out of it. There's no new nodes coming out that'll match what Intel 7 clocks to. Not from Intel or TSMC.

 

[ondma]

Which Arrow Lake-S chips have no E cores? I don't know the final clock speeds of the cores, so it is hard to estimate. But wouldn't 6 P cores doing worse without HT be outweighed by the 8 E cores doing better?

The Ultra 5 240 has only 4 E cores. I dont know if the improvement in the 4E cores will be enough to compensate for the loss of HT on 6 P cores. I would assume eventually there will be an i3 equivalent (Ultra 3??) with no E cores.

 

[jdubs03]

One thing to emphasize about this early result is that it is just one sample.

I just did a search of the i7-14700KF, and there are single-core values over 3300 and multicore over 22000. The average is of all results is 3004/19583. So yeah good bit of variance there which I imagine will be the case again with XMP/memory settings, etc.

Based on that image I reposted yesterday, the goal for Geekbench (v5.4.5) multicore is a gain of 17% over the 14th gen. These results show at best no improvement and almost a regression.

So for now lots of salt should be taken. When/if ARL releases and doesn’t show anywhere near at least 12% uplift, that’ll be a disappointment; really since I’m seeing 17% that’s what I expect. Single thread as was mentioned above seems right around 8% there and matches the bar graph.

 

[OriAr]

The Ultra 5 240 has only 4 E cores. I dont know if the improvement in the 4E cores will be enough to compensate for the loss of HT on 6 P cores. I would assume eventually there will be an i3 equivalent (Ultra 3??) with no E cores.

Core Ultra 3 is expected to have 4P + 4E cores.
Should be a huge upgrade in MT perf over current gen i3.

 

[ondma]

Core Ultra 3 is expected to have 4P + 4E cores.
Should be a huge upgrade in MT perf over current gen i3.

For highly parallel software that can use the E cores, sure. As a low budget gaming chip, it will be interesting to see how it holds up vs 4HT cores. I am still not convinced gaming can utilize the E cores effectively. Overall, I am still skeptical about removing HT, especially since removing it didnt lead to any larger gains in 1T performance than we saw for AL.

 

[jdubs03]

Below post references the QS data that we’ve been talking about past few weeks. He followed up with a post saying he thinks there will be 3% single core uplift (in GB6).

https://twitter.com/x/status/1824268748456595594

 

[coercitiv]

For highly parallel software that can use the E cores, sure.

Even for updating Windows, on some systems even for booting Windows. Browsers will also be quite happy about those extra cores.

As a low budget gaming chip, it will be interesting to see how it holds up vs 4HT cores. I am still not convinced gaming can utilize the E cores effectively.

For low budget chips, even in gaming, 4 Skymont cores will improve 60FPS gaming considerably. In fact if the Ultra 3 receives just a 5% speed bump, combined with the IPC uplift and L3 caches changes, will probably beat 14100 in gaming using P cores alone.