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.)

 

[Hulk]

My experience running Photoshop, Digital video and audio software, vector illustration, Topaz video/photo ai, Handbrake, ... the rest is easy on the CPU (Office, Chrome, etc..)
Given strong P cores and enough E's, HT isn't a benefit to most people outside of DC and benchmarking.
Most software only uses (heavily) around 6 to 8 cores and of those usually 2 or 3 are the ones carrying most of the load from what I see in HWinfo.
A lot of compute intensive applications harness the GPU pretty well.

For these reasons I have HT off on my 14900K. Run the P's and E's at 5.5/4.3 and it's always under 200W, generally under 150W for most tasks. It runs 70C max on air and around 1.1V under load. Except for Cinebench performance is no better with HT on but heat and voltage are much higher.

If the P's and E's in Arrow Lake are +20% over Raptor/Gracemont IPC-wise and the efficiency is better I'll be in for an upgrade. 15% might even do it for me assuming it can hit 5.3/4.0 all-core.

But then again over the years I've turned into an upgrade junkie. I kept my Haswell for like 7 years and then over the course of 3 ripped through a 12700K, 13900K (degradation warranty return), 13600K, and now 14900K. It's too much fun to upgrade and I have the disposable income to fool around a bit more than years past where I would eat beans and coos coos every night because it was cheap and keep the heat on 60 F.

 

[Thunder 57]

...Run the P's and E's at 5.5/4.3 and it's always under 200W, generally under 150W for most tasks. It runs 70C max on air and around 1.1V under load...

It's amazing what RPL can do when using sensible settings. Instead Intel decided to clock it to the moon to keep up with AMD.

 

[Ghostsonplanets]

It's amazing what RPL can do when using sensible settings. Instead Intel decided to clock it to the moon to keep up with AMD.

I still remember when Geekerwan underclocked and undervolted an i9 12900K while disabling 2P cores and locking it to 35W. It could still compete with M1 Pro. Everyone was so amazed and hopeful that ADL Mobile would be very efficient...But alas.

 

[Hulk]

It's amazing what RPL can do when using sensible settings. Instead Intel decided to clock it to the moon to keep up with AMD.

If I wanted to really work I could put a custom loop on it, maybe even a chiller and get maybe 8% more performance for like $500 and a ton of electricity use.
Not worth it for me and it doesn't sound like fun to buy or build.
5.5/4.3 is easy for the 14900K at under 1.1V under load with HT off.

Intel did what they had to do in order to post benchmarks. The non K parts and even the K parts in pre-built systems are set up to run in a sane manner. Most people buying K parts I think know how to set them up so I personally like Intel selling me the chip with specs indicating what it can do under extreme cooling and power. The user can back that off to what is reasonable for them.

 

[Ghostsonplanets]

This is certainly impressive:

https://twitter.com/x/status/1781029327980675237

Also, didn't knew that Xe LPG had support for DP4a. In hindsight it does need it due to XeSS, but I thought it was done on the XMX units on Xe 12.7.

 

[Hitman928]

This is certainly impressive:

https://twitter.com/x/status/1781029327980675237

Also, didn't knew that Xe LPG had support for DP4a. In hindsight it does need it due to XeSS, but I thought it was done on the XMX units on Xe 12.7.

Eh.

https://twitter.com/x/status/1781073943396315400

Also,

https://twitter.com/Tobias_Writes/status/1781054691704439027

 

[coercitiv]

Also,

https://twitter.com/Tobias_Writes/status/1781054691704439027

When joining Intel, Robert Hallock received a rear view mirror and a backpedal:

 

[poke01]

None of Intels current lineup is even close to Apple in performance per watt.

 

[coercitiv]

None of Intels current lineup is even close to Apple in performance per watt.

That's why the entire exchange is funny, it quickly went from "exclusive LLM functionat enablement on day 0" to an unfavorable comparison against a passively cooled, 3+ year old product from the competition.

It's the marketing bug I guess, that "exclusive" was too hard to resist. In Robert's partial defense though, it's also the AI bug, people can't help saying things without checking when it comes to generative AI claims.

 

[Markfw]

None of Intels current lineup is even close to Apple in performance per watt.

Or AMDs

 

[SiliconFly]

Or AMDs

Yep. Even none of AMDs current lineup is even close to Apple in performance per watt.

 

[Gideon]

But then again over the years I've turned into an upgrade junkie. I kept my Haswell for like 7 years and then over the course of 3 ripped through a 12700K, 13900K (degradation warranty return), 13600K, and now 14900K. It's too much fun to upgrade and I have the disposable income to fool around a bit more than years past where I would eat beans and coos coos every night because it was cheap and keep the heat on 60 F.

OT, but looking at your sig, and the taxing applications you listed, I'd say the biggest upgrade for you would be going for low-latency DDR5 6400 MT/s (or higher if your mem-controller can handle it).

Just going from 4400 MT/s DDR5 to 6400 DDR5 makes a notable difference in video editing and photoshop:

Impact of DDR5 Speed on Content Creation Performance (2023 update)

DDR5 memory has kits rated for up to 8400 Mbps, while desktop CPUs only officially support up to 5600 Mbps. How much does running at official specifications actually impact performance in common content creation applications?

www.pugetsystems.com

Let alone moving from DDR4 to DDR4 (TBF, they were using some horrific JEDEC timings for 3200 Mhz CL22) :

12th Gen Intel Core CPU Review Roundup

Intel has launched their new 12th Gen Intel Core desktop processors (code-named "Alder Lake") featuring support for DDR5, PCIe 5.0, as well as a completely new hybrid architecture using a mix of Performance and Efficient-cores. This is a lot of new technologies in one product which makes it hard...

www.pugetsystems.com

Those apps tend to like more memory bandwidth and the way DDR5 banks are set up, it is able to deliver more real-world bandwidth even at the same MT/s.

It would obviously make the most sense to time it with the Arrow Lake upgrade (as it's mem-controller hopefully is also improved).

I'm actually quite envious at the prices in the US for 64GB DDR5 kits, with excellent latency and speed that will run on most 12-14th gen CPUs:

https://www.newegg.com/g-skill-64gb/p/N82E16820374456

 

[Ghostsonplanets]

Our beloved Intel Core Ultra 5 115U makes another appearance(s). We have 2P Redwood Cove + 4E Crestmont + 2LPE Cresmont, for a total of 8C/10T. On the GPU side, it mark the first appearance for Intel Arc Xe LPG 48EU.

Basically, 115U is the binned/reject version for the MTL-U, akin to the role of i3 1215U on ADL-U line-up. The CPU side sees an entire 4C Cluster of Crestmont disabled. And the GPU sees an Xe Core (16 EUs) disabled within the Render Slice.

https://twitter.com/x/status/1781235319318180249

https://twitter.com/x/status/1781233859356443124

Also, for fun:

1C/1T@1GHz. Reject multicore, embrace single core😁

 

[Khato]

Basically, 115U is the binned/reject version for the MTL-U, akin to the role of i3 1215U on ADL-U line-up. The CPU side sees an entire 4C Cluster of Crestmont disabled. And the GPU sees an Xe Core (16 EUs) disabled within the Render Slice.

While I won't claim that there couldn't be some binned chips, almost all U series chips, including the 115U, are going to be using the dedicated 2P/8E and 64EU tiles rather than cut-down 6P/8E and 128EU tiles. Basically the exact same as the U series on ADL/RPL have a separate die.

Also, if anyone wants a full review of a 165U such is now available on notebookcheck - https://www.notebookcheck.net/Core-...ell-Latitude-9450-2-in-1-review.825254.0.html Pretty much as expected - slightly lower multi-thread performance, about 2/3 GPU performance, and notably better battery life. It is unfortunate that Intel didn't include the full GPU tile on any of the U series SKUs, but I guess we'll effectively be getting that with LNL.

 

[Ghostsonplanets]

While I won't claim that there couldn't be some binned chips, almost all U series chips, including the 115U, are going to be using the dedicated 2P/8E and 64EU tiles rather than cut-down 6P/8E and 128EU tiles. Basically the exact same as the U series on ADL/RPL have a separate die.

Aye. Hence why I said the 115U sees an 4C CRT cluster disabled and a Xe Core (16 EUs) out of 4/1RS.

Also, if anyone wants a full review of a 165U such is now available on notebookcheck - https://www.notebookcheck.net/Core-...ell-Latitude-9450-2-in-1-review.825254.0.html Pretty much as expected - slightly lower multi-thread performance, about 2/3 GPU performance, and notably better battery life. It is unfortunate that Intel didn't include the full GPU tile on any of the U series SKUs, but I guess we'll effectively be getting that with LNL.

Much thanks for the link! Was waiting for the review since they had added some results in the database some days ago.

 

[Hitman928]

While I won't claim that there couldn't be some binned chips, almost all U series chips, including the 115U, are going to be using the dedicated 2P/8E and 64EU tiles rather than cut-down 6P/8E and 128EU tiles. Basically the exact same as the U series on ADL/RPL have a separate die.

Also, if anyone wants a full review of a 165U such is now available on notebookcheck - https://www.notebookcheck.net/Core-...ell-Latitude-9450-2-in-1-review.825254.0.html Pretty much as expected - slightly lower multi-thread performance, about 2/3 GPU performance, and notably better battery life. It is unfortunate that Intel didn't include the full GPU tile on any of the U series SKUs, but I guess we'll effectively be getting that with LNL.

Better battery life (outside of load situation) is probably mostly due to the lower power screen being used compared to the previously reviewed MTL laptops with OLED screens.

 

[uzzi38]

This is certainly impressive:

https://twitter.com/x/status/1781029327980675237

Also, didn't knew that Xe LPG had support for DP4a. In hindsight it does need it due to XeSS, but I thought it was done on the XMX units on Xe 12.7.

MTL no haz XMX

 

[Ghostsonplanets]

MTL no haz XMX

Oh, i'm aware. I thought DP4a was executed through the XMX units on Xe HPG and thus MTL LPG wouldn’t support it due to lacking it. Bit of a "braindead" moment.

Also, if anyone wants a full review of a 165U such is now available on notebookcheck - https://www.notebookcheck.net/Core-...ell-Latitude-9450-2-in-1-review.825254.0.html Pretty much as expected - slightly lower multi-thread performance, about 2/3 GPU performance, and notably better battery life. It is unfortunate that Intel didn't include the full GPU tile on any of the U series SKUs, but I guess we'll effectively be getting that with LNL.

Coming back to this: CPU performance is unimpressive and fail behind its peers (And even RPL-U in some tests). GPU is actually pretty good! 64 EU LPG is ahead of Iris Xe 96 EU and relatively close to the 128 EU in some tests. Active battery life is an amazing improvement over the past generation. But, somehow, idle power has increased and thus idle battery life is lower than RPL-U🫠.

Overall, MTL-U seems to be a pretty good upgrade for anyone on TGL-U and before. ADL/RPL-U users won't benefit much due basically no improvement on CPU side, despite the others IPs and perf/W being an significant improvement.

On the matters of availability, I have seen very few offerings, if any, of MTL-U based devices in the wild. It seems it has been a very slow ramp up (Or the costs x benefit calculus is massively on RPL 282 side).

Let's see if Intel can get some design wins in the <$600 market until the end of the year. I assume Core Ultra 5 115U will be the key volume driver in this price bracket, if such ramp-up happens of course...(Projected 40M shipments until EoY).

Otherwise, I think it's best to wait until ARL-U announcement at CES, as it will be priced cheaper and will have a perf/W bump due to Intel 3. But certainly will be a challenging and extremely competitive landscape for MTL/ARL-U due to AMD and QCOM offerings.

 

[gdansk]

With MTL Intel now has an NPU in a laptop chip. Only a year after AMD and 3 years after Apple but despite that it has cutting edge AI marketing.

 

[Ghostsonplanets]

With MTL Intel now has an NPU in a laptop chip. Inly a a year after AMD and 3 years after Apple but despite that it has cutting edge AI marketing.

MTL NPU is basically a "saving face" IP. Specially given how late it is. But, if there's a thing Intel is really good at, is at at pushing marketing narratives and flexing their human capital and developer relationships with ISV so that their products are well supported. MTL NPU having so much support day 1 is an example of this.

 

[Khato]

Better battery life (outside of load situation) is probably mostly due to the lower power screen being used compared to the previously reviewed MTL laptops with OLED screens.

Yup. As some have been pointing out all along, MTL battery life looked bad on the initial models due to the OLED displays. Still don't have any MTL reviews from notebookcheck with a plain boring 1080p screen that'd compare against the majority of Phoenix laptops, but I'm sure it'll happen eventually. For now this is how the various MTL and PHX laptops which notebookcheck has reviewed stack up when sorted by their corresponding display size/resolution. Save for the PHX Tuxedo Pulse 14 Gen3, MTL is always on par or better on the WiFi test thus far. And MTL is clearly winning on the H.264 playback metric.

		Wh	H.264	WiFi 1.3	H.264 min/Wh	WiFi min/Wh
https://www.notebookcheck.net/HP-El...en-7-7840U-and-unbeatable-price.754141.0.html	14" 1920x1200 60Hz IPS	51	760	779	14.90	15.27	PHX
https://www.notebookcheck.net/HP-ZB...e-notebook-that-relies-on-Zen-4.759343.0.html	14" 1920x1200 60Hz IPS	51		782		15.33	PHX
https://www.notebookcheck.net/Lenov...n-power-in-the-compact-ThinkPad.786912.0.html	14" 1920x1200 60Hz IPS	52.5		729		13.89	PHX
https://www.notebookcheck.net/Lenov...-laptop-is-better-with-AMD-Zen4.763581.0.html	14" 1920x1200 60Hz IPS	57	988	793	17.33	13.91	PHX
https://www.notebookcheck.net/Lenov...w-fast-portable-straightforward.795741.0.html	14" 1920x1200 60Hz IPS	52	592	685	11.38	13.17	PHX
https://www.notebookcheck.net/HP-El...ness-laptop-ruined-by-Sure-View.809651.0.html	16" 1920x1200 60Hz IPS	76	839	924	11.04	12.16	PHX
https://www.notebookcheck.net/Frame...h-better-than-the-Intel-version.756613.0.html	13.5" 2256x1504 60Hz IPS	61		726		11.90	PHX
https://www.notebookcheck.net/Dell-...ce-RTX-for-integrated-Intel-Arc.813905.0.html	14" 2240x1400 60Hz IPS	64		749		11.70	MTL
https://www.notebookcheck.net/Core-...ell-Latitude-9450-2-in-1-review.825254.0.html	14" 2560x1600 60Hz IPS	60		815		13.58	MTL
https://www.notebookcheck.net/Lenov...Zen4-isn-t-automatically-better.787943.0.html	14.5" 2560x1600 90Hz IPS	73	737	618	10.10	8.47	PHX
https://www.notebookcheck.net/MSI-P...-Core-i7-Xe-to-Core-Ultra-7-Arc.785587.0.html	16 2560x1600 60Hz IPS	99.9		1100		11.01	MTL
https://www.notebookcheck.net/Lenov...Intel-Arc-confronts-Radeon-780M.810926.0.html	14.5" 3072x1920 120Hz IPS	73		680		9.32	MTL
https://www.notebookcheck.net/MSI-P...-7-and-OLED-weighing-under-1-kg.810670.0.html	13.3" 2880x1800 60Hz OLED	75	1002	694	13.36	9.25	MTL
https://www.notebookcheck.net/Tuxed...h-AMD-Zen4-and-a-120-Hz-display.795476.0.html	14" 2880x1800 120Hz OLED	60	595	620	9.92	10.33	PHX
https://www.notebookcheck.net/Lenov...ighter-with-faster-Arc-graphics.821145.0.html	14" 2880x1800 120Hz OLED	75		600		8.00	MTL
https://www.notebookcheck.net/Asus-...th-120-Hz-OLED-and-Core-Ultra-7.805236.0.html	14" 2880x1800 120Hz OLED	75	1010	707	13.47	9.43	MTL
https://www.notebookcheck.net/2-in-...Core-Ultra-allrounder-with-OLED.807431.0.html	16" 2880x1800 120Hz OLED	83	862	717	10.39	8.64	MTL
https://www.notebookcheck.net/Galax...ive-to-the-Apple-MacBook-Air-15.821005.0.html	16" 2880x1800 120Hz OLED	76	1144	644	15.05	8.47	MTL
https://www.notebookcheck.net/Lenov...-inside-this-workstation-laptop.796559.0.html	16" 3840x2400 60Hz OLED	84	797	635	9.49	7.56	PHX

 

[S'renne]

MTL NPU is basically a "saving face" IP. Specially given how late it is. But, if there's a thing Intel is really good at, is at at pushing marketing narratives and flexing their human capital and developer relationships with ISV so that their products are well supported. MTL NPU having so much support day 1 is an example of this.

Even though AMD has an NPU on their APU SoCs a year earlier than Intel, there's like no news on what software its actually supported on isnt it? Not even ROCm/Pytorch while Intel already has their own AI/ML support when people demo'ed running stable diffusion before MTL got released.

Earlier released hardware means nothing when there's barely anything you can do on it a year later...

 

[Ghostsonplanets]

Even though AMD has an NPU on their APU SoCs a year earlier than Intel, there's like no news on what software its actually supported on isnt it? Not even ROCm/Pytorch while Intel already has their own AI/ML support when people demo'ed running stable diffusion before MTL got released.

Earlier released hardware means nothing when there's barely anything you can do on it a year later...

iirc although AMD released Phoenix with an NPU, neither was the NPU exposed to software developers, outside of some (MS Teams), neither was the NPU mandatory as the OEM could enable it or disable the NPU functionality if they wanted it. So more or less what you said, yeah.

 

[S'renne]

iirc although AMD released Phoenix with an NPU, neither was the NPU exposed to software developers, outside of some (MS Teams), neither was the NPU mandatory as the OEM could enable it or disable the NPU functionality if they wanted it. So more or less what you said, yeah.

Idek why people tout that Intel lags behind with CPU SoC hardware like their NPU for the current topic, they've always went for the optimal software/hardware support for their processors(besides Arc) and that's more important than raw performance imo.

 

[SiliconFly]

Better battery life (outside of load situation) is probably mostly due to the lower power screen being used compared to the previously reviewed MTL laptops with OLED screens.

Yup. As some have been pointing out all along, MTL battery life looked bad on the initial models due to the OLED displays... ... MTL is always on par or better on the WiFi test thus far. And MTL is clearly winning on the H.264 playback metric.

In spite of everything, MTL performance is still only on par with previous gen (and that was a bit of a disappointment for me).

And it's battery life is also only on par with the competition (in spite of all the power saving tech). Well, there are a few scenarios where it does provide better results. But there is no clear win and that speaks volumes.

MTL no haz XMX

I thought DP4a was executed through the XMX units on Xe HPG...

XMX is pretty much necessary these days but isn't mandatory. MTL can definitely live without it.

Even though AMD has an NPU on their APU SoCs a year earlier than Intel, there's like no news on what software its actually supported on isnt it? Not even ROCm/Pytorch while Intel already has their own AI/ML support when people demo'ed running stable diffusion before MTL got released.

Earlier released hardware means nothing when there's barely anything you can do on it a year later...

Idek why people tout that Intel lags behind with CPU SoC hardware like their NPU for the current topic, they've always went for the optimal software/hardware support for their processors(besides Arc) and that's more important than raw performance imo.

I don't think it's a fair comparison. When AMD released it's NPU, there weren't many mainstream AI apps to begin with. Today, we have enough apps and more mainstream apps are integrating AI/LLMs as we speak. Things are just getting started and AMD was actually on the leading edge. Forget day 0 for a moment, today all those CPUs support AI and that matters a lot.