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.)

 

[Timorous]

People have limited buying capacity. They are not gonna replace an MTL laptop just coz AMD released something better six months later. That's a loss of sales for AMD. If AMD talks, it helps people make a better, informed buying decision by waiting for something better. People are not gonna wait based on rumors. At least, not average joe. But official confirmation from a company carries weight.

AMD / Intel are not selling to consumers (outside of DIY) they are selling to OEMs and I expect AMD are talking with them. If Zen 5 is good and MTL is bad then I expect MTL based machines to be minor revisions of RDL machines or the selection to be far smaller than average because the OEMs won't want to over commit.

 

[igor_kavinski]

They just don't care.
The policy is don't talk until it's ready.

Why? What's with this mentality of "Hey, we are good and even great in some cases but we don't want total market domination. People can buy Intel if they are desperate"?

 

[igor_kavinski]

because the OEMs won't want to over commit.

But if MTL laptops sales are good, OEMs may do just that and by the time AMD does come out with Strix Point laptops, MTL laptops will be fairly discounted so they may look to be a better buy to the cost conscious consumers. Again, loss of sales for AMD.

 

[Timorous]

But if MTL laptops sales are good, OEMs may do just that and by the time AMD does come out with Strix Point laptops, MTL laptops will be fairly discounted so they may look to be a better buy to the cost conscious consumers. Again, loss of sales for AMD.

If the current leaks are anything to go by it looks like MTL is going to be barely better than Hawk Point and we don't really know what battery life is like in an apples to apples test.

Also consumers will buy stuff when it is on sale or they have a specific need and for the most part it will either be recommended by a techy friend or by a sales rep in a shop so if AMD do want to improve their position in the laptop space they need to do 2 things. 1) Convince OEMs to use their products and get more design wins and 2) work with OEMs / retailers to train staff on the products. AMD are focusing on 1 and that is a lot easier of Intel is continually losing ground in terms of battery life, performance or price.

 

[csbin]

LMAO

https://twitter.com/x/status/1735255934262554668

 

[jpiniero]

Intel launches Core Ultra 100 "Meteor Lake" series, up to 16 CPU cores, Arc GPU with 8 Xe-Core and improved AI performance - VideoCardz.com

Core Ultra 100 series: up to 16 CPU cores & Arc GPU 8 with Xe-Cores Intel launches first client CPU series with new naming schema. CPU: “Redwood Cove” P-Core & “Crestmont” E-Core Intel’s Meteor Lake architecture marks a departure from a monolithic die, opting for four smaller dies (tiles)...

videocardz.com

Does seem like P core boosts are down across the board. The volume part (125U) is 4.3, which is even lower than the entire Raptor Lake mobile lineup.

 

[DAPUNISHER]

TAP discussing ARC in MTL

 

[Hitman928]

Nope. On the first run of CB15 MT the 1370P based Framework 13.5 score is 2492. Well above any of the scores for the 7840U variant due to the longer turbo power duration on Intel versus AMD. That would be the reason why we prefer looking at the steady state runs after the first, where the 7840U at 35W shows around 2170 on average compared to around 1850 for the 1370P running at 30W. So 17% higher performance for 16% higher power, which is again indicative of a clear minor advantage for AMD. But nowhere near what the AMD fans like to imply.

I already posted a graph where the 13700h and 7840hs are tested without boost at many power levels in Cinebench r23, here. The 7840hs has a significant performance advantage at every power level until you get above 55W. The 1360p/7840u and 13700h/7840hs aren't different enough to make a meaningful difference in the results. You can cross check if you'd like with 1370p and 7840u results though, like this review and this one which both show results for the 1370p and 7840u that are right in line with the scores shown in the 13700h/7840hs graph.

From that, I don't see how Intel could get results where the 1370p gets higher performance than a 7840u in a heavy multithreaded test at similar power, unless they are greatly stretching the definition of similar (i.e. similar PL1 level but PL2 and PL2 duration are much higher on the Intel side).

 

[ikjadoon]

LMAO

https://twitter.com/x/status/1735255934262554668

How did Intel lose IPC in a new CPU generation? Maybe Intel tweaked Golden Cove's most inefficient paths, which saved power, but also shaved off IPC? That might be worth it.

The footnote:

Among processors powering ultrathin systems, based on SPECrate*2017_int_base (n-copy) performance estimates for Intel® Core™ Ultra 7 165H on an Intel Internal development system with Intel Compiler 2023. 2.3 and in comparison to prior gen and comp; as of December 2023.

All figures tested on AC with Windows® "Best Performance" setting. Details at intel.com/performanceindex for details. Results may vary.

I hope Meteor Lake's 1T perf / W is a phenomenal upgrade, then. This may well be the largest historical perf regression by any CPU vendor, if VC's leaked slides are real.

Videocardz has not leaked any Intel slides on 1T perf / W. Intel did release those for nT for 12th gen mobile:

 

[Abwx]

From that, I don't see how Intel could get results where the 1370p gets higher performance than a 7840u in a heavy multithreaded test at similar power, unless they are greatly stretching the definition of similar (i.e. similar PL1 level but PL2 and PL2 duration are much higher on the Intel side).

They are actully comparing 45-64W chips to a 28W one, notice that they state that their GPU has 10% better perf than a 7840U@28W, they dont specify that it s at the same power.

The number that matter is that MTL has 8% better perf at same power than a 1370P according to their slides, that would be about 18% better perf/Watt at isofrequency with a same uarch, this doesnt align at all with their doctored slides with 7840U comparisons.

Also they were cautious by using a U version rather than a HS that would be at same power, and for a good reason, they are just selling snake oil.

How did Intel lose IPC in a new CPU generation? Maybe Intel tweaked Golden Cove's most inefficient paths, which saved power, but also shaved off IPC? That might be worth it.

The footnote:



I hope Meteor Lake's 1T perf / W is a phenomenal upgrade, then. This may well be the largest historical perf regression by any CPU vendor, if VC's leaked slides are real.

Videocardz has not leaked any Intel slides on 1T perf / W. Intel did release those for nT for 12th gen mobile:

In this slide the 5900HX perf increase by something like 5% from 25W to 45W, wich is completly fake, typical of Intel marketing, they know that most people understand jack to these curves...

 

[S'renne]

Why does Abwx sound so much like Userbenchmark but for AMD over time now?

 

[ikjadoon]

In this slide the 5900HX perf increase by something like 5% from 25W to 45W, wich is completly fake, typical of Intel marketing, they know that most people understand jack to these curves...

No, Intel's AMD 5900HX numbers actually appear pretty close. The 5900HX was an efficient CPU; basically a Ryzen 7 7735U, a 28W CPU, though clearly not the same class.

Intel's graph shows 5900HX perf went up +15%, not +5%, right? I'd measure it 80% to 95% on the left axis.

Using Cinebench R20 nT at set PL1s (not power consumption, but power limits):
5900HX at 25W PL1: 4703 pts (base line)
5900HX at 35W PL1: 5045 pts (+7.2%)
5900HX at 70W PL1: 5158 (+9.6%)

//

Intel's key deception is in not extending AMD's & Apple's lines to the left, which would show a much larger gap (and likely show both AMD & Apple more efficient at lower power draws).

 

[Abwx]

No, Intel's AMD 5900HX numbers actually appear pretty close. The 5900HX was an efficient CPU; basically a Ryzen 7 7735U, a 28W CPU, though clearly not the same class.

Intel's graph shows 5900HX perf went up +15%, not +5%, right? I'd measure it 80% to 95% on the left axis.

Using Cinebench R20 nT at set PL1s (not power consumption, but power limits):
5900HX at 25W PL1: 4703 pts (base line)
5900HX at 35W PL1: 5045 pts (+7.2%)
5900HX at 70W PL1: 5158 (+9.6%)

//

Intel's key deception is in not extending AMD's & Apple's lines to the left, which would show a much larger gap (and likely show both AMD & Apple more efficient at lower power draws).

15% more perf for 80% higher power..?.

The power evolution shown is more than bi quadratic, that s total non sense.

We re not at 80W to 100W here but between 25W and 45W, that s still within a good efficency range for the 5900HX.

 

[Hitman928]

No, Intel's AMD 5900HX numbers actually appear pretty close. The 5900HX was an efficient CPU; basically a Ryzen 7 7735U, a 28W CPU, though clearly not the same class.

Intel's graph shows 5900HX perf went up +15%, not +5%, right? I'd measure it 80% to 95% on the left axis.

Using Cinebench R20 nT at set PL1s (not power consumption, but power limits):
5900HX at 25W PL1: 4703 pts (base line)
5900HX at 35W PL1: 5045 pts (+7.2%)
5900HX at 70W PL1: 5158 (+9.6%)

//

Intel's key deception is in not extending AMD's & Apple's lines to the left, which would show a much larger gap (and likely show both AMD & Apple more efficient at lower power draws).

Here's a chart from HWUB for comparison.

 

[Abwx]

Here's a chart from HWUB for comparison.

Thanks, i was about to post it, perf increase by 15% from 35 to 70W and not 9.6% from 25 to 70W as stated by Ikjadoon, guess it show how misleading Intel marketing can be , and also how gullible people can be...

 

[coercitiv]

Hmm, so no review units after all, not even a reference platform for the press to tinker with on day one.

Quote from the recent Anandtech article on MTL:

Despite today being the official launch of the Core Ultra series and Meteor Lake platform, you won’t find any reviews for the hardware. And we’re not sure you’ll be able to find much hardware, either.
In a significant departure from how Intel has handled previous mobile-first architecture launches, the company has not sampled any laptops to the press for review ahead of today’s launch. So while you can go out and buy a Core Ultra laptop today (in theory, at least), we can’t actually show you how it performs, or even validate Intel’s own performance claims.

the lack of review hardware, combined with the fact that the flagship Core Ultra 9 185H SKU won’t even be available until Q1, gives us pause. If Meteor Lake is indeed launching in 2023, then it’s only by the skin of Intel’s teeth: 2 weeks before the end of the year, and what we strongly suspect will be a very limited number of laptops available at retail for today’s launch.
Which is not to kick Intel while they’re still trying to pick themselves back up. But if Intel hadn’t already promised a 2023 launch for Meteor Lake, then this would almost certainly would have been a CES launch, if not later.

 

[Hitman928]

First MTL laptop review that I've seen.

Acer Swift Go 14 review: An affordable Intel Core Ultra debut

The Acer Swift Go 14 is a good choice for those looking to buy a lightweight yet zippy laptop at a sub-$1,000 price.

www.pcworld.com

I just briefly skimmed through it but it seems to be more in line with the leaks than Intel's presentation yesterday.

 

[Hitman928]

Hmm, so no review units after all, not even a reference platform for the press to tinker with on day one.

Quote from the recent Anandtech article on MTL:

There's a few (e.g. this one).

Notebookcheck even did a nice performance to power curve for us. Seems Intel was fibbing about their, "similar power" claims. Shame on Intel marketing for being so deceptive about their power/efficiency. Their live demo scores must have been running at like 70W for the 155H while the 7840u was at 30W. Hallock promised they weren't playing any tricks with the power settings but they absolutely were.

 

[eek2121]

Intel launches Core Ultra 100 "Meteor Lake" series, up to 16 CPU cores, Arc GPU with 8 Xe-Core and improved AI performance - VideoCardz.com

Core Ultra 100 series: up to 16 CPU cores & Arc GPU 8 with Xe-Cores Intel launches first client CPU series with new naming schema. CPU: “Redwood Cove” P-Core & “Crestmont” E-Core Intel’s Meteor Lake architecture marks a departure from a monolithic die, opting for four smaller dies (tiles)...

videocardz.com

Does seem like P core boosts are down across the board. The volume part (125U) is 4.3, which is even lower than the entire Raptor Lake mobile lineup.

Intel’s P-cores definitely are not designed for efficiency and I suspect it will hurt them here.

How did Intel lose IPC in a new CPU generation? Maybe Intel tweaked Golden Cove's most inefficient paths, which saved power, but also shaved off IPC? That might be worth it.

The footnote:



I hope Meteor Lake's 1T perf / W is a phenomenal upgrade, then. This may well be the largest historical perf regression by any CPU vendor, if VC's leaked slides are real.

Videocardz has not leaked any Intel slides on 1T perf / W. Intel did release those for nT for 12th gen mobile:

It didn’t? There is a clock and power variance. It is also a completely new tiled design which will cause odd regressions in unexpected places. The cores themselves, however, have not regressed. They actually have better IPC.

No, Intel's AMD 5900HX numbers actually appear pretty close. The 5900HX was an efficient CPU; basically a Ryzen 7 7735U, a 28W CPU, though clearly not the same class.

Intel's graph shows 5900HX perf went up +15%, not +5%, right? I'd measure it 80% to 95% on the left axis.

Using Cinebench R20 nT at set PL1s (not power consumption, but power limits):
5900HX at 25W PL1: 4703 pts (base line)
5900HX at 35W PL1: 5045 pts (+7.2%)
5900HX at 70W PL1: 5158 (+9.6%)

//

Intel's key deception is in not extending AMD's & Apple's lines to the left, which would show a much larger gap (and likely show both AMD & Apple more efficient at lower power draws).

Cezanne was a great chip. I use a 5800 for some heavy workloads and it performs like a champ.

 

[Abwx]

Review at Computerbase, the 155H score 13936 pts in CB R23 with a 64W boost, TDP is officialy 45W.

Computerbase has the power consumption graphs, we can see that the 155H is at 64W for 35 seconds before reducing power, it does boost to 64W even when power is set to 35W but the boost is shorter at 20 seconds.

So Intel slides were a vast sell off of snake oil...

Intel Meteor Lake anhand des Core Ultra 7 155H im Test

Das Acer Swift Go 14 ermöglicht einen Ersteindruck zu Meteor Lake, der dieselbe Single-, mehr Multi-Core- und viel mehr GPU-Leistung bietet.

www.computerbase.de

 

[Abwx]

There's a few (e.g. this one).

Notebookcheck even did a nice performance to power curve for us. Seems Intel was fibbing about their, "similar power" claims. Shame on Intel marketing for being so deceptive about their power/efficiency. Their live demo scores must have been running at like 70W for the 155H while the 7840u was at 30W. Hallock promised they weren't playing any tricks with the power settings but they absolutely were.

The curve of the 155H is not accurate, no way that the perf slope increase that much over 35W, the score stated at 45W in this graph is actually performed at 64W since Computerbase measure 13936 pts at 65W.

 

[Hitman928]

Appears the 155H loses in gaming as well when it doesn't have a huge power advantage. There should be more powerful tGPU's from Intel though in the higher tier models.

Spoiler

Not really a significant battery life improvement during video playback either according to the PCworld review linked above:

Spoiler

Unless these first reviews got some terribly buggy hardware to review, seems MTL is indeed a bit of a dud. ST regression, small efficiency improvement, and no halo tier offerings. GPU improvement is significant but doesn't seem to convincingly beat the competition. AI is a huge improvement but not really a big selling point at this time, I don't think, and will soon be 2nd to the AMD 8000 series. I wonder if Intel is willing to eat it on margins to get this out or if they will limit volume and push RPL-R instead.

 

[yuri69]

People have limited buying capacity. They are not gonna replace an MTL laptop just coz AMD released something better six months later. That's a loss of sales for AMD. If AMD talks, it helps people make a better, informed buying decision by waiting for something better. People are not gonna wait based on rumors. At least, not average joe. But official confirmation from a company carries weight.

AMD need to pay for the development of Phoenix first. They have not sold that many yet.

 

[adroc_thurston]

Oh god, MTL seems to have caught PHXus and has some non-trivial FW issues on launch.

They have not sold that many yet.

They've literally said they'd shipped like a zillion AI PC units (that's all PHX) to date.
What's with the esoteric cope.

 

[tamz_msc]

Are the pro-AMDers blind? Do they need magnifying glasses to read the footnotes to find out the Intel ran SPECint_2017_base 1copy for 1T and ncopy for MT? Do they not know that SPEC2017 is especially demanding on the memory subsystem performance? And do they not see that the comparisons are with 7467 MT/s memory vs 6400 MT/s memory?