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.)

 

[IntelUser2000]

Tick overview: Generally the platform is more stable and fully featured, and has more SKUs to address more segments. Addresses competitive gaps.

Penryn: Companion G45 chipset had far better iGPU. More and balanced SKUs.
Westmere: Arrandale/Clarkdale for client featuring integrated graphics and also for low power mobile. Nehalem only had one mobile chip for high end. iGPU being on-package and introducing Hi-Z improved performance greatly. Westmere-EX for enterprise that Sandy Bridge didn't have.
Ivy Bridge: DX11 support, and (finally!) proper anisotropic filtering support. Introduction of Y chips with SDP for lower power.
Broadwell: Haswell brought 50% battery life gains. Broadwell did another 30%, achieving the original goal to have 2x the battery life. Core M introduced for official fanless chips.
Kabylake: VP9 support in GPU for Youtube acceleration which is big. Skylake was a sidegrade in terms of battery life and a lot of times even a regression. Kabylake made it better.

 

[IntelUser2000]

@Exist50 What are the chances Crestmont is a decent gain over Gracemont? Maybe Redwood Cove is a small gain but what about the E core?

 

[Exist50]

@Exist50 What are the chances Crestmont is a decent gain over Gracemont? Maybe Redwood Cove is a small gain but what about the E core?

I think it'll be a bigger gen/gen performance improvement than Redwood Cove, low bar though that might be, but I wouldn't expect miracles either. Think Skymont is supposed to be a much bigger change.

 

[IntelUser2000]

I think it'll be a bigger gen/gen performance improvement than Redwood Cove, low bar though that might be, but I wouldn't expect miracles either. Think Skymont is supposed to be a much bigger change.

The Atoms alternated between new features and expanding/optimizing. Wonder if it'll change with Crestmont.

Silvermont - New ideas
Airmont - Optimization
Goldmont - New ideas
Goldmont Plus - Expansion/Optimization
Tremont - New ideas
Gracemont - Expansion

 

[Kosusko]

@Exist50 What are the chances Crestmont is a decent gain over Gracemont? Maybe Redwood Cove is a small gain but what about the E core?

Intel Tremont delivers a massive 30% performance-per-clock jump – at least according to Intel – over its predecessor, Intel Goldmont Plus.

Intel Gracemont is direct ancestor, and represents a shift in Intel’s Atom strategy.

I assume that Cretmont will increase the IPC by ~30% over Gracemont.

 

[poke01]

. Core M introduced for official fanless chips

I remember those in the 12" MacBooks in 2015. Oh how I wished Apple used the A9X chip in them instead. Performance was not there and those chips got hot for simple tasks.

The A9X on the other hand was cool and performant in the iPad Pro.

I am confused why Intel never focused on pref/w. I mean if you lead in that you lead in mobile and datacentre.

 

[IntelUser2000]

I am confused why Intel never focused on pref/w. I mean if you lead in that you lead in mobile and datacentre.

They made a hoopla about Core M being this big change for lower power. The reality seemed much different to me. Yes it eventually led to fanless systems. Of course thermal technology was less advanced back then and couldn't handle more than 5W. Nowadays you can get fanless systems that handle 8W+. My new/used Yoga 710 11 is fanless and it has a PL1 of 5 to 7W depending on the environment(changes dynamically). Never runs hotter than 70C and usually stays under 50C for light tasks.

They struggled on the 14nm release and Core M was the lead product. I don't doubt the disappointment of the product are connected. When the pitfalls of BK management started showing.

But again mediocre/bad management was festering in that company for years, even back then!

 

[Exist50]

The Atoms alternated between new features and expanding/optimizing. Wonder if it'll change with Crestmont.

Silvermont - New ideas
Airmont - Optimization
Goldmont - New ideas
Goldmont Plus - Expansion/Optimization
Tremont - New ideas
Gracemont - Expansion

I think Gracemont doesn't really fit in there. It's a very large leap over Tremont in both performance and features (e.g. AVX2 support). Crestmont wasn't originally supposed to exist, so I'm expecting a "Goldmont Plus" kind of situation, but the Atom team seems to be more flexible about architectural changes than Core, so pretty large error bars on what that means.

 

[Doug S]

[QUOTE="poke01, post: 40851808, member: 478379"I am confused why Intel never focused on pref/w. I mean if you lead in that you lead in mobile and datacentre.
[/QUOTE]

Intel has done that several times. The original Core was targeted at perf/watt - but that was in comparison to the hot and power hungry Pentium 4. If you are less efficient than the market leader you can target perf/w but have the result of that effort still be less efficient than the market leader.

 

[mikk]

I think Gracemont doesn't really fit in there. It's a very large leap over Tremont in both performance and features (e.g. AVX2 support). Crestmont wasn't originally supposed to exist, so I'm expecting a "Goldmont Plus" kind of situation, but the Atom team seems to be more flexible about architectural changes than Core, so pretty large error bars on what that means.

Goldmont Plus was a huge improvement, at the same clock it was like 40-45% faster in Cinebench 1T. It was a big surprise at this time because they just added a plus.

In Raptor Lake the bigger IPC increase seems to come from Gracemont despite the unchanged little core codename whereas the big core got a new Cove codename.

 

[Exist50]

Goldmont Plus was a huge improvement, at the same clock it was like 40-45% faster in Cinebench 1T. It was a big surprise at this time because they just added a plus.

Ok, well certainly not that, lol. Kinda forgot how big an improvement GMT+ was. Anyway, don't think I know enough to meaningfully guess a range, but iirc it was originally created as basically a filler for MTL. Enough time to do better than Gracemont, but not enough to fit in everything Skymont was to bring.

But I have to say, it's kinda funny. You've got stuff like this diagram, but in reality, the gap between Core and Atom seems to be closing by the year. That's obviously not a stable long-term arrangement, so what'll happen? Royal replace Core? Core replace Atom? Should be fun.

 

[ashFTW]

Ok, well certainly not that, lol. Kinda forgot how big an improvement GMT+ was. Anyway, don't think I know enough to meaningfully guess a range, but iirc it was originally created as basically a filler for MTL. Enough time to do better than Gracemont, but not enough to fit in everything Skymont was to bring.

But I have to say, it's kinda funny. You've got stuff like this diagram, but in reality, the gap between Core and Atom seems to be closing by the year. That's obviously not a stable long-term arrangement, so what'll happen? Royal replace Core? Core replace Atom? Should be fun.

View attachment 68091

Since core can’t negate Atom’s area efficiency, it’s not a sensible replacement. If there is a newer (Royal) P-core, which is likely both more powerful and area plus power efficient, it will replace Core and Core will be phased out.

 

[IntelUser2000]

Goldmont Plus was a huge improvement, at the same clock it was like 40-45% faster in Cinebench 1T. It was a big surprise at this time because they just added a plus.

Goldmont Plus was about 30% faster per clock in Cinebench 1T. With clocks it was about 45% faster. The peak clock and sustained clock both increased by about 10%.

 

[eek2121]

Intel needs to get Atom into more products IMO. Thus far we have only seen Gracemont paired with a big core. I would love to see anSBC with Gracemont in it, for example.

 

[msj10]

I think Gracemont doesn't really fit in there. It's a very large leap over Tremont in both performance and features (e.g. AVX2 support). Crestmont wasn't originally supposed to exist, so I'm expecting a "Goldmont Plus" kind of situation, but the Atom team seems to be more flexible about architectural changes than Core, so pretty large error bars on what that means.

so the 'Next'mont here was supposed to be skymont?

 

[mikk]

Goldmont Plus was about 30% faster per clock in Cinebench 1T. With clocks it was about 45% faster. The peak clock and sustained clock both increased by about 10%.

I'm comparing N4100 Goldmont Plus to N3350 Goldmont. Both have a burst frequency of 2.4 Ghz. There are 9-12 Cinebench R15 1T results on Notebookcheck.

 

[Kosusko]

I compared Tremont vs LuJiaZui.
Small cores have evolved as well as revolutionized their development. That's why x86 big.LITTLE is a paradigm shift with the rise of small cores.
Crestmont will be an evolution of Gracemont and Darkmont will be another recolution of small cores.

 

[DrMrLordX]

paradigm shift

You like this term too much.

 

[Exist50]

so the 'Next'mont here was supposed to be skymont?

Yes, I believe so.

 

[Kosusko]

You like this term too much.

Small cores have never been more important than since Alder Lake's paradigm shift. With Raptor Lake the paradigm shift continues and Raptormont (sorry Gracemont) is up to 16c while Raptor Cove is "only" 8C.

Just the idea that AMD Zen 4 with 16C Ryzen 9 7950X will compete for the top spot in the TOP mainstream even with 16c atomic E-cores in Core i9 13900K completely solidifies this concept.

P.S. as you well know, I grew up on small cores VIA Isaiah (and before that VIA Samuel/Samel 2, Erza, Nehemiah, Esther), Intel Atom, AMD Bobcat/Jaguar/Puma or Intel Tremont as also atom line.

https://twitter.com/x/status/1570003272777203713

 

[IntelUser2000]

I'm comparing N4100 Goldmont Plus to N3350 Goldmont. Both have a burst frequency of 2.4 Ghz. There are 9-12 Cinebench R15 1T results on Notebookcheck.

Nah, 30% is based on real-world clocks of devices. 45% is overestimating it. They need the desktop 10W TDP in order to have any chance of reaching the burst frequency in MT workloads. Goldmont Plus was able to significantly improve on the power efficiency per clock part despite having substantial increase in performance per clock which is why it does better than list clocks suggest.

 

[DrMrLordX]

paradigm shift.

I sense the start of a drinking game.

 

[stebler]

Looks like FP64 support has been mostly added back in the MTL iGPU, with the exception of fdiv and fsqrt instructions. https://github.com/intel/intel-graphics-compiler/commit/aa829c9dd5880c6bbf3584fbdad7082d862a3e41

 

[moinmoin]

Looks like FP64 support has been mostly added back in the MTL iGPU, with the exception of fdiv and fsqrt instructions. https://github.com/intel/intel-graphics-compiler/commit/aa829c9dd5880c6bbf3584fbdad7082d862a3e41

Wait, I don't follow Intel iGPUs, when and why was FP64 support removed before?

 

[IntelUser2000]

Wait, I don't follow Intel iGPUs, when and why was FP64 support removed before?

In the Gen 11 Icelake generation they removed it. Up to Gen 9 Skylake it had 1/4 FP64 support. Removing was to reduce space so they can fit more for relevant features, since FP64 isn't for gaming.