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.)

 

[scineram]

	Process Node	Transistors	Die Size		Execution Units	ALU	RT units	TGP	Graphics Clock	FP32 (TFlops)
A370M	N6	7.2 B	157 mm2	8 Xe Cores	128 EU	1024 ALU	8	35 - 50W	1550 MHz	4.2 TF
GT2	N5	?	?	4 Xe Cores	64 EU	512 ALU	?	?	?	?
GT3	N5	?	?	8 Xe Cores	128 EU	1024 ALU	?	?	?	?

Since Kepler leaks the tGPU info on upcoming Meteor Lake, I have compiled the comparison chart of tGPU and A370M, make me think how much power MTL's tGPU needed, any ideas?

In Intel parlance an Execution Unit is how many shaders? How does it compare to a CU or SM?

 

[uzzi38]

There is no indication of a GT3 for Meteor Lake, github only has GT2 entries for mobile. It makes no sense, otherwise we would have to assume MTL only comes with 64EUs against 96 EUs on the previous generation. Kepler also claimed RDNA3 ships with a broken Shader Pre-Fetching which AMD denied. Kepler is not a good source. He claims something without any source and people believe in him.

Whether or not it fixed idk, but N31 retail silicon is definitely A0.

Also if MTL is 64EU only I'm going to be incredibly disappointed in it. Actually useless product, guess I'll be waiting for Arrow Lake to be interested in Intel's mobile lineup again.

 

[Exist50]

Intel literally said Intel 4 production has started.

No, they said it's ready for production. Which is a rather important distinction. Even if the production stepping for all MTL dies have taped out (which we don't know), it would still probably take months before they're ready to ship it.

They admittedly wanted to leapfrog AMD by leaving out an interconnect via organic package and this may still work out if AMD does not come to terms.

Honestly, AMD's strategy seems more than fine. Why would they bother pursuing chiplets for client when they have a perfectly viable strategy around monolithic dies?

Also if MTL is 64EU only I'm going to be incredibly disappointed in it. Actually useless product, guess I'll be waiting for Arrow Lake to be interested in Intel's mobile lineup again.

Isn't the split supposed to be 64 and 128EUs?

 

[uzzi38]

Isn't the split supposed to be 64 and 128EUs?

That's absolutely what I thought too, but when it comes to performance of MTL I know exceedingly little outside of vague statements made by a couple of people. They don't seem all that happy about CPU performance vs Raptor Lake anyway, I'm just saying I really hope the GPU performance doesn't disappoint as well.

Which it absolutely would if it's 64EUs only. 128EUs I can imagine would fall somewhere in the 1.5-2x range of performance relative to current Xe gfx which would be fantastic (especially if we see further improvements to Intel's driver suite). But 64EUs... that could only be a sidegrade at best I'd think.

 

[Joe NYC]

No, they said it's ready for production. Which is a rather important distinction. Even if the production stepping for all MTL dies have taped out (which we don't know), it would still probably take months before they're ready to ship it.

How the Intel 4 process node performs is also a variable in this equation.

Honestly, AMD's strategy seems more than fine. Why would they bother pursuing chiplets for client when they have a perfectly viable strategy around monolithic dies?

I think AMD will be sticking with monolithic in Thin and Light segment, until packaging technology with very low power overhead emerges.

 

[Tigerick]

In Intel parlance an Execution Unit is how many shaders? How does it compare to a CU or SM?

I am not familiar with Intel graphics design, even AMD has made some changes on CU with RDNA3. That reminds me to list the specs of upcoming Phoenix Point GPU on the comparison chart, we will know more about new APU platform from AMD. As for Intel Meteor Lake which supposedly competes with AMD's PP is nowhere to be seen

 

[mikk]

No, they said it's ready for production. Which is a rather important distinction. Even if the production stepping for all MTL dies have taped out (which we don't know), it would still probably take months before they're ready to ship it.

Honestly, AMD's strategy seems more than fine. Why would they bother pursuing chiplets for client when they have a perfectly viable strategy around monolithic dies?

Isn't the split supposed to be 64 and 128EUs?

Between GT1 and GT2. Assuming GT2 only gets 64EUs there won't be much GT3 128EU if any at all, as I said all the entries on github are GT2. They can surely clock it much higher but 64EUs against 96EUs can't be much more than a sidegrade update 2 years or 3 years after the introduction of Xe LP. That's why I think it makes no sense and Kepler is a bad Intel source.

 

[Tigerick]

Between GT1 and GT2. Assuming GT2 only gets 64EUs there won't be much GT3 128EU if any at all, as I said all the entries on github are GT2. They can surely clock it much higher but 64EUs against 96EUs can't be much more than a sidegrade update 2 years or 3 years after the introduction of Xe LP. That's why I think it makes no sense and Kepler is a bad Intel source.

GT3 might not be for MTL platform, it may be for future ARL platform. Since we know ARL is going to be made by TSMC N3 process, Intel might drop the ARL's tGPU which is rumored made by N3 process.

 

[Dayman1225]

Between GT1 and GT2. Assuming GT2 only gets 64EUs there won't be much GT3 128EU if any at all, as I said all the entries on github are GT2. They can surely clock it much higher but 64EUs against 96EUs can't be much more than a sidegrade update 2 years or 3 years after the introduction of Xe LP. That's why I think it makes no sense and Kepler is a bad Intel source.

I find It hard to believe it’ll only be 64 EUs when raja said this just 2 weeks ago

When XE first came out, I think a lot of us were excited that the integrated graphics on basic CPUs would get a significant bump, but we haven't seen that yet. Where do you see the baseline for entry-level integrated graphics, and why hasn't that progressed significantly recently?
Raja Koduri: Great question! In fact, that was always the plan. Meteor Lake was always the planbut what happened, and this is something we have publicly said, was delays in our Core CPU roadmap and 10nm process. We stayed on 10nm for a couple more generations [than intended]. Advanced XE graphics were on the next node, ready to go, but the Meteor Lake platform is the one that is going to ship that new Xe graphics and all those great features. So I can't wait for the world to see Meteor Lake, which will change the entire integrated graphics landscape.

link to article

 

[TESKATLIPOKA]

What's the problem with 64EU? Alder lake IGP has max boost of only 1.55GHz. Meteor Lake IGP will certainly have higher clocks and should have better performance per frequency thanks to the improved Alchemist architecture.
I wouldn't be surprised If 64EU IGP in Meteor Lake would perform similarly to the one in Alder or Raptor Lake If not better.
Then we still have the 128EU variant. I think this one should be pretty close to Rembrandt.
Ok, AMD will have Phoenix and will keep the IGP crown, but for casual gaming even this shouldn't be that bad, especially If they improve their drivers.
64EU IGP is basically a single Render slice or 1/8 of Arc A770, 128EU IGP is 1/4. Of course, performance will differ thanks to clocks and memory.

P.S. If they can't increase the clock significantly within limited TDP, then that's a pretty big problem.

 

[Glo.]

Without any form of integrated cache, I don't see any possibility how MTL iGPU could reshape entire integrated graphcis landscape.

The more EUs there are on the MTL package - the more it needs cache, instead of relying on simply DDR5 memory controller.

 

[TESKATLIPOKA]

Without any form of integrated cache, I don't see any possibility how MTL iGPU could reshape entire integrated graphcis landscape.

The more EUs there are on the MTL package - the more it needs cache, instead of relying on simply DDR5 memory controller.

Arc A770 uses 256bit 17.5gbps memory.
128EU would need 1/4 BW or 128bit 8.75gbps.
64EU would need 1/8 BW or 128bit 4.375gbps.
The bigger one will have a bit of a problem with insufficient BW.

 

[dullard]

Without any form of integrated cache, I don't see any possibility how MTL iGPU could reshape entire integrated graphcis landscape.

The more EUs there are on the MTL package - the more it needs cache, instead of relying on simply DDR5 memory controller.

Isn't the rumor that Meteor Lake has far more iGPU cache and conversely is no longer sharing the CPU's L3 cache?

 

[IntelUser2000]

What makes no sense is that 64EU is "GT2" when it's 96EU for GT2 now.

Yea something's wrong with Kepler.

@TESKATLIPOKA Remembrandt in mid-late 2023 at the earliest is not good at all. Meteorlake probably needs the original 192EU to be competitive with whatever AMD has out as the 2024 solution, because most Meteorlake will be 2024 chips. So it needs 2.5-3x Iris Xe, not 2x.

 

[TESKATLIPOKA]

What makes no sense is that 64EU is "GT2" when it's 96EU for GT2 now.

Yea something's wrong with Kepler.

@TESKATLIPOKA Remembrandt in mid-late 2023 at the earliest is not good at all. Meteorlake probably needs the original 192EU to be competitive with whatever AMD has out as the 2024 solution, because most Meteorlake will be 2024 chips. So it needs 2.5-3x Iris Xe, not 2x.

Does It really need to be competitive with Phoenix?
For 192EU there wouldn't be enough BW and most likely power budget.
Strix Point will be released in 2024, but I don't know what we should expect. I think more CU, at least 16CU should be expected.

 

[Glo.]

36W Iris Xe in 1280P scores 1830 pts in 3DMark TS. 28W Xe in 1260P - 1725 pts.

The performance in games is still lagging behind what those GPUs should perform, based on the graphics scores in TS, so there is a hope that Intel will fix the drivers.

But yes. I think, as well, Intel needs around 3500 Pts in 3DMark TS in order to reshape the entire integrated landscape.

P.S. I like what Raja said about changing the PC form factor .

 

[Glo.]

Isn't the rumor that Meteor Lake has far more iGPU cache and conversely is no longer sharing the CPU's L3 cache?

Considering its a separate tile altogether - it needs to have its own cache.

 

[uzzi38]

What makes no sense is that 64EU is "GT2" when it's 96EU for GT2 now.

Yea something's wrong with Kepler.

@TESKATLIPOKA Remembrandt in mid-late 2023 at the earliest is not good at all. Meteorlake probably needs the original 192EU to be competitive with whatever AMD has out as the 2024 solution, because most Meteorlake will be 2024 chips. So it needs 2.5-3x Iris Xe, not 2x.

Rembrandt tier performance doesn't sound good on paper, but it'll be fine. I think even compared to Phoenix the gap between AMD and Intel will shrink relative to now, although I may be over-estimating the gap currently when I say that.

But I don't expect Phoenix to be more than 20-30% faster at reasonable power limits for mobile. Maybe on desktop with unlimited power it might go further... Navi31 seems to love additional power from what we've seen.

Issue is that RDNA3 as we've seen so far just consumes a lot of power... even on desktop it's running <900mv in heavier workloads even at the maxed out TDP. They're quite low on the V/f curve already, so there's little ability to scale down for mobile.

 

[IntelUser2000]

Rembrandt tier performance doesn't sound good on paper, but it'll be fine. I think even compared to Phoenix the gap between AMD and Intel will shrink relative to now, although I may be over-estimating the gap currently when I say that.

Yes they absolutely need to be competitive! I am pretty sure that's what they want if they are able to do so. By leading in all categories you can sell it as a true halo product.

Interesting how the consensus seems to be 2x performance is going to be fine since Phoenix will be a relative small gain over Rembrandt. Always disappoints compared to rumors, always.

Although, 2024 is Strix Point not Phoenix.

The performance in games is still lagging behind what those GPUs should perform, based on the graphics scores in TS, so there is a hope that Intel will fix the drivers.

Remember how Intel said that they had to redo the driver for ARC because the iGPU mindset meant at much higher performance it's suboptimal?

The problem with ARC drivers is that the driver uses too much CPU power, and that's true in general not just old APIs.

But it's true even for their iGPU. You can see that while AMD's Cezanne had the CPU running at 1.2-1.5GHz, Tigerlake was running over 3GHz. So the CPU is sapping more power than it's necessary away from the integrated graphics.

That's probably one big reason why it performs so well in 3DMark, since at those scores it's very GPU bound since you are getting under 20 fps.

 

[TESKATLIPOKA]

Yes they absolutely need to be competitive! I am pretty sure that's what they want if they are able to do so. By leading in all categories you can sell it as a true halo product.

Interesting how the consensus seems to be 2x performance is going to be fine since Phoenix will be a relative small gain over Rembrandt. Always disappoints compared to rumors, always.

Although, 2024 is Strix Point not Phoenix.

Is It confirmed that Meteor won't come this year? We also don't know when Strix Point will be really announced next year.

Meteor IGP at Rembrandt level will be "enough", but of course Phoenix will be faster.
In my opinion, Meteor Lake will have stronger CPU while Phoenix the IGP, which one will be more efficient is unknown.

I would prefer the better IGP option, but If I can get It with a dGPU for a reasonable price, then I will choose that option.
I won't pay 1400-1500eur for Phoenix + 32GB + 1TB SSD If I can buy a Phoenix(Meteor) + 32GB + 1TB SSD + RTX 4050(Ti) for 1600-1700eur and gain at least 50% extra FPS. The second option provides better Bang for the buck.

 

[Glo.]

Is It confirmed that Meteor won't come this year? We also don't know when Strix Point will be really announced next year.

Meteor IGP at Rembrandt level will be "enough", but of course Phoenix will be faster.
In my opinion, Meteor Lake will have stronger CPU while Phoenix the IGP, which one will be more efficient is unknown.

I would prefer the better IGP option, but If I can get It with a dGPU for a reasonable price, then I will choose that option.
I won't pay 1400-1500eur for Phoenix + 32GB + 1TB SSD If I can buy a Phoenix(Meteor) + 32GB + 1TB SSD + RTX 4050(Ti) for 1600-1700eur and gain at least 50% extra FPS. The second option provides better Bang for the buck.

Mini PCs, NUC-like with Rembrandt are around 649$ for version with 6900HX, 16 GB RAM and 512 GB SSD.
Intel charges a lot more, actually around 1000$, for a NUC with 16 GB RAM, 512 GB SSD and i7-1260P.

Phoenix is a premium product, tho.

 

[TESKATLIPOKA]

Mini PCs, NUC-like with Rembrandt are around 649$ for version with 6900HX, 16 GB RAM and 512 GB SSD.
Intel charges a lot more, actually around 1000$, for a NUC with 16 GB RAM, 512 GB SSD and i7-1260P.

Phoenix is a premium product, tho.

I want a new machine to play with and It has to be a laptop, Mini PC is not for me.
What I wanted to say is that IGP performance wouldn't matter to me If a dGPU version is for a reasonable price increase.
I am interested in Meteor Lake, but I most likely won't wait for It, when we don't even know what performance increase to expect.

 

[Glo.]

I want a new machine to play with and It has to be a laptop, Mini PC is not for me.
What I wanted to say is that IGP performance wouldn't matter to me If a dGPU version is for a reasonable price increase.
I am interested in Meteor Lake, but I most likely won't wait for It, when we don't even know what performance increase to expect.

Oh you were talking about a machine for you .

I've just wanted to post something to think about in the context of Phoenix pricing. AMD APU based PCs are vastly cheaper than Intel's, usually.

 

[LightningZ71]

Essentially, that's the market though. Every day, you can go online and find half a dozen traditional laptops with a 5x00H and a 3050 dGPU in them for under $800. You can't find more than a rare one 6800u or 6800h (with no dGPU) laptop out there for notably less than a grand. Rembrandt is just not a good value unless you HAVE to have it in a particular form factor.

 

[TESKATLIPOKA]

Oh you were talking about a machine for you .

I've just wanted to post something to think about in the context of Phoenix pricing. AMD APU based PCs are vastly cheaper than Intel's, usually.

Yeah, It's for me. It's time to move on from 4C8T 9300H + GTX 1650M to something better.

Essentially, that's the market though. Every day, you can go online and find half a dozen traditional laptops with a 5x00H and a 3050 dGPU in them for under $800. You can't find more than a rare one 6800u or 6800h (with no dGPU) laptop out there for notably less than a grand. Rembrandt is just not a good value unless you HAVE to have it in a particular form factor.

Yeah, In my country 8C16T Rembrandt is simply overpriced and you don't even have many models to choose from, basically only premium laptops, although I must say they look nice. Cezanne on the other hand has a pretty good price, but It has weaker IGP.

If Meteor was released at the same time as Phoenix or just a bit behind, I think I would find a cheaper laptop with Meteor + Ada than Phoenix + Ada.
The biggest problem with Meteor Lake is that It's delayed to maybe H2 of 2023 If not Q1 2024.