Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads

[Tigerick]

Wildcat Lake (WCL) Preliminary Specs

Intel Wildcat Lake (WCL) is upcoming mobile SoC replacing ADL-N. WCL consists of 2 tiles: compute tile and PCD tile. It is true single die consists of CPU, GPU and NPU that is fabbed by 18-A process. Last time I checked, PCD tile is fabbed by TSMC N6 process. They are connected through UCIe, not D2D; a first from Intel. Expecting launching in Q2/Computex 2026. In case people don't remember AlderLake-N, I have created a table below to compare the detail specs of ADL-N and WCL. Just for fun, I am throwing LNL and upcoming Mediatek D9500 SoC.

		Intel Alder Lake - N	Intel Wildcat Lake	Intel Lunar Lake		Mediatek D9500
Launch Date		Q1-2023	Q2-2026 ?	Q3-2024		Q3-2025
Model		Intel N300	?	Core Ultra 7 268V		Dimensity 9500 5G
Dies		2	2	2		1
Node		Intel 7 + ?	Intel 18-A + TSMC N6	TSMC N3B + N6		TSMC N3P
CPU		8 E-cores	2 P-core + 4 LP E-cores	4 P-core + 4 LP E-cores		C1 1+3+4
Threads		8	6	8		8
Max Clock		3.8 GHz	?	5 GHz
L3 Cache		6 MB	?	12 MB
TDP		7 W	Fanless ?	17 W		Fanless
Memory		64-bit LPDDR5-4800	64-bit LPDDR5-6800 ?	128-bit LPDDR5X-8533		64-bit LPDDR5X-10667
Size		16 GB	?	32 GB		24 GB ?
Bandwidth			~ 55 GB/s	136 GB/s		85.6 GB/s
GPU		UHD Graphics		Arc 140V		G1 Ultra
EU / Xe		32 EU	2 Xe	8 Xe		12
Max Clock		1.25 GHz		2 GHz
NPU		NA	18 TOPS	48 TOPS		100 TOPS ?

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.

 

[TESKATLIPOKA]

We heard about 128EU, which is a 33% increase. Not bad but not great either. What I am interested in is whether the clocks stay the same or actually increase, then It could be pretty interesting.

 

[Tigerick]

We heard about 128EU, which is a 33% increase. Not bad but not great either. What I am interested in is whether the clocks stay the same or actually increase, then It could be pretty interesting.

GT2 with 128EU comes with 1024ALU which is actually bigger than Phoenix's 768. As for clock speed should be similar range with Phoenix cause they are made by N5/N4P. That's why AMD has to release Hawk Point with double ALU

 

[TESKATLIPOKA]

GT2 with 128EU comes with 1024ALU which is actually bigger than Phoenix's 768. As for clock speed should be similar range with Phoenix cause they are made by N5/N4P. That's why AMD has to release Hawk Point with double ALU

A770 with 512EU(4096 shaders) is 9% slower at 1080p than RX 7600 with 32CU(2048 dual-issue shaders), but 20% faster at 4k.
GT2 with 128EU(1024 shaders) shouldn't be faster than 780M with 12CU(768 dual-issue shaders).

 

[Tigerick]

A770 with 512EU(4096 shaders) is 9% slower at 1080p than RX 7600 with 32CU(2048 dual-issue shaders), but 20% faster at 4k.
GT2 with 128EU(1024 shaders) won't be faster than 780M with 12CU(768 dual-issue shaders).

A770 are made by TSMC N6 process at 2.1GHz clock speed. GT2 with newer process should clock up to 2.8GHz, go figure

 

[TESKATLIPOKA]

A770 are made by TSMC N6 process at 2.1GHz clock speed. GT2 with newer process should clock up to 2.8GHz, go figure

A770 clocks as high as 2400MHz and 2386Mhz as average(TPU) and we are talking about a 225W GPU.
Now an IGP with 1/4 of execution units should clock to 2.8GHz and under 50W? I am pretty skeptical.
Then there is still the question of how to feed It, because raw performance would have increased by ~2.5x.

 

[Tigerick]

A770 clocks as high as 2400MHz and 2386Mhz as average(TPU) and we are talking about a 225W GPU.
Now an IGP with 1/4 of execution units should clock to 2.8GHz and under 50W? I am pretty skeptical.
Then there is still the question of how to feed It, because raw performance would have increased by ~2.5x.

Yeah, part of the power might come from 256-bit memory bus. Anyhow, I suspect Intel has upgraded the process of tGPU from rumored N5 to N4P in order to boost efficiency, we shall see...

 

[eek2121]

Coercitiv, mikk, and eek2121 covered it pretty well. But I do want to highlight the word 'and' in your quote. That word should be 'or'. 20% more performance or 40% lower power. We aren't getting both. Not only that, but those numbers are only accurate when the processor is run at ~0.65 V. Look at the circles in this image. Intel 4 gets either 21% more frequency at the same power OR 40% less power at the same frequency.

AnandTech Forums: Technology, Hardware, Software, and Deals

Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

www.anandtech.com

But, now look at the square dots. At 0.85V and 3 GHz frequency, Intel 4 was only ~35% less power than Intel 7.

As you keep increasing the voltage, the possible frequencies get higher but the curves start to bend sharply upwards to even higher power. Extrapolate this out to 5 GHz or 6 GHz and the power could be massive. The Intel 4 node isn't optimized for high frequency use. The data isn't shown in the graph but it could be possible that MTL is just not going to perform well at ADL frequencies. Wait a few more months for Arrow Lake if that is your need.

...and if this chip only had process gains going for it, we likely wouldn't see such a great showing. However, Intel has improved power management, and I suspect battery life on this chips is going to be something else. We will see. Cautiously optimistic.

I'd disagree that MTL isn't going to perform much better than the non-ultra parts... just not on the CPU side The major improvements for MTL are efficiency as you note and graphics performance. Which makes sense as those are the two areas where they're lagging competition.

Well yes, I wasn't including the GPU. The GPU should be a decent jump over Raptor Lake. At least current gen Raptor Lake. The refresh might give us something new.

 

[A///]

Well yes, I wasn't including the GPU. The GPU should be a decent jump over Raptor Lake. At least current gen Raptor Lake. The refresh might give us something new.

the gpu is a departure from intel's current igpu and should open the doors for better integrated graphics in the future. I'd love to see the tech here advance enough where in 10 years provided most of us old farts on here are alive can play esports titles at 4k on a laptop or most games at 1440p on a laptop. it'll be better passing time on the planes than drinking multiple mini bottles whilst reading a book trying to ignore the blabbering fool 4 rows away talking about her bunyuns.

 

[Kepler_L2]

GT2 with 128EU comes with 1024ALU which is actually bigger than Phoenix's 768. As for clock speed should be similar range with Phoenix cause they are made by N5/N4P. That's why AMD has to release Hawk Point with double ALU

Hawk Point is not new silicon.

 

[Tigerick]

Hawk Point is not new silicon.

Hmm, so you meant PHX+ is still based on Phoenix? No changes in graphics engine?

 

[Kepler_L2]

Hmm, so you meant PHX+ is still based on Phoenix? No changes in graphics engine?

There is no "Phoenix+". PHX4 was a planned die a long time ago (when PHX1 was supposed to use N5), but that has been cancelled for a while. Hawk Point is just a firmware update like Lucienne/Barcelo.

 

[mikk]

We heard about 128EU, which is a 33% increase. Not bad but not great either. What I am interested in is whether the clocks stay the same or actually increase, then It could be pretty interesting.

Much improved, we have seen lots of 2.1 Ghz entries from ES samples.

 

[mikk]

Much improved, we have seen lots of 2.1 Ghz entries from ES samples.

Ultra 7 1003H seems to have a graphics frequency of 2200 Mhz, the media unit in the soc tile only 1300 Mhz.

GT0: Setting up Primary GT
GT1: Setting up Standalone Media GT

GT0: GUC: init took 10ms, freq = 2200MHz
GT1: GUC: init took 15ms, freq = 1300MHz


device info: EU total: 128
device info: EU per subslice: 16
device info: subslice total: 8


Intel(R) Core(TM) Ultra 7 1003H (family: 0x6, model: 0xaa, stepping: 0x4)

https://intel-gfx-ci.01.org/tree/linux-next/next-20230607/bat-mtlp-8/dmesg0.txt

https://intel-gfx-ci.01.org/tree/linux-next/next-20230607/bat-mtlp-8/boot0.txt

 

[eek2121]

Ultra 7 1003H seems to have a graphics frequency of 2200 Mhz, the media unit in the soc tile only 1300 Mhz.

https://intel-gfx-ci.01.org/tree/linux-next/next-20230607/bat-mtlp-8/dmesg0.txt

https://intel-gfx-ci.01.org/tree/linux-next/next-20230607/bat-mtlp-8/boot0.txt

Yep, and a 3.4ghz base clock with an unknown boost clock. Also unclear is the frequency of the “e” cores.

AMD’s 7840U has 8 (Zen 4) cores with a 3.3-5.1ghz clock range, so I suspect Intel will have no issues keeping up as long as boost clocks aren’t in the toilet.

 

[TESKATLIPOKA]

6nm A380 had 78W power consumption and clocked 2450MHz in Cyberpunk.TPU
It could be overclocked to 2696MHz.

Meteor Lake's IGP has pretty much the same specs except the memory subsystem and lower clocks.
If they can keep the average clocks close to Its turbo 2200-2300MHz even at lower TDP models, then that is already ~50% more, If you include the 33% increase then you have 2x more horsepower. I am not sure If It can beat Phoenix, but the difference will be pretty small.

The CPU part with 6P+8E cores will be performant, even Alder and Raptor Lake are performant, their problem was the power consumption.
If the power consumption goes down, which 3.4GHz base clock suggests, then I think It will be a pretty good CPU(APU).

P.S. TPU could retest A380 with newer drivers to see how much the performance improved, I would certainly be interested to see such a comparison.

 

[Khato]

6nm A380 had 78W power consumption and clocked 2450MHz in Cyberpunk.TPU
It could be overclocked to 2696MHz.

Meteor Lake's IGP has pretty much the same specs except the memory subsystem and lower clocks.
If they can keep the average clocks close to Its turbo 2200-2300MHz even at lower TDP models, then that is already ~50% more, If you include the 33% increase then you have 2x more horsepower. I am not sure If It can beat Phoenix, but the difference will be pretty small.

One open question with performance estimates based on existing Alchemist GPUs is whether MTL will be based on the same design or the tweaked refresh. While I know the latest rumors claim that Alchemist refresh GPUs were cancelled, that doesn't imply anything with respect to the MTL IGP implementation. And it'd be kinda odd if they had a more optimized design ready to go and didn't use it on MTL.

 

[mikk]

6nm A380 had 78W power consumption and clocked 2450MHz in Cyberpunk.TPU
It could be overclocked to 2696MHz.

This is the total board power including GDDR6. On Asrock A380 the GPU is limited to 55W PL1 and 60W PL2 at 2450 Mhz. MTL tGPU is build on a more efficient TSMC process and Xe LPG certainly will be even more optimized for a mobile low power environment versus the desktop A380.

 

[TESKATLIPOKA]

This is the total board power including GDDR6. On Asrock A380 the GPU is limited to 55W PL1 and 60W PL2 at 2450 Mhz. MTL tGPU is build on a more efficient TSMC process and Xe LPG certainly will be even more optimized for a mobile low power environment versus the desktop A380.

Didn't you read the rest of my post? I mean this part

Meteor Lake's IGP has pretty much the same specs except the memory subsystem and lower clocks.
If they can keep the average clocks close to Its turbo 2200-2300MHz even at lower TDP models, then that is already ~50% more, If you include the 33% increase then you have 2x more horsepower. I am not sure If It can beat Phoenix, but the difference will be pretty small.

I was talking about Meteor Lake IGP vs Raptor Lake IGP.
MTL's IGP looks like a big improvement.

 

[mikk]

Didn't you read the rest of my post? I mean this part

I was talking about Meteor Lake IGP vs Raptor Lake IGP.
MTL's IGP looks like a big improvement.

For sure it will be a big improvement with twice the computing power. I think depending on the IPC improvements on Xe LPG over Xe LP and bandwidth constraints an average improvement of 60-80% should be doable which is 680M-780M territory. MTL tGPU should be a better competitor than the dGPU version of Alchemist.

 

[TESKATLIPOKA]

For sure it will be a big improvement with twice the computing power. I think depending on the IPC improvements on Xe LPG over Xe LP and bandwidth constraints an average improvement of 60-80% should be doable which is 680M-780M territory. MTL tGPU should be a better competitor than the dGPU version of Alchemist.

But I also said in my post that MTL should be very close to Phoenix 780M. If they can somehow mitigate BW limit then maybe even faster, but would also depend on the game and If the drivers were optimized or not for It.

A380 was performing pretty badly at Full HD, but that review was done one year ago. Their drivers should have improved considerably, that's why I wanted to see how It performs now.

 

[SpudLobby]

Coercitiv, mikk, and eek2121 covered it pretty well. But I do want to highlight the word 'and' in your quote. That word should be 'or'. 20% more performance or 40% lower power. We aren't getting both. Not only that, but those numbers are only accurate when the processor is run at ~0.65 V. Look at the circles in this image. Intel 4 gets either 21% more frequency at the same power OR 40% less power at the same frequency.

AnandTech Forums: Technology, Hardware, Software, and Deals

Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

www.anandtech.com

But, now look at the square dots. At 0.85V and 3 GHz frequency, Intel 4 was only ~35% less power than Intel 7.

As you keep increasing the voltage, the possible frequencies get higher but the curves start to bend sharply upwards to even higher power. Extrapolate this out to 5 GHz or 6 GHz and the power could be massive. The Intel 4 node isn't optimized for high frequency use. The data isn't shown in the graph but it could be possible that MTL is just not going to perform well at ADL frequencies. Wait a few more months for Arrow Lake if that is your need.

I realize that it's not simultaneous, I know that RE: "And vs or" in the sense of iso-performance power savings vs iso-power performance gains.

However, it's not necessarily the wrong framing, because it's entirely possible that MTL at say 4GHz will save 40% on power, but won't realize 4.8GHz (which is the 20% higher frequency promised from Intel 4 iso-power albeit at a higher starting point than those VLSI slides, which is part of the discussion here) at the same 4GHz power that ADL had.

In other words, the "And" phrasing is not at all inappropriate, because it's a question of the product having the capability of displaying those qualities independently, which is theoretically possible as a straightforward implication of process node gains: you get A) savings or B) performance gain at the same power, and the product displays both of those qualities. However, we may not get this across the entire set of frequencies we care about. The fact that PL1 (apparently) does improve but PL2 does not I think makes this obvious.

I've seen those VLSI slides by the way. I'm well aware of all this, and remain fairly bullish on Intel 4, I would still be very surprised if Meteor Lake has negligible power gains in the modest frequency ranges^1.

Anyway, assuming TDP's across the leaked SKU's, if PL1 frequencies did improve due to some cache increases and process node gains, that's huge. I didn't realize that they did improve though, and it seems we have a strong disconnect between two caucuses on the subject of MTL leaks - albeit many don't seem to realize it.


1: Horrible yields and the IO draw might hurt but I don't see a reason to believe the former is as bad as 10NM was with the Ice Lake debacle and I don't see a reason to believe the latter will be as bad as say AMD's desktop IO dies which really don't emphasize power whatsoever.

 

[SpudLobby]

On the note of chiplets, importantly, something similar - I am assuming TSMC has a similar power-optimized interposer structure - is be a fantastic avenue to Intel, AMD - or even Qualcomm, Nvidia, MediaTek - building a cost-effective large APU at minimal cost to power and performance.

E.g. M1/2 Pro/Max lineup competitors with larger iGPU options. I think we will get more competition for the base lineup performance and Phoenix already basically offers that for casual gaming, MTL's iGPU will also be great, I'm sure Qualcomm will have a beefy Adreno.

But I don't think we'll see a non-Apple 300-400mm^2 SoC with larger iGPU's and larger bus widths to support that unless via some form of power-optimized chiplets.

 

[BorisTheBlade82]

And one more exciting bit of info that i came across recently:

AMD's Infinity Fabric is substrate-based (outdated tech). And Intel Foveros is interposer-based. Meaning, Foveros is light-years ahead of Infinity Fabric.

Interestingly, Foveros has up to 10X power-efficiency and very low latency compared to Infinity Fabric. Intel also has nextgen improved foveros too in the pipeline for ARL & LNL. Whereas, AMD has no such solution for Zen 5. I think AMD is in a lot of trouble. It appears MTL has managed to unleash the beasts Intel had in its unused arsenal.

Dude, this is extremely oooooold news. And just because you might only have stumbled across this recently does not make it interesting, much less game-changing.
As @Exist50 already pointed out, there are a lot of factors to, erm, factor in. One more being that the IFoP is much more flexible than Foveros. Intel could never produce something like a 12 CCD, 96c EPYC Server CPU with this.
And AMD is pretty much capable of using technology that is at least competitive to Foveros. See MI250, Navi31, MI300 for proof.
It all boils down to: Is it worth the costs in the grand scheme of things?

Meaning, it's not just going to be better than Zen 4. I think it's gonna be better than M2 macs as well in power efficiency. And it's coming soon...

If I were you, I wouldn't hold my breath for this - might get quite exhaustive.

 

[SpudLobby]

Just an MTL analysis. Rumors and leaks suggest MTL's ST performance uplift will be in a single digit which I totally agree. And we already know MTLs gonna be very power efficient. But by how much?

(1) The AT V/F curve suggests, Intel 4 provides 35% to 40% power efficiency which is in line with Intel's claims. Meaning, MTL can operate at significantly less power at iso-freq. I just did some math, and it clearly shows that a MTL cpu can easily run at less than 30W compared to an equivalent RPL cpu running at 45W. And thats just purely because of the node shrink alone.

(2) And rumors suggest MTL is getting DLVR. Thats another solid 15% to 20% boost in power efficiency.

(3) Rumors suggest that the LP E-cores in the SoC tile has the capability to completely turn off the CPU tile & GPU tile while idling which massively saves power.

(4) I've seen MTL ES idling at around 480 MHz (idle clock) which is impressive. Probably the SoC E-cores.

To summarize, if Intel can get all these tech to work in tandem with stable BIOS & proper Windows patches, MTL power-efficiency will be off the charts! Might easily provide all-day battery life for latops under standard productivity workloads. Meaning, it's not just going to be better than Zen 4. I think it's gonna be better than M2 macs as well in power efficiency. And it's coming soon...

The only sense I could even plausibly see MTL being more efficient than an M2 Mac is the same one in which Rembrandt is close or more efficient at 15-25W workloads where the M2 is at full tilt (measured from the wall, full package power) on a purely MT workload. With 4+4 and 4 of those littles only adding so much, if you measure at an arbitrary power point, you could absolutely find e.g. a MTL 2+8 or 6+8 unit more efficient at some high point on Apple's efficiency curve, in a purely multithreaded workload, even if the cores won't stand a chance on ST perf/w anywhere.

Otherwise though there's really no way. And in practice real workloads on a PC, mobile specifically, aren't invariably logging on to churn at 25W for two hours before plugging the system in. I mean, efficiency there is important still, but for me compilation/builds themselves aren't the majority of my time. Instead relatively snappy ST without gunning it to 15-20W & efficiency cores that can take care of background tasks at reasonable speeds & a fraction of the energy consumption a P core uses are invaluable, not to mention low idle draw.

Anyways, looking at where ADL is, optimistically shutting off the compute tile will be big for idle power, and more cache + Intel 4 could really take MTL closer to Zen 4 than currently. But I'm skeptical they're going to have an answer for Apple, even -40% on power iso-perf probably won't take Redwood Cove to Avalanche's league below 5-6 watts, nor put Intel's E Cores at Apple's class or even A7x class on energy consumption. Even AMD isn't there yet on the former.

The win for Intel would be putting MTL idle draw much closer to Ice/Tiger Lake, which seems feasible, and then some general power improvements under load from the process at moderate voltages. With that, I could see Intel back on the map as "acceptable" to good vs Zen 4. They'd have idle back to acceptable if not good ranges unlike ADL, but they'd have MT unlike with TGL and ICL SKUs (along with some ST improvements), some power improvement too, and notably I think ADL's iGPU is supposed to be pretty competitive.

 

[eek2121]

The only sense I could even plausibly see MTL being more efficient than an M2 Mac is the same one in which Rembrandt is close or more efficient at 15-25W workloads where the M2 is at full tilt (measured from the wall, full package power) on a purely MT workload. With 4+4 and 4 of those littles only adding so much, if you measure at an arbitrary power point, you could absolutely find e.g. a MTL 2+8 or 6+8 unit more efficient at some high point on Apple's efficiency curve, in a purely multithreaded workload, even if the cores won't stand a chance on ST perf/w anywhere.

Otherwise though there's really no way. And in practice real workloads on a PC, mobile specifically, aren't invariably logging on to churn at 25W for two hours before plugging the system in. I mean, efficiency there is important still, but for me compilation/builds themselves aren't the majority of my time. Instead relatively snappy ST without gunning it to 15-20W & efficiency cores that can take care of background tasks at reasonable speeds & a fraction of the energy consumption a P core uses are invaluable, not to mention low idle draw.

Anyways, looking at where ADL is, optimistically shutting off the compute tile will be big for idle power, and more cache + Intel 4 could really take MTL closer to Zen 4 than currently. But I'm skeptical they're going to have an answer for Apple, even -40% on power iso-perf probably won't take Redwood Cove to Avalanche's league below 5-6 watts, nor put Intel's E Cores at Apple's class or even A7x class on energy consumption. Even AMD isn't there yet on the former.

The win for Intel would be putting MTL idle draw much closer to Ice/Tiger Lake, which seems feasible, and then some general power improvements under load from the process at moderate voltages. With that, I could see Intel back on the map as "acceptable" to good vs Zen 4. They'd have idle back to acceptable if not good ranges unlike ADL, but they'd have MT unlike with TGL and ICL SKUs (along with some ST improvements), some power improvement too, and notably I think ADL's iGPU is supposed to be pretty competitive.

I've actually seen some comparisons between the 7950x, the 13900k, and the M2 Ultra. One of the issues I have with these comparisons is that the M2 is a laptop chip. If the comparisons had, say, a Ryzen 7945HX, I would agree with you, however, comparing Zen 4 or Raptor Lake S to an Apple mobile processor is an Apples to Oranges comparison.

Show me the lower power chips and where they stand.

Here is Ars Technica's chart on low vs. high power chips:

Even that one is suspect because with my own handbrake benchmarks, the 7950X at lower TDPs was MUCH more competitive than in this chart.

Also, the M2 Ultra is uncompetitive in terms of absolute performance. It only wins at perf/watt, so it is clearly a mobile oriented chip.