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.)

 

[Hitman928]

Is he related to Jensen Huang?

No idea, but I highly doubt it.

 

[Geddagod]

You didnt even look at the data, that just a remark out of clulessness; so much for the clownesque argument...

🤡

How could the 7840HS score be that lower than the 7840U at say 7-10W, this is exactly the same chip so it should have the same power/perf curve in this part of the graph.

Different SOC power, binning and preset voltage/frequency curves for yield, etc etc

Those are just the default 1t speeds of the respective CPUs. Where is this IPC scaling chart for RPC? At least according to the slides in the new video, RPC scales in IPC slightly better from 3.5 GHz to 4.7 GHz than RWC core scales from 3.6 GHZ to 4.4 GHz.

I think I see your point now. This is what Huang said.

 

[H433x0n]

When a new process is released


You didnt even look at the data, that just a remark out of clulessness; so much for the clownesque argument...

How could the 7840HS score be that lower than the 7840U at say 7-10W, this is exactly the same chip so it should have the same power/perf curve in this part of the graph.

The slope of its power/perf curve is just plain fairy tale since there s a segment where perf increase about linearly with power, whoever doesnt notice such a discrepancy is not qualified to comment the rest...

It’s a different gear for the memory controller.

 

[maddie]

Do 6VT and 8VT have area differences? I don't think they do, rather it's channel doping.

Then that's an expensive die. Multiple passes for the same layer.

 

[Thunder 57]

It seems that "Rentable units" is simply "Intelese" for the old concept of "reverse hyperthreading". Softmachines and their technology was discussed here many years ago and Anandtech even had an article. AMD, Intel and several other big names were investors until Intel bought the company outright. I always wondered when we would see real world results and I guess it will be soon.

For those interested and not wanting to make ridiculous ignorant claims, here it is. https://www.anandtech.com/print/10025/examining-soft-machines-architecture-visc-ipc

SPECint perf graph by same author. IPC regression on MTL is the biggest question mark, tile-base design is to blame?

https://twitter.com/x/status/1741391483578847613

View attachment 91288

Author with weifu girls? I don't trust any of those guys. No wonder birth rates are dropping.

 

[SiliconFly]

There IS. MTL has two different type of P core, one of the P core has lower voltage than the original one. Yeah I think it is a painful design.

I don't think it's possible. Mixing 6VT & 8VT cells in the same tile that too in the same P core cluster just doesn't sound feasible.

He says MTL has two P-cores because of 8VT/6VT Intel 4. The question would be how this works and what core is using the more efficient 6VT. It would make sense the two 1T cores are on 8VT because of the 1T Turbo.

Intel mentioned they're using power efficient cells for MTL on Intel 4. 8VT completely goes against those claims. Not for mobile. Doesn't sound right.

So MTL now is believed to have 4 core variations on the die?

Only 3 I guess.

There are 4 core types.

Probably only 3. 4th sounds more like a misinterpretation by a chinese blogger.

mikk is correct about this. MTL P cores has two different voltage settings. But unfortunately not all reviewers knew this.

Those 2 different voltage settings might not directly correspond to 2 types of P cores.

This seems to be quoting the same twitter source though? I was asking if there is anyone else making the same claim except this person on twitter. . .

Afaik, no.

Me neither, but Raichu also alluded to this.

Not sure. I might have missed it.

Do 6VT and 8VT have area differences? I don't think they do, rather it's channel doping.

8VT is the taller version with an extra fin I think to handle more power. Mainly for extremely high power & frequencies. Definitely not suitable for laptop CPUs.

Then that's an expensive die. Multiple passes for the same layer.

Too many passes (some for 6VT & some for 8VT) will severely lower yields.

 

[H433x0n]

Author with weifu girls? I don't trust any of those guys. No wonder birth rates are dropping.

He seems to be an AMD fan and only posts graphs with the best result possible for AMD as the comparison (comparing Windows vs Linux scores). If anything, he’s downplaying the results. The 2 graphs that give the most ideal apples/apples results he hasn’t produced (Crestmont v Gracemont) & (155H v 7840U v 7840HS).

Edit: Photoshop ninja from a discord I'm on created this graph that charts all of the relevant data points.

 

[FlameTail]

So... Will there a Meteor Lake N with only Crestmont E cores?

 

[Abwx]

Different SOC power, binning and preset voltage/frequency curves for yield, etc etc

Lol, you dont even know what you re talking about and are inventing difference that do not exist or at least gigantly overestimating the eventual tiny differences.

7840HS is the the same SoC as the 7840U and has same V/F curve, there s no binning that could result in such a vast difference, at 10W FI they should both be in the same curve but guess that when one is cluless about CPUs he s unable to spot even an elephant in a room...

 

[FlameTail]

Lol, you dont even know what you re talking about and are inventing difference that do not exist or at least gigantly overestimating the eventual tiny differences.

7840HS is the the same SoC as the 7840U and has same V/F curve, there s no binning that could result in such a vast difference, at 10W FI they should both be in the same curve but guess that when one is cluless about CPUs he s unable to spot even an elephant in a room...

Genuine Question: Then does AMD do no frequency binning at all for Phoenix?

 

[Geddagod]

Lol, you dont even know what you re talking about and are inventing difference that do not exist or at least gigantly overestimating the eventual tiny differences.

7840HS is the the same SoC as the 7840U and has same V/F curve, there s no binning that could result in such a vast difference, at 10W FI they should both be in the same curve but guess that when one is cluless about CPUs he s unable to spot even an elephant in a room...

massive 🤡 take
two CPUs can have the same SOC but have different uncore frequencies and V/F curves because they are targeting vastly different power levels.
But keep coping

dit: Photoshop ninja from a discord I'm on created this graph that charts all of the relevant data points.

For future reference, that's me lol
It was eyeballed, so if there are any mistakes I wouldn't mind them being pointed out so I can fix them

 

[Abwx]

massive 🤡 take
two CPUs can have the same SOC but have different uncore frequencies and V/F curves because they are targeting vastly different power levels.
But keep coping

Use less emoticons and more neurons, there s no CPU whose power can increase linearly with frequency.

The 7840HS has the same V:F curve as the 7840U, the difference between the two is that the former can be set at powers above 35W, this wont impact the V/F curve at the lower power range, actually the data about the 7840HS is bs and this cast a big doubt about the source s competence.

 

[Geddagod]

Use less emoticons and more neurons, there s no CPU whose power can increase linearly with frequency.

The 7840HS has the same V:F curve as the 7840U, the difference between the two is that the former can be set at powers above 35W, this wont impact the V/F curve at the lower power range, actually the data about the 7840HS is bs and this cast a big doubt about the source s competence.

Ye, the guy spent a thousand dollars on a SPEC licensee, went out and tested tens of machines, is incompetent, because the data doesn't hold up to your idiotic scrutiny, that the 7840hs and 7840u simply just can't have different v/f curves or different uncore presets lmao
Keep coping 🤡

 

[Abwx]

Ye, the guy spent a thousand dollars on a SPEC licensee, went out and tested tens of machines, is incompetent, because the data doesn't hold up to your idiotic scrutiny, that the 7840hs and 7840u simply just can't have different v/f curves or different uncore presets lmao
Keep coping 🤡

They have the same vf curve at low power, it s just that the 7840HS can be pushed at higher powers.

If this curve was right then it would mean that in MT with all 8 cores loaded the 7840HS woud use say 30W while the 7840U would be at 15W for a same Cinebench MT score for instance, that s just ridiculous...

Because that s basically what is displayed here, the 7840U is 2x more efficient in the lower power range, so it would still be so with all cores loaded at base frequency and something like 4W per core at most.

 

[H433x0n]

They have the same vf curve at low power, it s just that the 7840HS can be pushed at higher powers.

If this curve was right then it would mean that in MT with all 8 cores loaded the 7840HS woud use say 30W while the 7840U would be at 15W for a same Cinebench MT score for instance, that s just ridiculous...

Because that s basically what is displayed here, the 7840U is 2x more efficient in the lower power range, so it would still be so with all cores loaded at base frequency and something like 4W per core at most.

You’re missing the differences in both memory and fabric speed. The 7840HS is DDR5-5600, 7840U is LPDDR5-6400, both are running the fabric at different speeds (i.e. gears).

 

[Abwx]

You’re missing the differences in both memory and fabric speed. The 7840HS is DDR5-5600, 7840U is LPDDR5-6400, both are running the fabric at different speeds (i.e. gears).

That wont change things at that throughput and power level, beside we re talking of CPU package power, not full system power, the RAM has nothing to do here.

At base frequency each core take few watts, so just imagine the huge differences there would be between the HS and U PHX at say 30W with about 3.5W/core, one CPU would score 50% better than the other in MT at this power.

At some point when the discrepancy is that gigantic it s better to think about it rather than bringing infinitly small details with very few influence as likely explanations...

 

[Khato]

That wont change things at that throughput and power level, beside we re talking of CPU package power, not full system power, the RAM has nothing to do here.

At base frequency each core take few watts, so just imagine the huge differences there would be between the HS and U PHX at say 30W with about 3.5W/core, one CPU would score 50% better than the other in MT at this power.

At some point when the discrepancy is that gigantic it s better to think about it rather than bringing infinitly small details with very few influence as likely explanations...

While I'm not certain what exactly is being measured, I doubt it's just CPU package power. The 155H e-core power at 3W compared to 5W for the 155H LP e-core is the clearest example of such. Combined with the difference between 7840HS and 7840U I'd guess that the power figure also includes memory, potentially even full system.

There are three key differences between the 155H e-core and LP e-core - manufacturing process, frequency/power optimization, and cache. The manufacturing process and frequency/power optimization point probably end up with the LP e-core being just about as efficient as the e-cores on the compute tile, maybe a bit ahead. But the LP e-cores lack access to the L3 cache which may well be resulting in far higher memory utilization for the SPECint 2017 benchmark? That could explain the much higher power numbers.

Meanwhile between the 7840HS and 7840U traces the key difference is DDR5 5600 vs LPDDR5 6400. The DDR5 5600 is going to be slightly lower speed, but much lower latency and higher power. Which is exactly what we see in the different traces - shifted right by about 2W (higher memory power) and slightly higher performance (lower latency memory.)

 

[Geddagod]

While I'm not certain what exactly is being measured, I doubt it's just CPU package power. The 155H e-core power at 3W compared to 5W for the 155H LP e-core is the clearest example of such. Combined with the difference between 7840HS and 7840U I'd guess that the power figure also includes memory, potentially even full system.

There are three key differences between the 155H e-core and LP e-core - manufacturing process, frequency/power optimization, and cache. The manufacturing process and frequency/power optimization point probably end up with the LP e-core being just about as efficient as the e-cores on the compute tile, maybe a bit ahead. But the LP e-cores lack access to the L3 cache which may well be resulting in far higher memory utilization for the SPECint 2017 benchmark? That could explain the much higher power numbers.

Meanwhile between the 7840HS and 7840U traces the key difference is DDR5 5600 vs LPDDR5 6400. The DDR5 5600 is going to be slightly lower speed, but much lower latency and higher power. Which is exactly what we see in the different traces - shifted right by about 2W (higher memory power) and slightly higher performance (lower latency memory.)

I honestly think it's just worse power management on the SOC leading to a base increase of like a watt or something. I don't think any of these systems are being exactly tailored with their uncore/core power management for extremely low power consumption either, so that in combination with binning would easily make up the difference.
Zen 4's HS series idle power has been suspect for a while, with tests showing not much improvement or even regressions vs Rembrant in battery life. Infact, notebook check just ran a review citing that.
Plus, I think it's also important to note how Huang might have ran this testing. It's been brought to my attention that apparently you can't control Zen core's power directly, rather what Huang likely did was change package frequency limits or core frequency manually, and record data from there. This might impact the optimization of core/uncore power.
I do also think that your explanation might actually be another possibility though. That can also make a lot of sense, esp since he speaks a different language, and it could be a translation error.
I just don't think Huang pulled the data out of his butt lmao

 

[Abwx]

He seems to be an AMD fan and only posts graphs with the best result possible for AMD as the comparison (comparing Windows vs Linux scores). If anything, he’s downplaying the results. The 2 graphs that give the most ideal apples/apples results he hasn’t produced (Crestmont v Gracemont) & (155H v 7840U v 7840HS).

Edit: Photoshop ninja from a discord I'm on created this graph that charts all of the relevant data points.

View attachment 91302

You are unable to analyze the data, hence all what is left is to ridiculously brand someone a fan boy because of your limited knowledge.

From 7 to 12W the 7840HS score increase from 4.3 to 8.5, that s almost 2x the perf for 1.7x the power, so thats just a first point of total bs.

What is displayed is that with 8 cores loaded using each 7W, that is 56W, a 7840U would perform 62% better than a 7840HS, yet another prove that the 7840HS curve is totaly made up, no one serious with minimal knowledge would deem it as being credible.

 

[H433x0n]

You are unable to analyze the data, hence all what is left is to ridiculously brand someone a fan boy because of your limited knowledge.

From 7 to 12W the 7840HS score increase from 4.3 to 8.5, that s almost 2x the perf for 1.7x the power, so thats just a first point of total bs.

What is displayed is that with 8 cores loaded using each 7W, that is 56W, a 7840U would perform 62% better than a 7840HS, yet another prove that the 7840HS curve is totaly made up, no one serious with minimal knowledge would deem it as being credible.

I think the data is legitimate, he’s just annoyingly not providing all of the data points as I’d like to see them.

 

[Khato]

From 7 to 12W the 7840HS score increase from 4.3 to 8.5, that s almost 2x the perf for 1.7x the power, so thats just a first point of total bs.

How so? I'd agree if the 7W to 12W power figures were for the single active CPU core executing the benchmark, but that's clearly not the case. Given what I know of how these curves look, the active CPU core is probably in the 1-2W range for the 4.3 score and 6-7W range for the 8.5 score. That's 3-7x the power for 2x the performance.

 

[Abwx]

How so? I'd agree if the 7W to 12W power figures were for the single active CPU core executing the benchmark, but that's clearly not the case. Given what I know of how these curves look, the active CPU core is probably in the 1-2W range for the 4.3 score and 6-7W range for the 8.5 score. That's 3-7x the power for 2x the performance.

That would be the case for all other CPUs as well, and particularly for the 7840U, why would the 7840HS uncore use 3x the power while it s the same CPU..?.

If you look at some reviews the 7840HS idle power is the same as the 7840U, difference is the power at wich it can be boosted, since the 7840HS can use roughly 55W it will consume more when loaded with a lot of threads, but in single core it use the same power as the 7840U.

As i already said the 7840HS curve is boggus, there s a mistake that has been done by the guy who did the tests, there s no other explanation.

For the other curves that s another story, but given the blunder for the 7840HS i would be cautious to not take the rest at face value.

 

[Khato]

That would be the case for all other CPUs as well, and particularly for the 7840U, why would the 7840HS uncore use 3x the power while it s the same CPU..?.

If you look at some reviews the 7840HS idle power is the same as the 7840U, difference is the power at wich it can be boosted, since the 7840HS can use roughly 55W it will consume more when loaded with a lot of threads, but in single core it use the same power as the 7840U.

As i already said the 7840HS curve is boggus, there s a mistake that has been done by the guy who did the tests, there s no other explanation.

There are plenty of plausible explanations which have been provided in this thread.

I certainly won't claim that it's well presented and complete data. In fact, that's the primary reason why it's problematic to definitely state that there's a problem with any of the results. Certainly would be nice if there were reliable public sources for such benchmarks.

 

[mikk]

At minute 18 there he talks about different VT options on Intel 4:

Intel Tech Tour 2023-Intel4 Replay

Intel Tech Tour: Malaysia + Wiederholung des Meteor Lake Tech Day vom 21. bis 25. August 2023

www.intel.de

Here also a good explanation how it works or might work:

It is a little interesting that 2 types of options, 6VT (6 types of power supply voltage) and 8VT (8 types of power supply voltage), are offered. As shown on the right, to be precise, 3 to 4 types of voltage can be used for each of the NOTES and the MIMO (NMOS in increments of 190 mV and MIMO in increments of 180 mV), which allows a free balance between performance and power consumption. It is said that it can be adjusted to. Probably 6VT (3 types in short) does not have HVT, and the power consumption can be reduced by the amount that configures the circuit using 3 types of transistors of ULVT/LVT/SVT, but the power consumption increases sharply above 3GHz. On the other hand, 8VT can use HVT in addition to this, and by using this for the critical path, it seems that the increase in power consumption will be a little slower even if it exceeds 3GHz.

"Intel 4" process-Aiming for Commercialization in 2023 With 7nm, EUV Exposure and 20% Increase in Performance – DinoTechno

Intel 4 is a whole new process for Intel after a long time. Both 10nm SuperFin and Intel 7 are processes that are an extension of Intel 10nm.

dinotechno.com

 

[Hitman928]

8VT is the taller version with an extra fin I think to handle more power. Mainly for extremely high power & frequencies. Definitely not suitable for laptop CPUs.

The VTs refer to the threshold voltages or the necessary voltage applied to the gate to turn the transistor on. The taller or shorter you are referring to is the track height of the standard cells. They are not the same thing, though they tend to correlate somewhat. I'm not familiar with how Intel does their standard cell options though. In general, using a lower Vt device will allow you to reach a higher frequency without increasing the standard cell size at the cost of increased leakage.

At minute 18 there he talks about different VT options on Intel 4:

Intel Tech Tour 2023-Intel4 Replay

Intel Tech Tour: Malaysia + Wiederholung des Meteor Lake Tech Day vom 21. bis 25. August 2023

www.intel.de

Here also a good explanation how it works or might work:

"Intel 4" process-Aiming for Commercialization in 2023 With 7nm, EUV Exposure and 20% Increase in Performance – DinoTechno

Intel 4 is a whole new process for Intel after a long time. Both 10nm SuperFin and Intel 7 are processes that are an extension of Intel 10nm.

dinotechno.com

That quote for that website is either AI generated, a bad translation, from an author who doesn't really know what he is talking about, or some combination of the above.