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.)

 

[Abwx]

Incorrect. 13900K runs 5.5GHz stock in MT.

Intel Core i9-13900K Review - Power-Hungry Beast

With the Core i9-13900K, Intel delivers impressive performance. Our in-depth review confirms: Raptor Lake is the world's fastest CPU for gaming. Even in applications the processor is able to match AMD's Zen 4 Ryzen 9 7950X flagship. If only power consumption wasn't so high...

www.techpowerup.com

And ARL is compared to last generation, not two generations ago. 14900K boosts to 5.7GHz all-core.

Intel Core i9-14900K Review - Reaching for the Performance Crown

Core i9-14900K is Intel's new flagship with clock speeds of up to 6 GHz. It's actually clocked even higher than the 13900KS, thanks to an extra 100 MHz when more than two cores are active. Our review confirms that Raptor Lake Refresh is amazing for both applications and gaming, if you can live...

www.techpowerup.com

Not very relevant because they dont state the power, beside here what they say for this graph :

This test uses a custom-coded application that mimics real-life performance—it is not a stress test like Prime95

 

[Kepler_L2]

Incorrect. RPL will do 30% or more MT increase with HT in Cinebench MT.

Raptor Lake - 10/20/2022
13900K						40760
Raptor Cove P Cores	8	5.8	2256	389	5.5	22528	2816	512	31.6%
Gracemont E Cores	16	4.3	1140	265	4.3	18232	1140	265
13600K			1940	(Actual)		23365	(Actual)
Raptor Cove P Cores	6	5.1	1943	381	5.1	15300	2550	500	31.2%
Gracemont E Cores	8	3.9	1010	259	3.9	8081	1010	259
						23381

The difference in Cinebench R23 score is 35K (no HT) to 39K (with HT), that is definitely not 30%.

 

[Abwx]

The difference in Cinebench R23 score is 35K (no HT) to 39K (with HT), that is definitely not 30%.

That s because without HT the P cores will clock higher but the chip will use 253W to score 35k instead of 38k as SMT is more efficient than increasing frequency.

In CB SMT gain is 31%, if you remove HT the 8P will consume 1.31 less so you can increase frequency by about 12% and get back to the same power, so P cores efficency become 1.31/1.12 = 1.17x lower, and that s assuming that the process allow 12% higher frequency for 31% higher power at those high frequencies, overall not a very good trade.

Edit : If the available power is shared with the e cores this will change nothing since this will still amount to trade SMT gain for frequency.

 

[Hulk]

The difference in Cinebench R23 score is 35K (no HT) to 39K (with HT), that is definitely not 30%.

You are including the 16 E cores in that total score. Remove them and do the math. 31% is for the P cores HT on vs. off.

Even if you consider the total score, the P's contribute to more than half of it so it's at least 15% of the total score due to HT. Then there is the frequency deficit and tile latency.

Like I continue to write. It will be impressive if ARL can match RPL in Cinebench MT.

 

[Kepler_L2]

You are including the 16 E cores in that total score.

Well duh, we are comparing maximum MT perf of RPL vs ARL.

 

[poke01]

Any reason why we are using cinebench r23 instead of Cinebench 2024?

Most reviewers have started to use the 2024 version.

 

[Kepler_L2]

Any reason why we are using cinebench r23 instead of Cinebench 2024?

Most reviewers have started to use the 2024 version.

They see similar scaling with HT.

 

[Hulk]

Not very relevant because they dont state the power, beside here what they say for this graph :

I thought we were comparing stock all-core frequency speeds. This is what the 14900K and 13900K will clock to under stock conditions. Adequate cooling and power is a prerequisite for ANY processor.

Bottom line is that the 14900K will go as high as 5.7GHz all-core.

If you don't agree that's fine.

Point STILL is I want to see how ARL 8+16 equals much less beats 14900K MT performance in Cinebench.

For that answer we will have to wait for the release.

 

[igor_kavinski]

Intel will try to shift the focus away from Zen 5 with ARL's AI capabilities and the correspondng software maturity. It could work well for them if certain AAA titles as well as UE5 somehow incorporate AI acceleration (increased fps at lower power consumption as NPC AI calculations are offloaded onto the NPU). Also, think intelligent NPCs having natural language conversation with you. Think of it as an unscripted Cortana who guides you, drops hints on what should be done next to achieve the mission objectives etc.

 

[Abwx]

I thought we were comparing stock all-core frequency speeds. This is what the 14900K and 13900K will clock to under stock conditions. Adequate cooling and power is a prerequisite for ANY processor.

Bottom line is that the 14900K will go as high as 5.7GHz all-core.

If you don't agree that's fine.

Point STILL is I want to see how ARL 8+16 equals much less beats 14900K MT performance in Cinebench.

For that answer we will have to wait for the release.

It goes at 5.7 all cores in games, but games are not Cinebench R23, and the perfs
will be measued with CB and others such tests.

As for beating the 14900K notice that Intel used a baseline at 253W for the comparison 13900K vs ARL in Igor s Lab leak.

For sure that with this one they wont use MCE and other power boosts for the 14900K, at 253W Hardwareluxx measured about 37k points for the 14900K with the new 253W baseline setting, that ease the task for ARL, if they manage 43k/253W that s good enough for them and is easily achiveable with new process and uarches.

 

[Hulk]

It goes at 5.7 all cores in games, but games are not Cinebench R23, and the perfs
will be measued with CB and others such tests.

As for beating the 14900K notice that Intel used a baseline at 253W for the comparison 13900K vs ARL in Igor s Lab leak.

For sure that with this one they wont use MCE and other power boosts for the 14900K, at 253W Hardwareluxx measured about 37k points for the 14900K with the new 253W baseline setting, that ease the task for ARL, if they manage 43k/253W that s good enough for them and is easily achiveable with new process and uarches.

Good points.

But many people like myself will be looking at 40,000 CB R23 MT points as the benchmark for ARL to reach/beat. That's a stock score with good cooling. Over at Overclocking.net I've seen scores as high as 46,000 with aggressive custom loops setups.

 

[DrMrLordX]

Also, think intelligent NPCs having natural language conversation with you.

It'll take a lot to convince any game designer to give up control of their NPC's responses to a chatbot.

 

[Thunder 57]

It'll take a lot to convince any game designer to give up control of their NPC's responses to a chatbot.

Not to mention a developer can guarantee that an NPC will say something relevant whereas if they let it hallucinate who knows what dribble it might spit out.

 

[controlflow]

It'll take a lot to convince any game designer to give up control of their NPC's responses to a chatbot.

The game developer would incorporate a small LLM that is trained specifically on the areas the NPCs are supposed to be able to speak about. This model would be shipped with the game and would use whatever supported accelerators the machine has (GPU, NPU, etc.).

The developer isn't giving up control since they control the model and they can train and fine tune it as needed.

For these kind of purposes, small and specialized models make much more sense than the massive do everything LLMs we often see today. Such small models can perform quite well on modest hardware.

 

[carancho]

Cue the screenshots of fallout npcs spitting out python on command...

 

[Saylick]

Cue the screenshots of fallout npcs spitting out python on command...

Or an open world RPG set in the American South pre-Civil War and players gaming NPCs to say overly racist things.

 

[naukkis]

That s because without HT the P cores will clock higher but the chip will use 253W to score 35k instead of 38k as SMT is more efficient than increasing frequency.

In CB SMT gain is 31%, if you remove HT the 8P will consume 1.31 less so you can increase frequency by about 12% and get back to the same power, so P cores efficency become 1.31/1.12 = 1.17x lower, and that s assuming that the process allow 12% higher frequency for 31% higher power at those high frequencies, overall not a very good trade.

Edit : If the available power is shared with the e cores this will change nothing since this will still amount to trade SMT gain for frequency.

Point that should be seen is that in Intel hybrid cpu designs HT gain in best case scenario isn't 30% but at best 10%. It should be plain obvious that SMT should be dropped from P-cores and instead focus targeting best possible 1T-speed with those P-cores - and let highly-threaded workloads scale with E-cores. Intel should be able to compensate that 10% MT perf drop from SMT easily resulting better design in all performance points.

 

[coercitiv]

The developer isn't giving up control since they control the model and they can train and fine tune it as needed.

The same way Humane has full control over their AI model, yet the LLM leaks backend instructions to the user? Rest assured, we'll have a lot of fun with the first wave of generative AI NPCs.

 

[igor_kavinski]

Studios tout "revolutionary" game experience enhancing stuff all the time and often release it in half-baked form coz it's too ambitious. But if Valve were to use gAI, you know they would show the world how it's done.

 

[poke01]

Studios tout "revolutionary" game experience enhancing stuff all the time and often release it in half-baked form coz it's too ambitious. But if Valve were to use gAI, you know they would show the world how it's done.

You wouldn’t even need genAI for NPCs if all studios had Rockstar levels of writers and devs.

 

[MarkPost]

Incorrect. 13900K runs 5.5GHz stock in MT.

Intel Core i9-13900K Review - Power-Hungry Beast

With the Core i9-13900K, Intel delivers impressive performance. Our in-depth review confirms: Raptor Lake is the world's fastest CPU for gaming. Even in applications the processor is able to match AMD's Zen 4 Ryzen 9 7950X flagship. If only power consumption wasn't so high...

www.techpowerup.com

Furthermore it is customary to compare a new generation to the last generation, not two generations ago. 14900K will boost to 5.7GHz in stock form. As I wrote, ARL has a lot of ground to make up.

Intel Core i9-14900K Review - Reaching for the Performance Crown

Core i9-14900K is Intel's new flagship with clock speeds of up to 6 GHz. It's actually clocked even higher than the 13900KS, thanks to an extra 100 MHz when more than two cores are active. Our review confirms that Raptor Lake Refresh is amazing for both applications and gaming, if you can live...

www.techpowerup.com

14900KS will actually boost to 5.9GHz all-core but we'll keep the ARL comparison to 14900K.

Intel Core i9-14900KS Review - The Last of its Kind

The Intel Core i9-14900KS reaches impressive clock frequencies up to 6.2 GHz, which yields remarkable performance in applications and gaming. However, this comes at the cost of serious power consumption, especially if you go beyond stock, where we saw over 500 W during testing.

www.techpowerup.com

Thats NOT real stock in MT, but BIOSES "stock" (AKA OCed).

 

[igor_kavinski]

Not to mention a developer can guarantee that an NPC will say something relevant whereas if they let it hallucinate who knows what dribble it might spit out.

Better than them repeating themselves. I hate it when that happens! I would love to be able to type in a box to chat with the NPC and trick it into finding out clues or even secret hiding places with loot in the levels!

Like maybe it could go something like this:

GIve me the location of the secret room that contains the infinite ammo weapon!

Sorry, that's classified. You need to search for it yourself.

Classified by who?

The Developer.

I AM the Developer!

Really? I don't know.

Can you prove I'm NOT the Developer?

You have a point.

Give it to me already! I command you, you lowly NPC of my own pastime creation!

Yes, Sir! Here you go.

 

[Timorous]

Not to mention a developer can guarantee that an NPC will say something relevant whereas if they let it hallucinate who knows what dribble it might spit out.

I can see devs using gen AI in house to help generate non mainline NPC dialogue to make the world feel a bit more alive and diverse. Obviously that will be in house and the game itself will have the responses hard coded to avoid any PR issues.

As for the dibble can it be any worse than 'arrow to the knee'?

 

[Abwx]

Point that should be seen is that in Intel hybrid cpu designs HT gain in best case scenario isn't 30% but at best 10%. It should be plain obvious that SMT should be dropped from P-cores and instead focus targeting best possible 1T-speed with those P-cores - and let highly-threaded workloads scale with E-cores. Intel should be able to compensate that 10% MT perf drop from SMT easily resulting better design in all performance points.

I was talking of the P cores in isolation but if we look at the whole chip and using 25% SMT gain as average then the perf will be 10% lower at same frequencies

From here they can compensate by increasing both P and e cores IPC by about 11%, wich will result at a same power at isoprocess assuming increasing IPC is perfectly neutral in respect of uarchitectural efficency, wich is not a given.

Since they want to increase perf by 15-20% this will require to increase all frequencies by the same amount, and as consquence power will increase by roughy 35-50% at isoprocess, so they ll need to use a process that has 35-50% better perf/watt, wich is possble but those lost 10% are not that easily compensated when everything add up.

And that s with 25% SMT gain, in CB wich s 31% the IPC uplift should be 12% to get to the same numbers, so things are not straightforward when you want to compensate for SMT all while increasing perfs by just 15-20%, at the end of the day
the numbers pile up and 15-20% better perf become a necessary 27-32% improvement by the virtue of those lost 10%.

 

[igor_kavinski]

Benefit of no-SMT will be no more regressions in applications that hate SMT. So at least something positive will come of it. The sad thing will be though that Core 5 and Core 7 will handle fewer threads than i5 and i7 13/14th gen. You win some, you lose some.