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

Tigerick · Aug 22, 2022

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	12 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 (N3B)	Lunar Lake	Panther Lake
Platform	Mobile H/U Only	Desktop & Mobile H&HX	Mobile U Only	Mobile H
Process Node	Intel 4	TSMC N3B	TSMC N3B	Intel 18A
Date	Q4 2023	Desktop-Q4-2024 H&HX-Q1-2025	Q4 2024	Q1 2026 ?
Full Die	6P + 8P	8P + 16E	4P + 4E	4P + 8E
LLC	24 MB	36 MB ?	12 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.)

Josh128 · Oct 6, 2024

No way the 46K Passmark MT score is all 285K can push. Its an obvious MT sandbag. There is zero reason for it to be less than 14900K when in ST the 285K is ahead by 11%. It will be ahead of 14900K and likely a bit ahead or tied with the 9950X.

511 · Oct 6, 2024

OneEng2 said:
The numbers I have read have 18A somewhere between N3P and N2 and in some respects equal to N3P. N3B is better in all respects as I understand it to other N3X variants, just more expensive to use.

What do you base your statement on GNR having a superior memory accelerator than Turn on? I haven't heard this before.

Turn on 🤣🤣 anyway N3B is only good in SRAM by a very miniscule margin and slightly denser than other N3 flavours i forgot to put comma there it has MRDIMMs and the accelerators
IAA/QAT they are quite useful for the purpose they are designed they can lock few cores with accelerators and rest of the cores for their workload i remember seeing Third party benchmark for these

OneEng2 said:
Intel can't bleed cash forever. Currently it looks like their plan won't work without the foundry being lifted up by external customers using it. This makes sense as the exponential cost of new process technology has made it very hard for an IDM to justify new nodes when those new nodes must be exclusively paid for by a single chip program.

Yes that's the entire purpose for begging US Government anyway Manufacturing is nearly impossible without Government Support a good example is Ireland supports Intel quite well TSMC/Samsung has benefited from government support it is a bit unfair to their Manufacturing ( only manufacturing no Design) if US won't Support Intel they should be ready to accept defeat in leading edge semi manufacturing to TW/SK also Intel's best chance to gain foundry customers i semi custom chips for cloud customers something no body else can profit on as much as Intel

igor_kavinski · Oct 6, 2024

Josh128 said:
It will be ahead of 14900K and likely a bit ahead or tied with the 9950X.

It has better cores but fewer threads than both of those competitors. That's where it's gonna hurt especially in workloads that can be parallelized well. But there is hope that it will do well with something like 9600 MT/s RAM.

Hulk · Oct 6, 2024

Can someone with Zen 5 run CB R23 MT and ST and post the scores and the average clock during the run using HWinfo?

MS_AT · Oct 6, 2024

Hulk said:
Double checked the numbers.

8P @5.5GHz with HT on - CB R23 MT score is 22,783, package power is 173W

8P @5.5GHz no HT plus 5 E at 4.4GHz - CB R23 MT score is 22,620, package power is 166W.

So at 5.5GHz and 4.4GHz for P's and E's, respectively, it takes 5 E's to equal 8 P logical processors in CB R23 MT.

Package power is actually less without HT for the same score.

Just wanted to put out the correct numbers.

So at least for this scenario it would take 8/5*16=25.6 logical P cores at 5.5GHz to equal 16 physical E cores at 4.4GHz.
Not that the calculation means anything because physical cores scale with logical cores in this case but I just wanted to know what those 16E's are worth in terms of logical P cores.

Finally, Intel could have simply removed HT from the P's in Raptor and only had to add 5 E's to make up for it in MT in the most extreme (linear scaling) case.

They went one better and dramatically increased the compute with Skymont.

Now the question is what takes more area, the control hw for HT on P-cores or 5 e-cores

Additionally, while it is nice you have carried out those measurements, they really represent only one workload, that is CB23, so I wouldn't so easily reach the conclusion that 14900k with 8 P Cores and 21 E cores would be generally better than 14900k we got now.

Hulk · Oct 6, 2024

MS_AT said:
Now the question is what takes more area, the control hw for HT on P-cores or 5 e-cores

Additionally, while it is nice you have carried out those measurements, they really represent only one workload, that is CB23, so I wouldn't so easily reach the conclusion that 14900k with 8 P Cores and 21 E cores would be generally better than 14900k we got now.

HT transistors vs 5 additional E's? Don't know, probably comparible.

Considering the 21 E cores would distribute the heat on the die better than the concentrated hot spot resulting from HT in the P's, along with the fact that the P's would probably clock higher or last longer with less heat, I personally would prefer 8+21 with no HT. Apparently Intel agrees as well because Arrow Lake is following that that.

But we're in fantasy land regardless. My point was to simply post some data relevent to this thread.

naukkis · Oct 6, 2024

MS_AT said:
Additionally, while it is nice you have carried out those measurements, they really represent only one workload, that is CB23, so I wouldn't so easily reach the conclusion that 14900k with 8 P Cores and 21 E cores would be generally better than 14900k we got now.

That SMT-less cpu has for best MT 29 total threads and 8 strong threads. With HT current 14900k has 32 weak threads for MT. It's not even a contest which one would be better for every possible workload.

lightmanek · Oct 6, 2024

Hulk said:
Can someone with Zen 5 run CB R23 MT and ST and post the scores and the average clock during the run using HWinfo?

Score here, I will re-run with HWInfo in few hours.

Page 10 - Discussion - Zen 5 Builders thread

Page 10 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

This is slightly tweaked 103MHz FCLK, I can run stock if that's better for your comparison.

OneEng2 · Oct 6, 2024

Josh128 said:
No way the 46K Passmark MT score is all 285K can push. Its an obvious MT sandbag. There is zero reason for it to be less than 14900K when in ST the 285K is ahead by 11%. It will be ahead of 14900K and likely a bit ahead or tied with the 9950X.

It has 24 cores and 24 threads. 16 of those cores/threads are at a 25% performance deficit to even the Zen 5c cores running SMT and 9950x has 16 full Zen 5 cores which I am guessing are more like 40% more powerful than skymont cores and 15-25% more powerful than Lion Cove in multi-threaded workloads. It is only when you had the 14900K having 24 cores (8+16) and 32 threads that Intel had a multi-threaded advantage as the 8 p cores held their own against 8 of the 16 Zen cores, and the 16 e cores were more than a match for the 8 remaining Zen cores in multithreaded workloads.

Intel would have to have boosted its single core performance by a great deal more than 11% in order to make up for the loss of SMT and the loss of core count IMO.

igor_kavinski said:
It has better cores but fewer threads than both of those competitors. That's where it's gonna hurt especially in workloads that can be parallelized well. But there is hope that it will do well with something like 9600 MT/s RAM.

Exactly. The bandwidth should provide it with some benchmark wins for sure. We will see soon enough.

From a strategic standpoint, it feels like losing SMT wasn't that great an idea .... but we will see soon enough. Arrow Lake is out on the 10th of Oct, and we should see the Zen 5 based Turin soon.

jdubs03 · Oct 6, 2024

OneEng2 said:
It has 24 cores and 24 threads. 16 of those cores/threads are at a 25% performance deficit to even the Zen 5c cores running SMT and 9950x has 16 full Zen 5 cores which I am guessing are more like 40% more powerful than skymont cores and 15-25% more powerful than Lion Cove in multi-threaded workloads. It is only when you had the 14900K having 24 cores (8+16) and 32 threads that Intel had a multi-threaded advantage as the 8 p cores held their own against 8 of the 16 Zen cores, and the 16 e cores were more than a match for the 8 remaining Zen cores in multithreaded workloads.

Intel would have to have boosted its single core performance by a great deal more than 11% in order to make up for the loss of SMT and the loss of core count IMO.

Exactly. The bandwidth should provide it with some benchmark wins for sure. We will see soon enough.

From a strategic standpoint, it feels like losing SMT wasn't that great an idea .... but we will see soon enough. Arrow Lake is out on the 10th of Oct, and we should see the Zen 5 based Turin soon.

Intel’s own performance targets were showing a bigger uplift in MT than in ST. So that should have some credence when thinking about these things. Will post again as a reminder.

The rumors of ARL performance have been pretty much all over the place, it’s pretty hard to get a read on it.

IMO, treading water in MT would be a big disappointment even if they do get the power consumption down to Zen 5 levels. The expectation was 10 to 15% improvement.

Josh128 · Oct 6, 2024

OneEng2 said:
It has 24 cores and 24 threads. 16 of those cores/threads are at a 25% performance deficit to even the Zen 5c cores running SMT and 9950x has 16 full Zen 5 cores which I am guessing are more like 40% more powerful than skymont cores and 15-25% more powerful than Lion Cove in multi-threaded workloads. It is only when you had the 14900K having 24 cores (8+16) and 32 threads that Intel had a multi-threaded advantage as the 8 p cores held their own against 8 of the 16 Zen cores, and the 16 e cores were more than a match for the 8 remaining Zen cores in multithreaded workloads.

Intel would have to have boosted its single core performance by a great deal more than 11% in order to make up for the loss of SMT and the loss of core count IMO.

lol. no. The score given is 10% lower than a 24 thread 7900X from two years ago. And Intel has indeed boosted its single core performance by > 11% on 8 of the cores, and a great deal more (~30%+) on the remaining 16 cores. I dont know how many times it needs to be said, threads do not equal physical cores.

OneEng2 · Oct 6, 2024

Josh128 said:
lol. no. The score given is 10% lower than a 24 thread 7900X from two years ago. And Intel has indeed boosted its single core performance by > 11% on 8 of the cores, and a great deal more (~30%+) on the remaining 16 cores. I dont know how many times it needs to be said, threads do not equal physical cores.

AFAIK, Zen 5c has ~5% single core integer performance better than Skymont. In multi-threaded loads, Zen 5c would be about 30% better than Skymont. I do agree, SMT threads do not equal the performance of a full core; however, a SINGLE Zen 5c core will perform like 1.3 skymont cores in a multi-threaded application. A Single full Zen 5 core will perform more like 1.5 (or more) skymont cores in a multi-threaded application.

It is therefore not surprising that Arrow Lake will be bested in multi-threaded applications by both 14700K and 9950x by a decent margin. It is just math.

If you have differing information, please post.

OneEng2 · Oct 6, 2024

jdubs03 said:
Intel’s own performance targets were showing a bigger uplift in MT than in ST. So that should have some credence when thinking about these things. Will post again as a reminder.
View attachment 108779
View attachment 108780
The rumors of ARL performance have been pretty much all over the place, it’s pretty hard to get a read on it.

IMO, treading water in MT would be a big disappointment even if they do get the power consumption down to Zen 5 levels. The expectation was 10 to 15% improvement.

I hear you, and I see what Intel is saying. I just can't reconcile how the math works against 9950x in multi-threaded applications.

Since 14700K was essentially equal to 9950x in single core performance, and Arrow Lake is 11% higher than both, it is hard to see how Arrow Lake would keep up with the Zen 5 X3D versions being released this year. In fact, I am sure AMD moved the X3D launch forward this year when they realized that Arrow Lake had gained the single thread advantage to ensure the lead would be short lived.

Its the multi-treaded workloads, specifically in data center applications, that I see as being strategic since this is where the highest growth and highest profit margin is. I am trying to see Intel's logic for removing SMT from their core architecture and all I can come up with is that they have shifted focus to laptops and desktops where low thread count applications are the norm and even current CPU's are more than enough for the task. Laptops in specific will greatly benefit from a design that focuses more on power efficient low thread count applications. LNL is a great example of how this pays off.

It's kind of funny when you think about it. Before AMD had SMT in their core design, Intel dominated most server benchmarks. AMD blustered that they thought full cores were better. My how times have changed.

Within the CPU core, I suspect that SMT design is the best way to maximize multi-threaded performance within a power window. If that is true, then Intel made a strategic mistake removing it.

We will see.

Doug S · Oct 6, 2024

OneEng2 said:
Every other manufacturer that brought GAA to production was able to get the process to the point of creating chips, and had very bad issues trying to get yields up to the point where the process made money and was able to support high volume.

That "every other manufacturer" you mention is exactly one, Samsung. And they've had problems ever since they started using EUV so I don't think this necessarily says anything about GAA.

TSMC reported reaching their target yield to enter risk production of N2 in late May - that's 80%. So already N2 yields with GAA are significantly higher than their N3B yields were when they entered mass production for Apple (not publicly revealed but rumored to be in the 50-60% range, and Samsung's "3nm" yields were rumored to be much worse than even that) and they officially entered risk production phase the first week of July.

Now that's no guarantee they won't hit some unknown roadblock and won't continue to their usual smooth improvement to hit their 90% mass production target, but they seem to be on their usual trajectory, and usually hit mass production targets in under a year after reaching their risk production target.

While you can't say "because TSMC can do GAA so can Intel" you also can't say "because Samsung can't do GAA neither can Intel".

lightmanek · Oct 6, 2024

Hulk said:
Can someone with Zen 5 run CB R23 MT and ST and post the scores and the average clock during the run using HWinfo?

Here are almost stock settings - FCLK 100, PPT 205, I forgot to disable Core Boost so ST can boost +100 from stock, no Core Optimizer.

HWInfo while running ST

HWInfo while running MT

Run with HWInfo closed for proper result at these settings

Josh128 · Oct 6, 2024

OneEng2 said:
Since 14700K was essentially equal to 9950x in single core performance, and Arrow Lake is 11% higher than both, it is hard to see how Arrow Lake would keep up with the Zen 5 X3D versions being released this year. In fact, I am sure AMD moved the X3D launch forward this year when they realized that Arrow Lake had gained the single thread advantage to ensure the lead would be short lived.

Where do you see 14700K equaling 9950X in ST perf? Im seeing 9950X ST in the charts as 6% faster.

If you have differing information, please post.

Here you go. Geekbench 6 runs of ES samples are in line with 9950X in MT, and slightly better than 14900K.

Det0x · Oct 6, 2024

igor_kavinski said:
In games mostly. MT perf looks bad.

Must be something wrong with MT

My ES sample

lightmanek · Oct 6, 2024

Josh128 said:
Where do you see 14700K equaling 9950X in ST perf? Im seeing 9950X ST in the charts as 6% faster.

Here you go. Geekbench 6 runs of ES samples are in line with 9950X in MT, and slightly better than 14900K.

View attachment 108788

I think he was referring to that PassMark test, where it shows 14700K = 9950X. There is a story with that test and website regarding ranking (Reddit thread ), but to me it is still a data point I just tend to put less weight on it.

Hulk · Oct 6, 2024

lightmanek said:
Score here, I will re-run with HWInfo in few hours.

Page 10 - Discussion - Zen 5 Builders thread

Page 10 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

This is slightly tweaked 103MHz FCLK, I can run stock if that's better for your comparison.

Thanks, I would prefer stock just to compare to other unoptimized scores.
You could run without HWinfo running for the scores, then run again with HWinfo running to get the average frequency during the run as well as package power while you're at it if possible? This way HWinfo doesn't interfere with the actual score.

Hulk · Oct 6, 2024

lightmanek said:
Here are almost stock settings - FCLK 100, PPT 205, I forgot to disable Core Boost so ST can boost +100 from stock, no Core Optimizer.

HWInfo while running ST
View attachment 108787

HWInfo while running MT
View attachment 108789

Run with HWInfo closed for proper result at these settings
View attachment 108790

I don't understand why the clocks are all over the place? Can you reset the counter right after the boxes start getting rendered and then take the screenshot right before it's completed so we can see what the average clocks look like during the run? I understand the clocks throttle down as more cores are engaged but these seem to go up and down.

Hulk · Oct 6, 2024

Some charts from my own historical data. I've owned most of these processors or found enough corroborated data I felt confident enough with it to include in the chart.

My systems are day-to-day work, not optimized. I would just reboot, run CB, then run again restarting HWinfo counter at the beginning of the run and then checking average clock at the end of the run. Not trying to "win" contests, just seeing how the CPU's do basically at stock.

For the preliminary Zen 5 data here I used scores from this thread and clocks I acquired from TechReport. They always get boost frequencies very nicely. https://www.techpowerup.com/review/amd-ryzen-9-9950x/26.html
Assuming Lightmanek's 9950X is running approximately the same clocks. I took the average of the TechReport clocks.

I had a Broadwell system but never got any good data from it. Anyone have any Broadwell numbers?

Alder/Raptor/Raptor Refresh are a little harder to isolate E's and P's. I've gone into it before but I'm pretty confident with my numbers.

Again, I'm not posting this to show who is better or worse. Just trying to look at historical performance trends for this one specific benchmark, which of course only tests rendering and in a specific way.

Det0x · Oct 6, 2024

Hulk said:
Some charts from my own historical data. I've owned most of these processors or found enough corroborated data I felt confident enough with it to include in the chart.

My systems are day-to-day work, not optimized. I would just reboot, run CB, then run again restarting HWinfo counter at the beginning of the run and then checking average clock at the end of the run. Not trying to "win" contests, just seeing how the CPU's do basically at stock.

For the preliminary Zen 5 data here I used scores from this thread and clocks I acquired from TechReport. They always get boost frequencies very nicely. https://www.techpowerup.com/review/amd-ryzen-9-9950x/26.html
Assuming Lightmanek's 9950X is running approximately the same clocks. I took the average of the TechReport clocks.

I had a Broadwell system but never got any good data from it. Anyone have any Broadwell numbers?

Alder/Raptor/Raptor Refresh are a little harder to isolate E's and P's. I've gone into it before but I'm pretty confident with my numbers.

Again, I'm not posting this to show who is better or worse. Just trying to look at historical performance trends for this one specific benchmark, which of course only tests rendering and in a specific way.

View attachment 108800

View attachment 108801

If it helps you

ST score @ static 5.7ghz

MT score @ static 5.4ghz

DavidC1 · Oct 6, 2024

OneEng2 said:
If you have differing information, please post.

Why would it be slower than Raptorlake? The 16 E cores were roughly equal to 8 P cores in MT, and that one is getting boosted by 30%, while the other half is losing maybe 10%. It's a net gain.

lightmanek · Oct 6, 2024

Hulk said:
I don't understand why the clocks are all over the place? Can you reset the counter right after the boxes start getting rendered and then take the screenshot right before it's completed so we can see what the average clocks look like during the run? I understand the clocks throttle down as more cores are engaged but these seem to go up and down.

Clocks are quite stable for me as I'm on a custom loop, so the 5025MHz and 4725MHz are about right. They fluctuate +/-50MHz during the whole test as I'm far from throttling temperatures and pegged at PTT limit the whole time.

Josh128 · Oct 6, 2024

lightmanek said:
I think he was referring to that PassMark test, where it shows 14700K = 9950X. There is a story with that test and website regarding ranking (Reddit thread ), but to me it is still a data point I just tend to put less weight on it.

I know, but my belief on it is that this sample was purposely power limited for the MT test as to not give away too much too early. Can anyone here say with a straight face that they think a 7900 non-X is going to score higher in the MT of this test than a 285K? To me, thats a clear giveaway. I referenced the GB6 scores because if it matches Zen 5 16 core there, it shouldnt be possible that it loses to Zen 4 12 core in MT in Passmark.

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

Senior member

Attachments

Senior member

Golden Member

Lifer

Diamond Member

Senior member

Diamond Member

Senior member

Senior member

Senior member

Golden Member

Senior member

Senior member

Senior member

Platinum Member

Senior member

Senior member

Golden Member

Senior member

Diamond Member

Diamond Member

Diamond Member

Golden Member

Golden Member

Senior member

Senior member