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.

 

[poke01]

It is a complete waste of time trying to compare Apple CPU's to X86 processors. Apple stuff is geared towards power efficiency and multimedia performance. They kick ass in arcade style games that you play on smart phones. Not AAA graphics intensive multiplayer games. Apple is great for Netflix and battery life is beyond 10 hours. Apple processors should be compared to Snapdragon processors and the new MediaTek Dimensity 9400 CPU.

No offence but you have no idea what you’re talking about cause we are not talking about the phone CPUs in iPhones but comparing how ST in Cinebench2024 compares between Intel, Apple and AMD CPUs that run desktop operating systems. No idea why Mediatek is even in your reply let alone battery life.

 

[DavidC1]

https://quasarzone.com/bbs/qc_bench/views/91773

Haha, the core-to-core latency within Skymont clusters are super low, about half the predecessor.

Those L1-L1 transfers are doing good.

 

[poke01]

@DavidC1 here is R23

 

[gdansk]

@DavidC1 here is R23

I had screenshotted without the cartoon boy for the anime-averse people here but I posted it in Zen 5 thread by accident

 

[511]

@DavidC1 here is R23View attachment 110258

Very nice as expected
Just Watched Derbauers video should have not included DLVR and should have let it be a MT Beast

 

[DavidC1]

@poke01 better, but still only two datapoints.

TPU's Gracemont cores are getting 1089 points @ 3.9GHz, so the 32% Integer gains are carrying out quite nicely here.

At 90% scaling and at peak 4.6GHz clocks, it's roughly 5950X's ST performance.

 

[poke01]

At 90% scaling and at peak 4.6GHz clocks, it's roughly 5950X's ST performance

I wonder what the power consumption is?

 

[DavidC1]

I wonder what the power consumption is?

In Computerbase.de's test and CB2024,

8P 14900K - 65-67W for 719 points
8P 14900K HT - 80W for 936 points
1P HT+16E 14900K - 90W for 1100 points
8P 285K - 65-67W for 769 points
1P+16E 285K -100W for 1444 points

I would say E cores are pretty darn efficient at 4GHz.

 

[Hulk]

Computerbase has that tested on CB2024.
View attachment 110244
If you divide 285K 8P, it's 96, but scaling isn't perfect so let's say it's 100. 1P+16E = 1444, so without the P, it's 1340.

1340 divide by two is 670, but scaling isn't perfect. Cinebench is about 90% scaling but let's assume few different points:
-95%: 687
-90%: 705

So roughly Golden Cove levels. By the way, in Cinebench Lion Cove is only 7% faster than Golden Cove. It's not a perfect way of comparison but gives us a rough idea.

Skymont comparison shows 31%, but it's a bit skewed as the 1P has HT disabled on 285K but enabled on 14900K. So, E-to-E it's higher than 31%. If you assume 117 points for the 14900K's HT enabled P core, then it's 1340 vs 983, or 36%.

Excellent find! "Throughput" is CB R24 MT points/GHz.

Skymont 33.5% better than Gracemont and 6.6% behind Raptor Cove.

Lion Cove barely 5% better than Raptor Cove in this one.

Skymont carrying 60% of load vs. Gracemont carrying 42%.

Skymont is all that. Lion Cove seems to be underperforming.

Core	Throughput
Zen 5 w/HT	31.3	6.8%	100.0%
Raptor w/HT	29.3	22.1%	93.6%
Lion Cove	24	1.7%	100.0%
Zen 5 w/o HT	23.6	4.9%	98.3%
Raptor w/o HT	22.5	6.6%	93.8%
Skymont	21.1	33.5%	87.9%
Gracemont	15.8		65.8%

 

[dttprofessor]

In Computerbase.de's test and CB2024,

8P 14900K - 65-67W for 719 points
8P 14900K HT - 80W for 936 points
1P HT+16E 14900K - 90W for 1100 points
8P 285K - 65-67W for 769 points
1P+16E 285K -100W for 1444 points

I would say E cores are pretty darn efficient at 4GHz.

Skymont on LNL is setted on 3.7Ghz, maybe this is the best balance point.

 

[DavidC1]

In Computerbase.de's test and CB2024,

8P 14900K - 65-67W for 719 points
8P 14900K HT - 80W for 936 points
1P HT+16E 14900K - 90W for 1100 points
8P 285K - 65-67W for 769 points
1P+16E 285K -100W for 1444 points

I would say E cores are pretty darn efficient at 4GHz.

Perf per watt in CB2024:
8P 14900K - 11.06 per W, Base(100%)
8P 14900K HT - 11.7 per W(105.8%)
1P HT+16E 14900K - 12.2 per W(110.3%)
8P 285K - 11.83 per W(106.9%)
1P+16E 285K -14.4 per W(130.2%)

HT improves perf/W by 6%, and makes it close to E core efficiency. However, Lion Cove improves efficiency by 7%, slightly better than HT enabled Raptorlake. Skymont improves on Gracemont by 18%.

 

[desrever]

Perf per watt in CB2024:
8P 14900K - 11.06 per W, Base(100%)
8P 14900K HT - 11.7 per W(105.8%)
1P HT+16E 14900K - 12.2 per W(110.3%)
8P 285K - 11.83 per W(106.9%)
1P+16E 285K -14.4 per W(130.2%)

HT improves perf/W by 6%, and makes it close to E core efficiency. However, Lion Cove improves efficiency by 7%, slightly better than HT enabled Raptorlake. Skymont improves on Gracemont by 18%.

All the hype of Skymont and it's only 18% more efficient than Gracemont. N3B vs Intel 7 and all they got was 18% improvement?

 

[lopri]

If your main app is Cinebench, your choice is the Arrow Lake. If your main app is Geekbench, Get Zen 5.

 

[gdansk]

All the hype of Skymont and it's only 18% more efficient than Gracemont. N3B vs Intel 7 and all they got was 18% improvement?

It's smaller, it's somewhat efficient, it seems pretty good and the monts have actually been improving gen-on-gen. 🤔
Maybe some people overhype it. I'm not sure if it's any better performance per area than N3 Zen 5C.

 

[gdansk]

If your main app is Cinebench, your choice is the Arrow Lake. If your main app is Geekbench, Get Zen 5.

Here's my flow as of today:
Main App = Blender -> It doesn't matter, about the same
Main App = Chrome/Electron -> Zen 5
Main App = Games -> Zen 4 3D (or 14900K if you like to live dangerously)
Main App = Python maybe? -> Arrow Lake S

 

[MangoX]

If your main app is Cinebench, your choice is the Arrow Lake. If your main app is Geekbench, Get Zen 5.

What if neither? Those aren't my main apps I use daily.

 

[TESKATLIPOKA]

If I were leading Intel, I would

1) re-hire Jim Keller as chief technology consultant and fire ANYONE on the spot who disagrees with him

2) shut off all fabs other than the extremely necessary ones

3) acquire the Oryon team from Qualcomm for $$ billions (give them an offer they won't refuse) and start making both x86-64 and WARM CPUs

4) create the top three flagship CPUs with dissimilar architecture cores (probably souped up Darkmont for P cores and Oryon cores for E-cores) and bribe Microsoft to include both architecture executables in Windows so that demanding ones always run on P-cores whereas services type of executables run on Oryon cores

2, 3 and 4 subject to change based on Keller's advice.

We are lucky you are not leading Intel, because you would truly finish It off.

 

[Doug S]

It is a complete waste of time trying to compare Apple CPU's to X86 processors. Apple stuff is geared towards power efficiency and multimedia performance. They kick ass in arcade style games that you play on smart phones. Not AAA graphics intensive multiplayer games. Apple is great for Netflix and battery life is beyond 10 hours. Apple processors should be compared to Snapdragon processors and the new MediaTek Dimensity 9400 CPU.

Ah yes, the old "just ignore that Apple beats x86 in ST using a fraction of the power, Macs can't do graphics intensive games nearly as well as PCs so those aren't 'real' CPUs".

If it were possible to run Apple's CPU with a top of the line Nvidia GPU, it would likely kick x86 butt in gaming as well. The fact Apple is weak in AAA gaming is down to two reasons. 1) their GPU isn't close to the equal of their CPU, and there are no options for third party GPUs; 2) their GPU's pipeline operates differently (TBDR and Metal) so even games that are ported from the PC are generally poorly ported and don't perform nearly as well as they could.

Not that I blame game devs for taking shortcuts when porting to the Mac - the game market on the Mac is tiny compared to PC gaming so it isn't worth putting a lot of resources toward making the best possible port.

 

[511]

Arrow Lake MemSS Overclocking - SkatterBencher

We have a closer look at tuning the performance of the the Arrow Lake MemSS, or Memory Sub System, located on the SoC Tile.

skatterbencher.com

Skaterbench memory oc

 

[511]

I think ARL Aapproach is a bit wrong here it's desktop No one is crying over 10-20W more just give us the damm performance in ST/MT it is a hit or miss in ST while being Okay in MT if you wanted to damm gaming performance fine but should have kept Application Performance

 

[yuri69]

Can someone expand on which factors may contribute the 9950X and 285K having such similar IPC in 1T when it was shown Zen 5 operates as a 4-wide decode design outside of SMT and Lion Cove is apparently up to 8-wide decode?

Too many mispredictions? There isn't more ILP to exploit? Lion Cove is bottlenecked at execution? Then why does Lion Cove have 8 wide decode?

Every design is full of tradeoffs. Intel has lagged behind AMD's uOp cache effectiveness. Intel might have aimed at a workload (or part of it) which does not play well with uOp cache. Are those decoders fully-fledged or are those some quirks? We don't know.

What we know is that Lion Cove has a bit odd branch prediction accuracy. Also the very exec-wide Zen 5 is bottlenecked mainly by the frontend latency. We need to profile Lion for that too.

It just so happens you're 2 months away from that.

Wow, over 40% IPC!!!

 

[DAPUNISHER]

For the moment, it's DOA for gamers. There has to be significant gains waiting on microcode and OS updates, because something is as broken as ARC at the moment. I dislike using geomean, here's why -



20 games with double digit losses, 15 where it's basically 20-50%. That's- "Yes police, I'd like to report a murder" territory. If she'd added F1 '24 that's another 20%+ beatdown. This can't be all there is without overclocking and budget breaking stupidly expensive ram, it can't be...

 

[reaperrr3]

I would expect 9950X on N3B, to yield the same power efficiency as it has now .... which is already better than Arrow Lake and be able to do so with about a 10-15% clock increase.

Why do you think differently?

TSMC themselves gave these approximate figures for their respective processes:

N3(B): +15% perf (aka clocks at iso power/transistor count) or -30% power vs. N5
N4P: +11% perf or -22% power vs. N5

So the only area where N3B is a notable uplift over N4P is transistor density.
N3E is slightly better than N3B in terms of perf/efficiency (+18% or -32% vs. N5), but loses some density.

Additionally, the technical differences between N5 class and N3 class processes could be significant enough that a quick-n-dirty shrink doesn't necessarily give you the maximum clockspeed benefit "just like that".
In any case, a 3nm Zen5 would've likely used N3E.
But 10-15% higher clocks would mean 6.3-6.6 GHz, and I highly doubt that Zen5 would've hit that so easily.

Consumer-Zen6 with the FPU scaled back to 256bit might hit that with N3P, if we're lucky.

 

[alcoholbob]

For the moment, it's DOA for gamers. There has to be significant gains waiting on microcode and OS updates, because something is as broken as ARC at the moment. I dislike using geomean, here's why -

View attachment 110261

20 games with double digit losses, 15 where it's basically 20-50%. That's- "Yes police, I'd like to report a murder" territory. If she'd added F1 '24 that's another 20%+ beatdown. This can't be all there is without overclocking and budget breaking stupidly expensive ram, it can't be...

Still there are going to be people who watched de8bauer's video and will go out and drop several grand on a new build just to see if they can tweak this thing to the high heavens just to match last gen performance lmao.

 

[TESKATLIPOKA]

Intel makes way more money just on desktops compared to AMD's entire client business.

How does It help them exactly, when their foundries will no longer make any client CPUs for them? Not sure how long they plan to continue producing Raptor Lake.