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

[Tigerick]

Wildcat Lake (WCL) Specs

Intel Wildcat Lake (WCL) is upcoming mobile SoC replacing Raptor Lake-U. 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 Q1 2026.

		Intel Raptor Lake U	Intel Wildcat Lake 15W?	Intel Lunar Lake		Intel Panther Lake 4+0+4
Launch Date		Q1-2024	Q2-2026	Q3-2024		Q1-2026
Model		Intel 150U	Intel Core 7	Core Ultra 7 268V		Core Ultra 7 365
Dies		2	2	2		3
Node		Intel 7 + ?	Intel 18-A + TSMC N6	TSMC N3B + N6		Intel 18-A + Intel 3 + TSMC N6
CPU		2 P-core + 8 E-cores	2 P-core + 4 LP E-cores	4 P-core + 4 LP E-cores		4 P-core + 4 LP E-cores
Threads		12	6	8		8
Max Clock		5.4 GHz	?	5 GHz		4.8 GHz
L3 Cache		12 MB		12 MB		12 MB
TDP		15 - 55 W	15 W ?	17 - 37 W		25 - 55 W
Memory		128-bit LPDDR5-5200	64-bit LPDDR5	128-bit LPDDR5x-8533		128-bit LPDDR5x-7467
Size		96 GB		32 GB		128 GB
Bandwidth				136 GB/s
GPU		Intel Graphics	Intel Graphics	Arc 140V		Intel Graphics
RT		No	No	YES		YES
EU / Xe		96 EU	2 Xe	8 Xe		4 Xe
Max Clock		1.3 GHz	?	2 GHz		2.5 GHz
NPU		GNA 3.0	18 TOPS	48 TOPS		49 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.

 

[DavidC1]

Pantherlake looking like a newborn kitten next to M5.

It's 80% faster per clock than Lion Cove/Zen 5 in Cinebench R24 ST and 50% faster per clock in Spec. These kind of things remind me over and over again how whether AMD or Intel cores are better is made moot when you get yourself out of the Wintel RDF bubble. Both can do MUCH, MUCH better.

GPU looks 20-40% faster than Pantherlake, assuming the presentation slides are correct. Though GPUs are harder to differentiate because they are more at mercy to process.

 

[poke01]

GPU looks 20-40% faster than Pantherlake, assuming the presentation slides are correct. Though GPUs are harder to differentiate because they are more at mercy to process.

GPU is still better on Panther lake for gaming, although for productivity M5 GPU takes a huge lead

 

[Hulk]

Pantherlake looking like a newborn kitten next to M5.

It's 80% faster per clock than Lion Cove/Zen 5 in Cinebench R24 ST and 50% faster per clock in Spec. These kind of things remind me over and over again how whether AMD or Intel cores are better is made moot when you get yourself out of the Wintel RDF bubble. Both can do MUCH, MUCH better.

GPU looks 20-40% faster than Pantherlake, assuming the presentation slides are correct. Though GPUs are harder to differentiate because they are more at mercy to process.

The M5 is a wonder. Too bad it can't do the one thing I need it to do. Run Windows.

 

[poke01]

The M5 is a wonder. Too bad it can't do the one thing I need it to do. Run Windows.

This is biggest reason why Intel and AMD barely do anything in client. Windows dominates

 

[adroc_thurston]

This is biggest reason why Intel and AMD barely do anything in client.

Intel sure did.
LNL is pgood.
AMD doesn't care about stuff not server socket perf CAGR.

 

[511]

Pantherlake looking like a newborn kitten next to M5.

It's 80% faster per clock than Lion Cove/Zen 5 in Cinebench R24 ST and 50% faster per clock in Spec. These kind of things remind me over and over again how whether AMD or Intel cores are better is made moot when you get yourself out of the Wintel RDF bubble. Both can do MUCH, MUCH better.

GPU looks 20-40% faster than Pantherlake, assuming the presentation slides are correct. Though GPUs are harder to differentiate because they are more at mercy to process.

I am more curious about the instructions mix of CBR24 cause Snapdragon X2 is 162 on reference device kind of expected but M4/M5 is league ahead did they do something on it for cinebench?

 

[adroc_thurston]

I am more curious about the instructions mix of CBR24 cause Snapdragon X2 is 162 on reference device kind of expected but M4/M5 is league ahead did they do something on it for cinebench?

yeah, that thing's off, underperforms heavily on Zen5.

 

[poke01]

I am more curious about the instructions mix of CBR24 cause Snapdragon X2 is 162 on reference device kind of expected but M4/M5 is league ahead did they do something on it for cinebench?

Here’s your answer
“Cinebench 2024 uses scalar and 128-bit packed floating point operations. Wider vector execution units are not useful. Scalar integer performance plays an important role in keeping the FP execution units fed.”

 

[poke01]

yeah, that thing's off, underperforms heavily on Zen5.

CB2024 is fine, its nT results reflects in Blender too


Ever since M4 Apple just went crazy with FP.

 

[511]

View attachment 132617

Here’s your answer
“Cinebench 2024 uses scalar and 128-bit packed floating point operations. Wider vector execution units are not useful. Scalar integer performance plays an important role in keeping the FP execution units fed.”

it is for x86 not ARM also Glymer is close to Apple M5 in ST but in Cinebench it's like 20% difference.

 

[poke01]

it is for x86 not ARM also Glymer is close to Apple M5 in ST but in Cinebench it's like 20% difference.

It’s the same for ARM lol. It’s scalar integer based.

Well like I said before Apples cores are very good at FP, which is what rendering applications like. They are also much wider cores than Qualcomms.





In R23 ST, M5 is 5% faster than the 285K at 5.7GHz with 8400MT/s CUDIMMs where it runs under x86 emulation. Remove the emulation and it’s not suprising to see the M5 doing so well in native apps.

 

[511]

It’s the same for ARM lol. It’s scalar integer based.

Big doubt especially on windows ARM vs MacOS ARM

Well like I said before Apples cores are very good at FP, which is what rendering applications like. They are also much wider cores than Qualcomms.

Glymer is similar to apple cores in ST but the Cinebench score Delta is way wide

View attachment 132624

In R23 ST, M5 is 5% faster than the 285K at 5.7GHz with 8400MT/s CUDIMMs where it runs under x86 emulation. Remove the emulation and it’s not suprising to see the M5 doing so well in native apps.

Yeah R23 was never optimized for ARM

 

[Magio]

In R23 ST, M5 is 5% faster than the 285K at 5.7GHz with 8400MT/s CUDIMMs where it runs under x86 emulation. Remove the emulation and it’s not suprising to see the M5 doing so well in native apps.

Yeah R23 was never optimized for ARM

We can discuss whether it was fully optimized (which is really why crossplatform benchmarking never makes sense), but R23 has had a native ARM/Apple Silicon build since the M1 days, it's not running in "emulation" or through Rosetta so the original claim is bogus.

 

[poke01]

Big doubt especially on windows ARM vs MacOS ARM

Its the same on both ARM platforms... They both use Intel Embree and use scalar.

Glymer is similar to apple cores in ST

Not in rendering where Apple has a sizable lead over Qualcomm, you can see this in GB6 too. Both cores are ~4.6GHz

Mac17,2 vs vivo V2505A - Geekbench

browser.geekbench.com

 

[poke01]

We can discuss whether it was fully optimized (which is really why crossplatform benchmarking never makes sense), but R23 has had a native ARM/Apple Silicon build since the M1 days, it's not running in "emulation" or through Rosetta so the original claim is bogus.

R23 is based on v3.11 Intel embree which Maxon just ported over to ARM with no optimisation. Support for Apple cpus wasn't even added till v3.13 and then v3.13.4 added more support. R23 is "native" but the backend is still very much x86 based.

Release Embree v3.13.0 Release · RenderKit/embree

Added support for Apple M1 CPUs. RTC_SUBDIVISION_MODE_NO_BOUNDARY now works properly for non-manifold edges. CMake target 'uninstall' is not defined if it already exists. Embree no longer reads the...

github.com

Release Embree v3.13.4 Release · RenderKit/embree

Using 8-wide BVH and double pumped NEON instructions on Apple M1 gives 8% performance boost. Fixed binning related crash in SAH BVH builder. Added EMBREE_TBB_COMPONENT cmake option to define the co...

github.com

This is why you see a much bigger uplift with CB2024 which uses a newer version of Intel Embree.

 

[Nothingness]

R23 is based on v3.11 Intel embree which Maxon just ported over to ARM with no optimisation. Support for Apple cpus wasn't even added till v3.13 and then v3.13.4 added more support. R23 is "native" but the backend is still very much x86 based.

That still doesn't prove it was emulated in the Rosetta sense. It perhaps just defaulted to using no vector instruction at all due to poor compilation flags (and/or lack of ISPC ARM as hinted in v3.13.1 release). If I wasn't lazy, I'd look into it, but I am. Anyway we all agree that CB23 is a poor choice to make x86 vs Arm comparisons, don't we?

Note that Embree seems to still use some intrinsics translation layer: https://github.com/RenderKit/embree/releases/tag/v4.3.1

Known issue: Embree build using Apple Clang 15 and ARM support (via the SEE2NEON library) may cause "EXEC_BAD_INSTRUCTION" runtime exceptions. Please use Apple Clang <= 14 on macOS.

Given last Intel announcements about open source and not giving an edge to competition, I don't expect this to improve.

 

[511]

Apple is a Trillion$ Company sure they could afford few big fixes in EMBREE also EMBREE has Better kernels for AVX2/AVX-512 the biggest it's complied using ISPC.

 

[Hulk]

Apple is a Trillion$ Company sure they could afford few big fixes in EMBREE also EMBREE has Better kernels for AVX2/AVX-512 the biggest it's complied using ISPC.

Why would Apple invest 10 cents to support Windows?

Everything about the Apple eco-system is closed. They have complete control and no competition. They can charge just about whatever they want for hardware and software. No modifications are allowed. There is no DIY market. The software is also tightly controlled, which allows for software and hardware to be optimized more easily since there are no arguments among CPU developers or software developers as to how things should be. This is how Apple has been able to do the impossible many times over the past 50 years. By that I mean moving from Motorola to PowerPC to x86 to M Series CPU's. They have switched the entire code bas 3 times!

Everything that is great about x86 and Windows, namely the huge open "source" market for millions of applications going back 50+ years and the ability to custom configure and build/upgrade your own system at great prices, have a downside in that there are compatibility and optimization issues on a far larger scale than an "East Berlin's" Apple culture.

Now having lived through building computers from the 80's to date, I can tell you that Windows 11, as much as it is maligned is a wonder compared to Window 1, 2, or even the groundbreaking 3.0. Not to mention DOS, which I used at my first engineering job in the late '80's.

While it is natural to compare Apple CPU's to x86, it's kind of silly. It's like comparing an F1 car to one that is legal to be driven daily. One is purpose built to run in a tiny loop on perfect pavement, the other goes slower, but basically anywhere.

I NEED to build the PC to my exact specifications. What I need exactly isn't manufactured. I need it to fit under my desk, I need a certain form factor, I need it to be quiet, and need custom fan profiles, I need specific SSD and memory configurations, etc... I guess you could say "desire" instead of need but you get the point.

If the performance gap between Apple and x86 was so great that I simply couldn't work then I'd switch. But in reality, while we are always looking for more compute, my 9950X handles everything I throw at it with aplomb. 50 plugins on an audio mix I'm working on? No problem. But as I mentioned in an earlier post, it's all a nonstarter as I need Windows. I think in directories and roots, not folders and desktops.

Finally, honestly I wonder without the iPod and iPhone if we'd even be talking about Apple today? My kids laugh at my Pixel 7, which cost 1/3 of their iPhones. I ask them what my Pixel can't do they their phones can. They can't tell me anything that has value other than "easier to use facetime" or some other nonsense.

Apple is great, "from a certain point of view" as a wise man once said.

Different stroke for different folks.

 

[MS_AT]

EMBREE has Better kernels for AVX2/AVX-512 the biggest it's complied using ISPC.

Which CB24 does not use

 

[511]

Which CB24 does not use

sadly it's still stuck on AVX for majority of it's instructions sadly not even AVX 2

 

[511]

Why would Apple invest 10 cents to support Windows?

Not windows but embree it's an OSS library and it is used in Blender as well.

 

[Nothingness]

Everything about the Apple eco-system is closed. They have complete control and no competition. They can charge just about whatever they want for hardware and software.

I agree HW is closed and prices are outrageous, but on the SW front I disagree. I happily run thousands of apps on macOS over which Apple has no control. Basically the same that Linux can run. Something I never was able to achieve with Windows (which I stopped using before WSL was introduced).

Everything that is great about x86 and Windows, namely the huge open "source" market for millions of applications going back 50+ years and the ability to custom configure and build/upgrade your own system at great prices, have a downside in that there are compatibility and optimization issues on a far larger scale than an "East Berlin's" Apple culture.

Many old applications don't run anymore on modern Windows. I had to use full machine emulation in some circumstances to get old games to run. And that also means that for these apps I could run them on an Apple or a Linux machine if I had to.

Your experience is so far from mine. I never could switch from UNIX/Linux to Windows, despite trying several times over the last 30 years. But I easily switched from Linux to macOS. As you wrote, different stroke for different people, but some of your claims are just plain wrong.

 

[Magio]

R23 is based on v3.11 Intel embree which Maxon just ported over to ARM with no optimisation. Support for Apple cpus wasn't even added till v3.13 and then v3.13.4 added more support. R23 is "native" but the backend is still very much x86 based.

That's still not what emulation is. As I said we can discuss optimization but it was still native code.

And it's such a crapshoot to estimate which OS/ISA a benchmark is most optimized and how much of that is representative of library optimization in real world use and which is just the benchmark's own failings. Just need to look at Linux consistently outperforming Windows (pretty sure it also beat macOS when they were still on Intel) on the same hardware in Geekbench and as big a Linux fan as I am there's no evidence it translates 1-to-1 to real world use.

The only conclusion from a debate around a benchmark's cross platform optimization should be that cross platform benchmarks are just never really reliable.

 

[511]

SPEC is pretty reliable if done properly

 

[Doug S]

Finally, honestly I wonder without the iPod and iPhone if we'd even be talking about Apple today?

Without the iPod to generate cash flow Apple never could have made the investments required to develop the iPhone. Without the iPhone driving massive economies of scale for developing their own silicon, ARM Macs wouldn't exist so yes we wouldn't be discussing those.

Without IBM making their ill-fated deal with Bill Gates you wouldn't be talking about Windows because Microsoft would be a footnote in computing history like Broderbund. You can play the "what if" game any number of ways. In one maybe Acorn ends up as the dominant PC and Intel dropped out of CPU manufacturing to focus exclusively on DRAM.