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.





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.)

 

[gdansk]

ST performance is the best currently in x86 world

I didn't see that. Even in browser benches it's generally behind Zen 5.

 

[GTracing]

Hopefully someone tests gaming with e-cores disabled. That should shed some light on whether the windows scheduler is the problem

 

[511]

I don't even know if ST Workload is getting scheduled correctly

 

[cannedlake240]

Some of the regressions are wild though. . .

View attachment 110206

View attachment 110207
View attachment 110208

Are these running on the E cores?

 

[Hitman928]

Are these running on the E cores?

No idea. Investigation needed.

 

[cannedlake240]

Hopefully someone tests gaming with e-cores disabled. That should shed some light on whether the windows scheduler is the problem

E cores should be tested too find if chadmont lives up to the hype

 

[AcrosTinus]

I'm not so sure it takes the ST crown and MT is very hit and miss.

Sure, but based on what I have seen in Phoronix as well, I will make the claim that ST performance is the best currently and MT is competitive.

 

[cebri1]

It’s an absolute mess of a launch. Results are over the place. It’s an unfinished product with alpha software support.

 

[AcrosTinus]

I didn't see that. Even in browser benches it's generally behind Zen 5.

Even behind the Raptor which is confusing as well, might be related to the higher latency.
I'm waiting for the OC people to unleash it.

 

[MS_AT]

Intel Core Ultra 9 285K "Arrow Lake" Delivers Strong Linux Performance Review - Phoronix

www.phoronix.com

Look good despite lack of AVX512

I hoped the code compilation would fare better with 24 physical cores where SIMD does provide little advantage if any at all.

Check here and the following posts :

Page 449 - Discussion - Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

Page 449 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Ok, I concede the code is written so as to use FMA optimized implementation for Intel cpus only. Thanks for linking the original message it was helpful to navigate around the code.

 

[adroc_thurston]

Even behind the Raptor which is confusing as well, might be related to the higher latency.

Slower L3, worse memory perf. Unfortunately an additional meg of L2, L1.5d and L1d latency improvement did not offset any of that.

Even in browser benches it's generally behind Zen 5.

Behind many things even. That's rough.
Also not that good at code compile etc either. Probably not SKT's best workload.

 

[igor_kavinski]

For stock config 285K in TechPowerUp review:

Spoiler: My opinion

For stock config 285K in TechPowerUp review:

Jetstream: Fail
Speedometer: Fail
GPT2 AI: Decisive win
Stable Diffusion: Decisive win
Topaz Photo AI: Fail
NLP AI: Decisive win
Image classification: Decisive win but massively outshined by 265K
RPCS3: Respectable score slightly behind 7950X and 14900K
Ryujinx: Fail
CB R24 and Blender: Decisive win
Corona: Respectable score slightly behind 9950X
Keyshot 2024: Respectable score and four seconds slower than 7950X
V-Ray: Beats 7950X but can't touch 9950X
UE5 Build Cook Release: Respectable score
Git: Fail
Visual Studio: Respectable score four seconds slower than 9950X
AV1 encoding: Decisive win
HEVC encoding: Just behind 14900K
H.264: Respectable score
MP3 encoding: Fail
MS Office: Fail
Adobe Premiere: Respectable score
Adobe After Effects: Decisive win
3DF Zephyr: Decisive win
COMSOL: Decisive win
NAMD: Decisive win
Genome analysis: Decisive win but outshined by 265K
Antivirus: Fail
Altium Designer: Almost a fail as it can't catch 14900K but otherwise doesn't look that bad compared to 9950X
Tesseract OCR: Respectable score less than 3 seconds slower than 9950X
VirtualBox: Respectable score
MySQL: Fail as it's behind 14900K quite a bit and can't catch 9950X
MongoDB 6: Decisive win
OpenJDK: Win
ASP.NET: Tie with 14900K
WinRAR: Respectable score
7-zip: Fail
AIDA64 AES: Fail
AIDA64 SHA3: Fail
Games fail everywhere except Spiderman RT: Win
Min FPS fail everywhere except Spiderman RT and Last of US: Good

 

[Hitman928]

Sure, but based on what I have seen in Phoronix as well, I will make the claim that ST performance is the best currently and MT is competitive.

Based on which benchmarks?

 

[gdansk]

I am not trying to be snarky but what is the advantage of Intel's advanced packaging if it performs similarly and with similar energy consumption to AMD's cheapo glue?

Is it more relevant at low TDPs like ARL-H for mobile?

 

[adroc_thurston]

Is it more relevant at low TDPs like ARL-H for mobile?

indeed.

I am not trying to be snarky but what is the advantage of Intel's advanced packaging if it performs similarly and with similar energy consumption to AMD's cheapo glue?

Lower pJ/b and overall lower power floor.

 

[cannedlake240]

I am not trying to be snarky but what is the advantage of Intel's advanced packaging if it performs similarly and with similar energy consumption to AMD's cheapo glue?

Is it more relevant at low TDPs like ARL-H for mobile?

It's not an advantage. It's a result of fab issues

 

[Hulk]

So, I'm done consuming some of the biggest reviews:

Here is the autopsy:

• Power consumption is indeed lowered
• ST performance is the best currently in x86 world
• MT performance is great and Skymont compensated the lack of HT
• CUDIMM advantage consumed by the latency of the Tile architecture
• If you don't game on a 4090 or game in 4K, these CPUs and any CPU will make no difference
• AMD wins in all cases when it comes to gaming
• Basically a nice entry level workstation chips that games well enough for GPUs up to 4070Ti Super or 4080.
• For anyone on Raptor or Alder, SKIP
• The foundation is not as bad as I though, It might need a new revision

I agree with this assessment. I'm curious as to why application performance isn't better, especially things like MS Office, which tend to rely on ST performance?

As we had thought, the combination of high Raptor clocks and tile latency did create a high bar for Intel to cross for ARL vs. Raptor. They just got over that bar, but it looks like the 9950X bar is just a little higher...

Also, Techpowerup is showing nT frequency as 5.4GHz, not 5.5GHz as listed in the Ark.

 

[Hulk]

I am not trying to be snarky but what is the advantage of Intel's advanced packaging if it performs similarly and with similar energy consumption to AMD's cheapo glue?

Is it more relevant at low TDPs like ARL-H for mobile?

Perhaps this is one of those design decisions where Intel is too clever by half.

 

[Timorous]

Perhaps this is one of those design decisions where Intel is too clever by half.

Funnily enough I think if they kept iterating on Clarkdale they would be miles ahead in terms of chiplets given that was basically 1 CCD and 1 IO die, it even had an iGPU unlike Zen2 and Zen3 IO dies.

 

[adroc_thurston]

It's not an advantage. It's a result of fab issues

no, freebie 2.5d slab is definitely an advantage.

 

[511]

C2C is not that bad but C2C -> IMC is

 

[511]

The supply issue is everyone lining at TSMC which Intel is not used to they need their fab back with comparable Technology

 

[adroc_thurston]

C2C is not that bad

Uh. Yeah. It's all a single ring!

The supply issue is everyone lining at TSMC which Intel is not used to they need their fab back with comparable Technology

TSMC has wafers to spare.

 

[511]

Only good slide out of all ARL Reviews

 

[SiliconFly]

.