Intel Arrow Lake-S CPU gets delidded ahead of launch, showcasing tiled architecture

[zohaibahd]

What just happened? Intel's new Arrow Lake desktop processors are set to hit shelves later this week, but one hardware enthusiast has already managed to get an early look under the hood. Streamer Madness727 got their hands on an upcoming Core Ultra Arrow Lake-S model and did the unthinkable – he delidded it.

The delidding process involves removing the integrated heat spreader (IHS) from the top of the CPU. It's a risky move that can damage the chip if not done correctly, but it allows for direct cooling of the silicon dies inside, which is something hardcore overclockers love for pushing performance to the limit. Of course, it also voids your warranty.

Madness727 shared multiple photos of the delidded Arrow Lake-S chip, giving us the first real-world look at the novel chiplet design Intel is using for this new generation. While the specific model remains unidentified, this isn't particularly important, as all Arrow Lake-S models share the same tile configuration.

At the center, you've got the big compute tile packing those 8 performance (Lion Cove) and 16 efficiency (Skymont) cores we've heard so much about. Flanking it is the graphics tile with 4 Xe GPU cores embedded. Then there are the obligatory I/O tile that integrates a Thunderbolt 4 controller, and SoC tile handling other chipset functions.

I just can't stop myself from delidding stuff pic.twitter.com/XmniBHY5U8

– Madness! (@Madness727) October 18, 2024

Interestingly, there's also a vacant "dummy" tile that seems to just be there for structural support.

For the uninitiated, a chiplet design with its tiled architecture allows Intel to mix-and-match different tiles using optimized process nodes for better performance-per-watt. It also makes it easier to swap in cutting-edge components each generation without overhauling the entire CPU design.

Also read: What Are Chiplets and Why They Are So Important for the Future of Processors

Case in point, Arrow Lake-S uses different process nodes for various components. The compute tile, which houses the cores, is built on TSMC's advanced N3B node, while the GPU tile uses N5P, and the SoC and I/O tiles are based on the more mature N6 process. Remarkably, all five tiles are assembled on a base layer fabricated using Intel's 22nm FinFET technology.

While the desktop Arrow Lake-S models utilize this 5-tile layout, Intel's mobile "Arrow Lake-H" chips headed for laptops in Q1 2025 will rock a more stripped-down configuration. These mobile CPUs will feature a smaller compute tile with a 6P+8E core complex, an expanded graphics tile with 8 Xe cores, and a reduced I/O tile.

Permalink to story:

Intel Arrow Lake-S CPU gets delidded ahead of launch, showcasing tiled architecture

 

[VitalyT]

Some beautiful images of the Core Ultra 200S

You must be a big fan of Kazimir Malevich's art, and his Black Square creation.

 

[Manoa]

I would liked this frankestein lake more if his performance was as high as his number of tiles

 

[trparky]

Why is it so rectangular?

 

[HardReset]

Why is it so rectangular?

There are lot more components on CPU that you can see from top picture. Probably those components are easier to fit when it's shaped that way and not square.

 

[trparky]

I’ve always contributed the fact that it’s rectangular to how the chips bend.

 

[Axeia]

It is impressive how they managed to use so many different processes, hopefully they can put this to good use to keep things scalable and have a good base to iterate on.
We kinda need Intel around as AM5 has shown that AMD is fine with being almost as complacent as Intel was for many years.

 

[NumberNine]

Nope. I do not want or need an NPU, no matter how much performance or efficiency they tag onto it. I won't pretend it's okay.

 

[zamroni111]

Nope. I do not want or need an NPU, no matter how much performance or efficiency they tag onto it. I won't pretend it's okay.

Wait 2 or 3 years for Microsoft to realize that 40 tops npu is ridiculous requirement

 

[techyrounder]

What an exciting preview of the Intel Arrow Lake-S CPU! Delidding before launch really gives us a fascinating look at its tiled architecture. It's impressive to see how Intel is pushing the boundaries with design innovations. I'm particularly curious about how this architecture will impact performance and efficiency compared to previous generations. It would be great to hear more about potential applications in gaming and content creation. Thanks for sharing this update!

 

[Rudyduu]

It is impressive how they managed to use so many different processes, hopefully they can put this to good use to keep things scalable and have a good base to iterate on.
We kinda need Intel around as AM5 has shown that AMD is fine with being almost as complacent as Intel was for many years.

Intel wasn't complacent, their foundry just couldn't keep up with their designs.

 

[HardReset]

Intel wasn't complacent, their foundry just couldn't keep up with their designs.

They were. Many experts warned that 10nm tech was too big simple improvement and it will cause trouble. Intel was complacent and we all know what resulted from that decision.

 

[ioliveirafarias]

Wait 2 or 3 years for Microsoft to realize that 40 tops npu is ridiculous requirement

MS will use your NPU on background to track your data. Its the evolution of User Metrics enabled by default today.
Thats why NPU will be a requirement for windows.

 

[trparky]

Thats why NPU will be a requirement for windows.

If the data processing is done on my system and nothing more than a few Boolean values are sent to Microsoft, then I don't care.

 

[ioliveirafarias]

If the data processing is done on my system and nothing more than a few Boolean values are sent to Microsoft, then I don't care.

Tecnically, youre buying an NPU for Microsoft spies you better. Even if you dont use it in another application

 

[Axeia]

Intel wasn't complacent, their foundry just couldn't keep up with their designs.

They barely changed their architecture for years, they gave us quadcore after quadcore. Whilst requiring motherboard upgrades for almost every tiny generational improvement.

Some weird AliExpress motherboards with the 'wrong' chipset for the socket showed that the new motherboard costs were something that weren't always needed to boot, they just liked the chipset sales I suppose.

Once they got competition suddenly they could do more than 4 cores and lately they're even overhauling the architecture.

The better node lead is what made them lazy, AMD being no threat until Ryzen made them complacent. Now that they have done financial struggles they're probably regretting having done so little for so many years. Competition and the need for innovation it brings benefits us all.

At the moment both Intel and AMD have been rather boring with the last few generations. AMDs generational improvements on AM5 have been small compared to AM4.
Intel needed to fix their structure and this seems like a step in the right direction. Hopefully the gen after it gives them a step as big as Ryzen 2000 was compared to 1000.

 

[HardReset]

Intel needed to fix their structure and this seems like a step in the right direction. Hopefully the gen after it gives them a step as big as Ryzen 2000 was compared to 1000.

That's what I hope too. Ryzen 2000 had no architectural improvements vs Ryzen 1000.