Intel's 18A Era: Panther Lake, Clearwater Forest, and the Return of Foundry Leadership

18A represents Intel’s most critical manufacturing leap, combining new transistor technology and packaging breakthroughs for next-generation products.

Hardware by RereRara on  Nov 28, 2025

Intel is once again in the news, but not for the reasons you might think of when you think of a chip company. Making things has become a strategic issue, and politicians are paying more attention to it.

Because there is so much going on around Intel, the attention often moves away from the hardware itself. These new developments—Panther Lake, Clearwater Forest, and the 18A process node—mark a turning point.

Intel's, 18A Era, Panther Lake, Clearwater Forest, the Return of Foundry Leadership, NoobFeed

The big bet that Pat Gelsinger made on the whole company was on 18A. Its success depends on how competitive US-made chips are. Knowing about these platforms shows where Intel stands in a rapidly changing business.

Previewing 18A and Intel's New Wave of Silicon

We were recently in the US for Intel's Tech Tour. There, we got to see Panther Lake and Clearwater Forest, and we went inside FAB 52, where 18A is made. 18A is a big step forward for Intel because it is the first production-ready node to combine Gate-All-Around transistors and Backside Power Delivery Networks. On paper, it looks like the world's most advanced high-volume process node, and it was built completely in the US.

Panther Lake and Clearwater Forest both depend heavily on 18A. Figuring out how they are put together shows how Intel is trying to get back to the top. We need to look at how these tools work and what they tell us about the company's direction to make sense of it all.

Panther Lake vs Lunar Lake

Panther Lake might look like a simple follow-up to Lunar Lake at first glance. Lunar Lake is another mobile SoC designed for thin-and-light devices. But two big changes make everything different: Panther Lake adds more than one type of silicon—not just different configurations of the same die—and Intel is making core silicon again at Intel Foundry. Many people saw that Lunar Lake and Arrow Lake were mostly developed at TSMC as a sign that Intel Foundry was becoming less important. Panther Lake changes that course.

Panther Lake comes in three forms, each made with a different silicon mix optimized for a different use. This change shows a lot about Intel's current state and how its design mindset is changing.

Disaggregated Tile Architecture Takes Shape

Intel has been using a disaggregated tile approach since Meteor Lake. This means chips are broken into modular tiles that can be mixed to create custom products. Instead of just saving money like traditional chiplet approaches do, the goal is to be flexible, reuse, scale, and build in ways that fit your needs.

Six active tiles make up Panther Lake's toolbox. These are small and large compute tiles, small and large GPU tiles, and small and large platform driver tiles. These tiles are put together in different ways for each type of Panther Lake.

Panther Lake-U: The Entry Variant

Panther Lake-U takes the three smallest tiles and combines them to form the smallest possible shape. With four Cougar Cove P-cores, 12 MB L3 cache, and four Darkmont low-power E-cores, the small compute tile is both quick and powerful. The small platform controller tile supports PCIe Gen 4/Gen 5, Thunderbolt 4, USB, and Wi-Fi 7. The small GPU tile has four Xe3 "Celestial" GPU cores. It has all the key features a modern mobile SoC should have while still meeting entry-level requirements.

Panther Lake-H: Designed to work best with systems that have discrete GPUs

Panther Lake-H has a mix of Intel's big compute tile, its small GPU tile, and its big platform driver tile. The compute tile keeps the four Cougar Cove P-cores but adds eight Darkmont E-cores right on top of the P-core fabric. This makes multi-threading performance much better. The low-power island that is connected still has four more E-cores, making the setup a total of 4+8+4.

The small GPU tile is best for this version because it's made for computers with separate GPUs. The big platform tile adds more PCIe 5.0 lanes, enough to connect a separate mobile GPU and a Gen5 SSD simultaneously.

Panther Lake-H12Xe: Designed for iGPU-Only Systems

Panther Lake-H12Xe pairs the large compute tile with the large GPU tile. It has 12 Xe3 GPU cores, which is three times as many as the small GPU tile and far more than Lunar Lake. The small platform controller tile is used because it's intended for computers without a separate GPU. This configuration works with this tile. The CPU and GPU performance in this version is much better than in past generations.

The first fully realized disaggregated generation from Intel

Meteor Lake was the prototype, but it struggled because it wasn't widely available and lacked a desktop version. Lunar Lake went in a different direction and was mostly one big rock, even though it was called a "tile." Arrow Lake used some tile IP again, but they didn't fully use the modular idea either.

Panther Lake is the first generation to showcase Intel's disaggregated design as it was meant to be—with real tile mixing, planned silicon reuse, and a range of specialized versions. Intel went back to the design theory it laid out many years ago with this.

Intel's, 18A Era, Panther Lake, Clearwater Forest, the Return of Foundry Leadership, NoobFeed

How Intel and TSMC Plan to Use Process Nodes

Panther Lake tiles can all be made on different nodes, even ones from different foundries. Active silicon for Lunar Lake and Arrow Lake came only from TSMC. Intel made its own interposer and packaged the chips itself.

Panther Lake is not the same. Each of its small and big compute tiles is based on Intel's 18A. In other words, each Panther Lake chip has 18A silicon on it.

The platform controller tiles all use TSMC N6, but the GPU tiles are built on different chips. The small GPU tiles are made on Intel 3, while the large GPU tiles are made on TSMC N3E. Both Panther Lake-U and Panther Lake-H have an 18A CPU, an Intel 3 GPU, and a TSMC N6 platform. Panther Lake-H12Xe is the only one that uses all exterior tiles, except compute.

Even so, the large 18A compute tile covers more space than the big GPU tile and the platform controller tile combined. This means that Intel Foundry still makes most of the active silicon.

Panther Lake shows that Intel's foundry approach is stronger and more balanced, and it gives people more faith in Intel's manufacturing.

Clearwater Forest: A New Era for Server E-Cores

Some might say that Clearwater Forest is even more amazing. It is Intel's next-generation E-core server CPU, composed of 17 tiles manufactured across three process nodes and linked together using two Intel packaging technologies.

The 12 compute tiles each have 24 cores, comprising six Darkmont quad-core groups. That's 288 Darkmont cores spread out over 12 tiles. On 18A, where economy and density are most important, these tiles are made. But the base tiles might be even better.

Base Tiles with Massive Cache and Active Silicon

Clearwater Forest has three live base tiles that are built on Intel 3. Each of these tiles has memory controllers, DDR5 PHYs, and an amazing 192MB of last-level cache. That's 576 MB of cache spread across all three base tiles.

Since E-cores don't have L3 cache, this serves as a large last-level cache that all compute tiles can use. Intel's Foveros Direct 3D hybrid bonding is used to attach the compute tiles to the base tiles. This is the same type of technology that AMD's 3D V-Cache uses, but it is built into Intel's packaging environment.

Each base tile is hybrid-bonded to four compute tiles that are stacked on top of each other. This makes for a very dense and efficient pattern.

EMIB bridges link the two I/O tiles on either side of the base tiles. These tiles were used on earlier Xeon 6 chips. Clearwater Forest has more than twice as many cores as Sierra Forest (from 144 to 288) and a higher IPC because Darkmont replaces Crestmont, with better process nodes and huge new cache structures.

Clearwater Forest shows Intel's most advanced multi-tile integration to date and marks the start of a new era of competition in the server market.

18A: Intel's Make-or-Break Node

The 18A is important for both Panther Lake and Clearwater Forest. When we went to FAB 52, it was clear that Intel wants everyone to know how important this node is. It has Gate-All-Around transistors and Backside Power Delivery, features that no other high-volume rival has yet.

Intel seems to be optimistic about 18A's progress. Production has begun, and the landings at Panther Lake and Clearwater Forest should line up with high-volume production. Yields are not made public, but Intel probably wouldn't put multiple big goods on 18A if performance or yields fell far short of expectations.

There will have to be direct comparisons, especially since Lunar Lake's N3B design is identical to Panther Lake-U's compute configuration. Intel's desire to offer this kind of match-up shows that it is confident in its power, area efficiency, and ability to compete.

Intel's, 18A Era, Panther Lake, Clearwater Forest, the Return of Foundry Leadership, NoobFeed

Where Intel Stands Today

Intel is in a much better place now than it was a year ago, given the path it's taken. Panther Lake and Clearwater Forest are the start of true disaggregation, the reuse of tiles for a purpose, better designs, and Intel Foundry's return. Intel could once again be the winner in the data center, thanks to Clearwater Forest in particular.

18A seems to be on the right track, and what it teaches will have a direct effect on 14A. Chips made in the US could gain a lot of steam if they could find the right customers.

Intel is doing things right again, and the next generation of chips shows that the company is not just trying to stay alive, but to become a winner.


Also, check our other Intel chips articles:

Tanisha Aria

Contributor, NoobFeed

Latest Articles

No Data.