Intel Panther Lake, architecture based on Intel 18A
Although we do not have commercial details, such as number of models, frequencies or different configurations, during ITT 2025 in Arizona we were able to discover what Intel is preparing for its next generation of processors for laptops and low-power edge solutions.
Processors that, with a consumption range of around 17 W, They promise to evolve Lunar Lake with the performance of Arrow Lake-H thanks to architectural optimizations and a revolutionary new manufacturing process: the Intel 18A.

Intel Panther Lake offers a more scalable and versatile design with three unique combinations that enjoy the same design and format, with the same packaging, and a lot of freedom in energy profiles, so that its integration in different designs, with different configuration options and RAM, is much more versatile than in Lunar Lake, which you surely remember integrated DDR5 memory into the “die” of the processor.

Among the most interesting new features is a modular GPU, which has two different versions and scales performance with up to three times more features depending on the design. This generation also returns low-power efficiency cores and configurations of up to 16 cores.

This generation is made up of different tiles, with GPU on a tile connected using the new Intel Scalable Fabric 2 technology, which allows homogeneous connectivity with agnosticity both in the partition and in the connection of processing units. It is made up of a total of seven elements, including the packaging, a base tile and the different elements or systems. It is a SoC that redefines the concept of “system on chip” (SoC – system on chip) with a “System of Chips”. All of them connected using Intel’s Foveros technology, both for the system tiles and the base tile with the packaging.

This new generation once again extracts the memory system out of the “die” and now supports the incorporation of up to 96 GB of LPDDR5 memory with speeds of 9600 MT/s in dual channel or DDR5 memory, in different formats such as CAMM2 or SODIMM, with up to 128 GB capacity and speeds up to 7200 MT/s. We forget about the predefined configurations per SKU of up to 32 GB.

We can find this generation in three basic configurations; Then we will have to see if there are different versions based on base and turbo frequencies, we still do not have details about commercial versions. These three base configurations share the same design and packaging, so it’s very easy for integrators to offer different capabilities in the same format, without touching any of the rest of the platform implementation.

Among the three versions we find:

- 8 cores: 4 P-Core + 4 LP E-Core with a 4-core Xe3 graphics card and four Ray Tracing units. You have the rest of the details in the characteristics table that follows this summary.
- 16 cores: 4 P-Core + 8 E-Core + 4 LP E-Core. The graphics of this model are the same as the basic model, with 4 Xe3 cores and 4 RT. This model has more PCI Express 5.0 lanes, with two available, perfect for integrating with other dedicated graphics.
- The most advanced model: 16 cores, 4 P-Core + 8 E-Core + 4 LP E-Core with 12-core Xe3 graphics with 12 RT. It reduces its number of PCI Express 5.0 (4x) lanes to make room for a noticeably more powerful GPU.
| Model | Core configuration | Graph | Ray Tracing Units | PCI Express 5.0 |
|---|---|---|---|---|
| Essential | 8 cores: 4 P-Core + 4 LP E-Core | 4-core Xe3 | 4 units | – |
| Intermediate | 16 cores: 4 P-Core + 8 E-Core + 4 LP E-Core | 4-core Xe3 | 4 units | 12 lines |
| Advanced | 16 cores: 4 P-Core + 8 E-Core + 4 LP E-Core | 12-core Xe3 | 12 units | 4 lines |
Architectures of the new P cores and E cores
One of the great advances of this generation is found in the growth of cache memories, which take up more space on the die and offer more capacity, in addition to multiple optimizations. An added cost that translates into more efficiency and also a double-digit increase in performance per clock cycle (IPC).

Intel also takes advantage of the modular format of this new processor and its modular packaging to compensate for the removal of die memory, introducing a cache on the memory sidein the interface that communicates with the system DRAM, with 8 MB capacity that will reduce data traffic to memory, with a significant impact on very specialized applications.

This cache increase, with a significant last level cache (level 3) shared by both the efficiency and performance coreshas a significant impact on the performance of the processor, which equals the performance of the Arrow Lake-H, being even more efficient than the Lunar Lake. This is an architectural optimization that brings the performance of processors with up to 45 W consumption closer to those designed for half the consumption, increasing performance per watt by up to 50%.

But surely the most important leap in this generation is not so much in its Cougar Cove, the performance cores, but in the Intel Darkmontwhich we have seen underpin the powerful performance growth of the Xeon 6+, which doubles its core density with the same consumption as the previous generation.

Completely independent graphics tile
One of the most powerful improvements of this generation is found in the GPU. Intel has completely redesigned it – you can see the details here -, being in the 12 Xe3 units where we will find the most important improvement. Best of all, the design, whether using one or the other GPU, remains unchanged, so integrators will have an easy time offering both variants in their different laptop designs or other EDGE solutions.
The graphics tile is now more efficient: the processor uses its lower consumption Darkmont cores by defaultleaving more space for the graphics to develop more stable performance and greater impact in games for this generation. Important improvements maintaining the consumption and cooling design of previous generations.

Next generation NPU 5
Another of the improved tiles is found in the integrated NPU. It does not present major performance improvements, it is around the same 50 TOPS, but here we list all the improvements that have been introduced in it, which result in the efficiency of this AI processing unit that should do nothing more than reduce load on the GPU.
This new NPU is not only just as fast, consuming less and taking up less space on the die, but it also introduces new modes that were not available before, unifying FP8 capabilities to all IA units of the processor.

Wireless connectivity
The new Intel Panther Lake processors maintain hardware MAC support within the same processor, but not the communications chip, which remains modular using PCI Express or USB expansion cards. What Intel does do is combine this processor hardware support, which they have been integrating for multiple generations, with new wireless chips that update the performance of these processors until they reach the best that technology can currently offer.

Devices that appear with Panther Lake starting in January 2026—we will surely be able to see them during CES 2026— They will come equipped with Wi-Fi 7 R2 and Bluetooth 6.0. In this analysis we go into detail about the features that are incorporated in this new generation, which are important and will revolutionize the way in which high-speed wireless networks are managed in portable and edge devices.

Video Processing Unit IPU 7.5
This generation also includes a new video processing unit. IPU 7.5 delves into the real-time video processing capabilities of these processors and enhances them through the use of AI located in the processor itself.
Introduces new, more realistic HDR techniques, with more dynamic width, but also two new techniques based on AI: on the one hand, noise reduction, with more defined images in low-light environments, and also localized tone adjustment, improving the contrast of the images. The result will be more realistic management of the cameras integrated into this type of devices, in addition to offering new features to devices connected to cameras that perform video processing functions in real time.

All this with more performance and less consumptionan average of 1.5 W less consumption in this unit, with native support for 4K and high-speed cameras with up to 120 FPS at 1080p resolutions. This unit supports up to three simultaneous cameraswithout the need for dedicated solutions.
When and how Panther Lake will arrive
As you can see, during ITT 2025—which we were lucky enough to experience in person in Arizona last week—the announcements are about technology, with brief demonstrations of real products that will soon hit the market. From laptops to EDGE products such as servers also based on the new Xeon 6+ ClearWater Forest architecture.

All these products, especially the most commercial ones, They will arrive in early 2026surely with announcements in January of next year and real availability during the first quarter of this year.
Intel has already begun mass producing the processors, in its three variants, in at least two plants located in the US. FAB 52 at the Ocotillo complex in Chandler, Arizona, is already manufacturing these new processors using its Intel 18A process; a process with important innovations that should support the next generations of Intel, where the brand also has a lot at stake.

On paper, Panther Lake offers important new features; added to a more ambitious attitude that we have been able to see in the speeches of its managers, It makes us think that Intel has a very interesting product that should translate into cutting-edge devices.
Intel Panther Lake Feature Summary
- Up to 16 CPU cores
- Up to 96GB LPDDR5, 128GB DDR5
- New generation IPU 7.5
- Up to 180 TOPS across platform (120 GPU + 50 NPU + 10 CPU)
- Next generation NPU 5
- New Xe3 GPU
- XeSS Multi-Frame Generation Technology
- Integrated Thunderbolt 4
- Improved Intel Thread Director and power management
- E-core Darkmont cores
- Cougar Cove P-core Cores
- Settings:
- 8c – 4Xe
- 16c – 4Xe
- 16c – 12Xe
- Up to 12 Xe-cores
- Support for Wi-Fi 7 (R2) and Intel Bluetooth 6 Dual
- Up to 12 PCIe Gen 5 lanes
- Intel Total Storage Encryption
- Intel Partner Security Engine
- New LPCAMM support
