Intel's 2026 calendar is busy. Between the refreshed Arrow Lake desktop CPU, new data center processors, the Arc C-series graphics cards, and the Panther Lake mobile SoC, Team Blue's newest tech will be coming in the new year, with Panther Lake being the flagship chip manufactured on Intel's latest 18A silicon manufacturing process.

Today, Intel confirmed a 2026 launch for Panther Lake and details on the chip architecture. Along with Intel's Clearwater Forest data center chipset which packs the same E cores and is produced on the same Intel 18A process, Panther Lake was the other, consumer-focused subject of Intel's recent tech tour, which offered deeper insights into the Panther Lake platform.

About this article: Intel paid for my accommodation and travel to the Intel Tech Tour. The company provided pre-briefings and documentation ahead of the official announcement, but did not have any input or oversight on this article.

Panther Lake: architecture overview

What do we know about the new CPU?

Panther Lake was first teased last year after the Lunar Lake launch, and is hoped to rescue Intel's reputation after many felt that the company had overpromised when it came to the release of Arrow Lake. Specs for Panther Lake leaked earlier this year, and the launch confirms that many of the details ended up being correct.

As an Intel 18A chipset, Panther Lake utilizes Intel's newest silicon manufacturing technology, RibbonFET transistors, and PowerVia front-side interconnect to enable area scaling and power efficiency along the silicon.

For context, RibbonFET is simply another name (given by Intel) to a Gate All Around Field-Effect Transistor, or GAAFET. TSMC is moving to GAAFET for its 2nm N2 node, whereas Samsung uses it with its 3nm 3GAE process node. RibbonFET uses ribbon-shaped channels (hence the name) for improved control, higher performance, and greater energy efficiency than the previous FinFET architecture. By fully surrounding the ribbon-like channels with the gate, RibbonFET minimizes current leakage, allows for faster switching speeds, and enables the creation of denser chips.

PowerVia, meanwhile, allows for backside power delivery throughout a chip, where signal wires and power wires are decoupled and optimized separately. With frontside power delivery, the standard of the industry now, there's potential for bottlenecking due to space while also potentially opening up to issues like power integrity and signal interference. PowerVia separates signal and power lines, resulting in theoretically better power delivery and a more effective yield overall from the fabrication and packaging process.

Like other recently released previous integrated chips, Panther Lake separates the GPU tile so it can scale independently next to the central compute tile and memory sub-system. Panther Lake's platform controller tile is off to the side of the SoC package, and is said to be manufactured externally. Intel didn't say which external provider was used, but the company has been looking to TSMC over the past two years to bolster its production capabilities. This architectural design follows the trend that Intel has been moving with for the past two years, kickstarted at the company as a response to competitors with Meteor Lake and continuing thereafter.

Panther Lake will ship in three configurations:

• 8-core CPU with 4 Xe GPU cores
• 16-core CPU with 4 Xe GPU cores
• 16-core CPU with 12 Xe GPU cores

Intel's Senior Principal Engineer, Arik Gihon, told the crowd at ITT, "Panther Lake builds on the architectural strength of both Lunar Lake and Arrow Lake. It inherits Lunar Lake power efficiency and power scaling from Arrow Lake." It's a bold promise to make, so long as Intel can stick the landing.

Panther Lake CPU tile

More than just core counts

Like previous generations of Intel chips, the Panther Lake package consists of the compute tile with I/O IP on the power control device tile, a separate GPU tile, and Intel's proprietary Foveros 2.5D packaging. All the Panther Lake configurations come with a mix of Cougar Cove P cores and Darkmont E and LPE cores, with an IPU 7.5 and NPU5 on the xPU tile, Xe media and display engines, and memory support for DDR5 and LPDDR5 memory. Plus WiFi 7, Bluetooth 6, and an Xe3 iGPU.

The new core designs for P and E cores are designed and optimized for Intel's 18A foundry process and offer better performance and additional IPC optimization. However, there's some weirdness happening with regards to the graphics tile, which we'll get to. There are three distinct configurations across the series, each targeting different performance tiers while sharing a common architectural foundation.

The 8-core Panther Lake variant combines four Cougar Cove performance cores (P-cores) with four Darkmont low-power efficiency cores (LP E-cores). It includes an 8 MB memory-side cache, 12 PCIe lanes split between eight PCIe Gen 4 and four PCIe Gen 5 lanes, and connectivity support for up to four Thunderbolt 4 ports, two USB 3.2 ports, and eight USB 2.0 ports. On the graphics side, the integrated Xe GPU is manufactured on the Intel 3 node and features four Xe cores and four ray tracing units. Memory support includes LPDDR5x up to 6800 MT/s and DDR5 up to 6400 MT/s.

The mid-tier 16-core Panther Lake configuration steps things up with four Cougar Cove P-cores, four Darkmont LP E-cores, and an additional eight standard Darkmont E-cores, retaining the same 8 MB memory-side cache. It expands PCIe connectivity to 20 lanes total, featuring eight PCIe Gen 4 and twelve PCIe Gen 5 lanes, alongside four Thunderbolt 4 ports, two USB 3.2 ports, and eight USB 2.0 ports. The integrated Intel 3-produced GPU remains similar to the 8-core model, offering four Xe cores with four ray tracing units, but memory support gets faster, packing support for LPDDR5x up to 8533 MT/s and DDR5 up to 7200 MT/s.

At the top of the stack sits the flagship 16-core Panther Lake with 12 Xe cores, which is a true showcase of Intel's hybrid design and the scalability of the graphics tile. Like the mid-tier model, it features four Cougar Cove P-cores, four Darkmont LP E-cores, and eight Darkmont E-cores, alongside the same 8 MB memory-side cache. However, its GPU tile is dramatically upgraded to 12 Xe cores with 12 ray tracing units, pushing integrated graphics performance well beyond previous designs.

Somewhat worryingly, this GPU-heavy configuration trades off PCIe connectivity, offering just 12 lanes total; eight PCIe Gen 4 and only four PCIe Gen 5 lanes. This is a pretty sharp drop from the 20 lanes available on the standard 16-core part. As well, this graphics tile is not manufactured on Intel 3, and is instead manufactured using an "external" process. Again, Intel didn't say who, but Intel has developed a strong relationship with TSMC over the past two years, and both Lunar Lake and Meteor Lake's graphics tiles were manufactured there. Connectivity remains robust, with four Thunderbolt 4, two USB 3.2, and eight USB 2.0 ports, while memory support climbs to LPDDR5x speeds up to 9600 MT/s.

Panther Lake's GPU tile raises questions

Reducing PCIe Gen 5 lanes is a bold strategy, but NPU 5 looks alright

Panther Lake's integrated GPU marks one of Intel's most aggressive steps forward in on-die graphics since the introduction of Xe-LPG in Meteor Lake. Based on the Xe3 architecture, these new cores will form the basis of Intel's Celestial dedicated GPUs, and they aim to bridge the gap between integrated and discrete-class graphics. Even the smallest configuration, featuring four Xe cores, represents a meaningful evolution over Lunar Lake’s Xe2 tile, offering improved ray tracing efficiency, a ton of optimisations, and a claimed improvement of more than 50% over Lunar Lake. Finally, Intel's Xe3 architecture supports Intel's XeSS 3 AI super-scaling and frame generation technology.

The flagship 12 Xe-core variant is particularly noteworthy. Panther Lake's iGPU builds on the integrated graphics tiles of the Lunar and Arrow Lake platforms, The Panther Lake GPU tile features up to 12 Xe-cores, up to 16 MB of dedicated L2 cache, and up to 12 ray tracing units. The adoption of LPDDR5x-9600 memory provides enormous bandwidth, offering up to 19.2 GB/s per channel.

Intel has also massively focused on AI workloads, with its Xe Matrix Extensions (XMX) units gaining some big improvements. Panther Lake's iGPU is said to offer up to 120 TOPS of performance, though this appears to be at INT2/INT4, so real-world use will likely clock in at a lot lower.

With all of that said, the decision to only pack 12 total PCIe lanes in the highest-end SKU, compared to 20 on the mid-tier model, shows that Intel appears to be prioritizing GPU tile real estate and memory bandwidth over external expansion. This could limit high-performance SSD or discrete GPU configurations in thin-and-light designs, though for integrated systems and ultrabooks, it's likely an acceptable trade-off.

Another curious detail is the manufacturing process of the graphics tile itself. Unlike the CPU and SoC tiles, which use the Intel 3 process, the GPU tile is reportedly produced externally, likely at TSMC, though this has not been confirmed.

When it comes to the NPU, Intel is continuing its strategy of featuring a built-in option for on-device AI workloads. The NPU itself is said to be capable of up to 50 TOPS (though precision wasn't specified, either), and just like NPU 4, this means it meets the minimum threshold for Copilot+ certification.

There are some pretty big improvements to write home about here: not only does it support FP8 precision,but Intel claims that it also has a more-than 50% increase in performance per watt while increase the TOPS per area. This means it takes up far less space on the SoC overall, freeing up room for other, arguably more important componenets that users will see the benefits of.

Right now, not many applications actually use the NPU, as most local workloads are easier to target at the CPU and are light enough that it's feasible to do so. For more intensive computation, developers will likely target the GPU instead. NPUs are incredibly fragmented, and while there are some applications you can run on them, these improvements are unlikely to be experienced by most users.

Can Intel stick the landing?

Panther Lake sounds good, but the proof is still to come

As is the case at all of Intel's tech tours, the company makes its latest CPU sound fantastic. Between Lunar Lake's leading power efficiency and Arrow Lake's decent performance, Panther Lake could be the chip that helps Intel reclaim its near dominance of the mobile Windows device market and combines the best of both worlds, especially given that Intel has been big on power efficiency for this generation, too.

"Lunar Lake already established that x86 has efficiency leadership," Gihon explained. While this is true in many cases, Lunar Lake wasn't entirely able to beat out the Arm-based efficiency of Qualcomm's Snapdragon platform or Apple's M-series silicon, though it certainly proved that there is potential for improvement.

We expect to get our first experiences of these chips in January of 2026, and we'll have to see whether Intel can truly recapture several segments all at once. With a powerful CPU, GPU, and NPU combination, Panther Lake just might have the raw power and efficiency to do so.