Everybody was spending so much time basking in the schadenfreude of Intel’s financial woes, they forgot about the company’s legitimately successful launch of the Lunar Lake consumer chips last year. Now that Intel has some support from Nvidia and backing from the current nativist regime occupying the White House, Intel is once again trying to push laptop battery life further without slacking on performance with Panther Lake chips. What’s more interesting this time is how Intel hopes to dethrone AMD for graphics capabilities on a single chip.
(Full disclosure: Intel invited me to its chipset fab in Phoenix, Ariz. Travel and lodging were paid by Intel, but Gizmodo did not guarantee any coverage as a condition of accepting the trip.)
Intel promised Panther Lake chipsets would maintain the battery life expectations of the past chip generation, but now it will have options for better graphics performance thanks to a novel way of positioning the GPU, or graphics processing unit. Modern chips build entire GPU sections onto the die, namely, the small block of the semiconductor that accounts for most of the processing potential. This naturally limits the max size of the GPU section. Despite that limitation, recent AMD APUs, or accelerated processing units, like the Ryzen 9 AI Max+ 365 show enormous potential for graphics processing. Intel’s solution is simple, at least on paper. Intel removed the GPU onto its own die, and included a special “die-to-die interconnect” that it claims offers seamless communications, as if the GPU tile existed together with the compute tile.
Panther Lake builds a bridge from CPU to GPU
Intel detailed three different versions of Panther Lake using Intel’s new 18A 3nm process with an 8MB side cache. The company stopped short of mentioning how many varieties of chips or SKUs customers would need to learn before buying a PC (last year’s Lunar Lake had nine total SKUs). The smallest of the three is an 8-core CPU featuring four performance (P) cores and four low-power efficiency (LPE) cores. The performance cores are based on the new Cougar Cove architecture, while the efficiency cores are now labeled Darkmont. There’s another 16-core version of the chip with an additional eight Darkmont efficiency cores.
Panther Lake is supposed to one-up both Lunar Lake and the much-maligned Arrow Lake chips. Its CPU performance is supposed to hit 10% better single-thread performance over last year’s light-laptop chips at the same power, but more importantly, it could do 50% better with multi-thread performance, which could make for faster work in intense tasks like video editing.
Panther Lake’s GPU now sports the new Xe3 GPU cores. The top-end version of Panther Lake can sport 12 Xe3 cores compared to just four. Last year’s Lunar Lake chips maxed out with eight Xe2 GPU cores. Suffice it to say, the GPU is more powerful than before, especially with its new 12 ray tracing units to support more accurate lighting effects in 3D programs or games. Intel claims the new GPU has 50% better performance compared to the top-end Lunar Lake and Arrow Lake chips. Bigger is going to be better, at least according to Intel, especially since Panther Lake is supposed to offer the same power as Arrow Lake-H chips at 40% less wattage. Xe3 is also supposed to hit 50% better top-end performance over Lunar Lake, but of course, that’s going to demand more juice from your machine.
Top-end Panther Lake chips can expect to support up to 96GB of LPDDR5 RAM, plus Bluetooth 6 and Wi-Fi 7. Otherwise, the chips may be able to make video calls look a little clearer thanks to improved signal processing from the webcam. We’ll all need to be patient, little PC buyers, for accurate numbers in gaming and graphics benchmarks, which likely won’t be around until the company starts shipping the chips later this year. For the rest of us who care more about how these CPUs perform on a machine you can buy, we’ll see more as major laptop makers share their wares at CES 2026 in January.
What’s the competition going to be like?
Last month, Qualcomm shared the details on its next-gen Snapdragon X2 Elite and Elite Extreme chips designed for lightweight PCs. The ARM-based CPUs were supposed to show up Intel’s Lunar Lake and AMD Strix Point with better performances at much lower wattages, meaning they could be even better for battery life than the previous generation. Intel’s promising something very similar, but with the same x86 microarchitecture, which won’t lead to the numerous issues with app and driver compatibility faced by Qualcomm-powered PCs. As I mentioned last month, gaming is still Qualcomm’s Achilles Heel for its PC endeavors.
Qualcomm is still fighting to keep in front of the AI wars. TOPS, or trillions of operations per second, is normally how companies describe the performance of their separate neural processing unit (NPU) dedicated to low-end or repetitive AI tasks. And still, it’s rarely a good indicator of how strong the chip is for processing AI. Qualcomm’s Snapdragon X2 Elite features a new NPU promising 70 TOPS of performance. The Panther Lake’s new NPU 5 on all varieties of the chip goes up to 50 TOPS of performance. What does any of that mean? Nothing, at least in practice. A PC’s GPU will still be better for more-intensive AI processing than the NPU can handle.
So we still need to see what comes out of AMD, but I wouldn’t be surprised if even more chips down the line feature a die-to-die GPU connection. Chipmakers may keep pushing AI performance, but if it means we get laptop chips that can play games without the need for battery-sucking discrete GPUs, there’s reason to get excited.