Intel has presented the Core Ultra Series 3 processors at CES 2026. A new generation of processors that aim to be a turning point to overcome one of the worst moments in the history of the chip giant. And, on paper, they really promisein terms of performance, energy efficiency, design, chip integration and manufacturing.
The CEO of Intel has assured in the presentation that the Core Ultra Series 3 will mark the beginning of the “next evolution of PCs”. According to the company, it will be the most widely adopted AI PC platform globally that Intel has ever released. They will be more than 200 partner designs those that will be launched under this platform in pre-orders starting today, January 6, and global availability starting January 27.
They will have to compete with the new Ryzen AI 400 and AI Max+ that AMD has also presented at the Las Vegas fair and with what comes from the ARM platform, from manufacturers such as Apple and also from others such as the new Qualcomm Snapdragon.
Core Ultra Series 3, basics
Intel had already advanced at Computex 2025 the main characteristics of its new generation of processors that are part of the family codenamed ‘Panther Lake’ and our media was able to see them first hand in an exclusive presentation that we talk to you about in depth in this special article.
Its technical foundations were known, starting with a design in blocks (chiplets) where each of them included a series of key components. Each block communicates with the others through a coherent cross-block encapsulation and unification system to ensure proper operation. This type of design combines several different silicon plates on one fundamental motherboard using Intel’s Foveros 3D packaging technology, offering highly scalable solutions integrating the CPU, GPU and neural processing unit in the same package.
The Core Ultra Series 3 also stands out for its manufacturing, since it is the first AI PC platform based on Intel 18A process node. A novel process technology, in an effort by the company to reach TSMC’s chip manufacturing capabilities, produced directly in Intel factories in the United States and that the company intends to use for its Intel Foundry division to manufacture chips for third parties. External sources speak of conversations with potential clients as important as Intel or NVIDIA.
Going into what may interest a consumer the most, say that the new chips combine “Cougar Cove” performance cores with “Darkmont” efficiency and low-consumption cores. Intel does great performance promises for high-end Core Ultra Series 3 processors: up to 60% faster multi-core CPU performance compared to previous Core Ultra 200V chips, and up to 77% faster integrated GPU performance.
All Panther Lake chips will include the same neural processing unit (NPU), capable of reaching up to 50 billion operations per second (TOPS). This puts it above Microsoft’s Copilot+ PC platform requirement, although it falls slightly short of the 60 TOPS that AMD claims for its Ryzen AI 400 series and the 80 TOPS that Qualcomm says its Snapdragon X2 chips can achieve.
Intel also says it prioritizes power efficiency, so while some of the new chips can run at up to 80 watts for short periods, the company says a basic laptop could stream Netflix videos for up to 27 hours.
The graphic section deserves a separate chapter
The Intel Xe3 architecture that we review in depth in this article represents Intel’s biggest advance in several generations. The high-end Core Ultra Series 3 will use the new Arc B390 GPU with 12 Xe cores, which promise a performance increase that the company places at a whopping 77% between generations. The company says that it far surpasses the Radeon in both scaling and resolution. Intel even says it is developing an independent SoC that will be comparable in performance to an RTX 4050 for laptops.
Intel is working with developers to integrate its latest XeSS3 multi-frame scaling technology into numerous titles since the launch of these chips, with Battlefield 6 being the first to offer official support. With the Frostbite engine now supporting XeSS3, we can expect many EA games to adopt this technology soon. Additionally, Cyberpunk 2077 is also supported on day one, which triples the frame rate compared to XeSS2. Intel also claims to surpass NVIDIA’s DLSS with the RTX 4050, especially since XeSS3 includes a multi-frame generation, unlike the RTX 4050.
Core Ultra Series 3, families and versions
The chiplet-based approach allows Intel to combine the aforementioned modules to offer solutions other than these Panther Lake. In fact, the launch will begin with 14 chips integrated into different product families which we can summarize in:
- The Core Ultra X9 and Core Ultra
- The Core Ultra 9 and 7 processors will use the same technologies, but with only four GPU cores and support for LPDDR5x-8533 or DDR5-7200 DIMM modules. However, they will offer 20 PCI Express lanes, compared to 12 for the X9 and X7 processors, meaning they will integrate better with dedicated GPUs.
- Core Ultra 5 chips are mostly lower-end models with fewer CPU cores and 4- or 2-core GPUs. But, as usual at Intel (complicating the understanding of the offer) this group includes the Core Ultra 5 338H, which has 12 CPU cores and a 10-core Intel Arc B370 GPU.
At launch, the top of the range platform will be the Core Ultra X9 386H. A development with 16 cores (4P + 8E + 4LPE), maximum frequency of 5.1 GHz, 18 Mbytes of cache, 50 TOPS NPU, integrated Arc Pro B390 graphics with 12 cores, support for LPDDR5/X 9600 memory and a TDP consumption from 25 to 80 watts.
In terms of connectivity, the Core Ultra Series 3 will support Wi-Fi 7, Bluetooth 6.0, four Thunderbolt 4 ports and in the high range up to Thunderbolt 5.
Core Ultra Series 3 Edge
Intel has also presented a variant specially designed for integrated devices. For the first time, along with their PC counterparts, Series 3 edge processors (typically offered on a segmented platform) will be certified for embedded and industrial use casesincluding extended temperature ranges, deterministic performance and 24×7 reliability.
Intel Core Ultra Serie 3 promises competitive advantages in critical edge AI workloads with up to 1.9x large language model (LLM) performance, up to 2.3x performance per watt per dollar in end-to-end video analytics, and up to 4.5x performance in vision language action (VLA) models. Integrated AI acceleration will enable higher total cost of ownership (TCO) through a single system-on-chip (SoC) solution versus traditional multi-chip CPU and GPU architectures.
In this regard, note that Intel has a new platform AI Super Builderwhich looks like an interconnected system of on-premises and cloud AI. We’ve heard a lot about hybrid AI, but it seems like Intel is promoting it. As for the bet on local computing AI This is very relevant at a time when RAM prices are out of control and service costs are increasing in areas where AI data centers are operating.
