Pushing Open the Door to a New Era of PCs: An Analysis of Intel's New 'AI Processor'
-
Earlier this September, our team at SanYi Life combined the latest architectural information and some exclusive internal data to provide an early 'analysis' of Intel's newest generation of mobile processor architecture.
Of course, by now everyone knows that the subject of our 'analysis' was an ES version of the just-released first-generation Core Ultra processors.
To be fair, the materials we received earlier were still in the preliminary stages. Although they were quite detailed, there remained some gaps compared to the final version that would eventually hit the market. This is precisely why we were particularly keen on Intel's new product launch event on December 15, 2023. In the end, the event did not disappoint.
Larger and More Powerful: The Fifth-Generation Scalable Xeon 'Game Changer' Arrives
It must be said that the biggest difference between Intel's latest product launch and previous ones lies in the fact that, for the first time, they combined enterprise-grade cloud server CPUs with consumer-focused edge computing device CPUs in the same event.
The first to debut was the fifth-generation Xeon scalable processor. According to the technical documentation, compared to the fourth-generation Xeon scalable processor, the new CPU, codenamed 'Emerald Rapide-SP,' still utilizes the Intel 7 process but features a significantly redesigned architecture.
The fifth-generation Xeon Scalable processors now feature a "complete specification" with 64 cores, up from 60 in the previous generation. These cores are organized into two MCM modules instead of the previous four, meaning each CPU module in the new architecture packs over twice the core count of its predecessor.
Beyond core scaling, the fifth-gen Xeon Scalable processors transition from Golden Cove to Raptor Cove architecture—mirroring the 13th/14th Gen Core series—with significant enhancements to cache and memory subsystems. For example, comparing the new Xeon 8592+ to the previous Xeon 8490H reveals a leap from 112.5MB to 320MB of L3 cache. Memory bandwidth also sees major gains, with baseline eight-channel DDR5-5600 replacing DDR5-4800.
These architectural improvements primarily serve cloud providers like JD.com, Alibaba, and Baidu, accelerating AI model deployment and iteration in cloud environments. However, Intel's consumer-focused Xeon W lineup will likely inherit these advancements, making this launch a preview of next-gen workstation processors.
More Cores, Stronger iGPU, and NPU: Core Ultra Arrives
Now comes the highlight of this launch event - the brand-new Intel Core Ultra processors.
In terms of architecture, the first-generation Core Ultra is Intel's inaugural processor featuring a multi-module, multi-process hybrid packaging design. Each core consists of four modules: a CPU module using Intel4 process technology, a GPU module utilizing TSMC's 5nm process, an SoC module based on TSMC's 6nm process, and an IO module also built on TSMC's 6nm process.
What are the advantages of this design? On one hand, it allows the processor to integrate more functional components. Taking the flagship Core i9-185H as an example, its CPU module integrates 6P+8E cores (14 CPU cores in total), while the SoC module contains 2 additional E-cores that can completely shut down the CPU module during ultra-low-load operations to achieve better power efficiency.
For instance, the ARC GPU module in Core Ultra now features a more "complete" Xe architecture. It boasts up to 128 execution units (compared to the previous generation's 96), with a maximum operating frequency of 2250MHz, 64 texture units, and 32 raster units. Additionally, it supports hardware ray tracing and XeSS AI-powered super-resolution anti-aliasing. With this configuration, the new integrated GPU achieves FP32 performance of up to 4.608TFlops—double that of its predecessor.
Beyond this, Core Ultra marks Intel's first mobile processor to integrate an NPU, delivering up to 10TOPs of computing power according to technical documentation. It also includes a built-in Thunderbolt 4 controller.
Apart from major architectural and design improvements, another standout feature of Core Ultra is its remarkably low default TDP of 28-45W for its high-performance variants, with a maximum TDP of just 115W. Since the 11th generation, Intel's high-performance mobile Core processors haven't had such low official TDP ratings—flagship models like the 14900HX even carry a 157W official maximum TDP. Based on various leaks, exceptional energy efficiency is likely to be the primary highlight of the first-gen Core Ultra in real-world usage.
How to Make 'AI PC' Truly Practical: Intel Has Indeed Set a Good Example
If you've been following recent developments, you may know that Intel's competitors, such as Qualcomm and AMD, have already launched mobile CPU product lines equipped with independent NPUs. In comparison, Intel's Core Ultra series arrived relatively late.
However, this doesn't mean Intel started later than its competitors on the 'AI PC' technology roadmap. In fact, as early as 2019, Intel had already integrated the DL Boost instruction set—designed to accelerate deep learning computations—into its Core X-series processors (Cascade Lake).
Later, with the 10th and 11th generation Core mobile processors, the DL Boost and AVX-512 instruction sets were further extended to more mainstream products. This enabled many laptops at the time to support AI features like camera centering, microphone noise reduction, and video super-resolution.
With Intel's introduction of the Xe architecture into the integrated graphics of Core processors, these GPUs have also gained AI acceleration capabilities. For example, in some 12th and 13th-generation Core laptops and even desktop PCs, AI-powered features like voice assistants and background blurring can be implemented using the GPU's computing power.
As a result, although the first-generation Core Ultra processors appear to have only an NPU as their "AI unit," Intel has actually developed a heterogeneous acceleration system called "XPU." This system enables the CPU, GPU, and NPU to handle different AI tasks based on code requirements and even collaborate on computations.
More importantly, Intel showcased at today's launch event that they have partnered with numerous domestic software vendors to demonstrate how their products leverage AI acceleration on Core Ultra. According to Intel, even at this early stage of the product's release, Core Ultra has already been officially optimized for over 100 software applications, allowing them to harness the AI acceleration capabilities.
It's worth noting that Intel's unique heterogeneous processing design and their extensive efforts to promote "AI PC" adoption through software optimization may be the most significant advantages of Core Ultra compared to its competitors.