Throughout the better part of 2023 since the generative AI hype grabbed and held on to tech headlines, the idea of the "AI PC" has emerged as a great hope for the PC sector that has suffered a brutal post-pandemic reckoning with unprecedented declines in PC shipments and OEM revenues.
Intel's AI Everywhere event and the launch of its Core Ultra family of processors Dec. 14th at NASDAQ headquarters in New York City is an important AI PC moment for the chip giant. According to Intel CEO Pat Gelsinger, the company's newly launched Core Ultra family represents a seismic shift in their mobile processor product line that will usher in the next era of the PC.
What is an AI PC anyway?
Pat Gelsinger coined a new catchphrase during his opening presentation that caught my eye, "Augmenting Intelligence," which spoke to the essence of what the AI PC is--an AI augmented PC. Is this really a novel thing?
We have seen AI compute augmentation in smartphones since 2017 when Apple and Qualcomm introduced NPUs (neural processing units) into their mobile SOCs. This AI augmentation was borne out of the necessity of lowering the power envelope of AI processing largely centered around computational camera and sensor fusion functions for a wide range of applications such as computational photography and motion tracking and analytics for fitness.
Although the ability to run large language models (LLMs) on the PC has become a staple in the AI PC talk tracks of Intel, Qualcomm, AMD and others, it is more practical and likely that low-power AI functions such as gaze correction for PC cameras will benefit the most from the "AI PC" in terms of lower power consumption and performance.
While the Core Ultra is the first Intel PC processor to feature an integrated NPU, Apple introduced a 16-core Neural Engine to Macbooks with their Apple Silicon M1 SoC more than three years ago. More recently, AMD launched their Ryzen 7000 series at the beginning of this year which Lisa Su, CEO of AMD claimed was the first x86 PC processor with an integrated NPU.
A First Phase in Intel-driven PC Reinvention
When it comes to the PC, Intel is still the 800-lb titan in the room with its well-rooted silicon and underappreciated software portfolio, ecosystem partnerships, and developer community support. Specs aside, Intel is still the company poised to drive the reinvention of the PC in an era of "Augmenting Intelligence."
Going into the AI Everywhere event, I didn't expect Intel would have a leadership offering. In my view, all they needed was something competitive to get the ball rolling with their scale and strong OEM relationships from Lenovo to HP which Michelle Johnston Holthaus, EVP of Intel's Client Computing Group, prominently emphasized. It seemed that Core Ultra delivered.
Frankly, it’s difficult to tell. Most vendor benchmarks have been nothing short of diversionary and unhelpful, suffering from baskets full of choice apples-to-oranges comparisons. For example, Intel benchmarked their top-of-the-line Ultra Core SKU (at least until the Core Ultra 9 185H becomes available Q1 2024) against Qualcomm's 2-year-old Snapdragon 8cx Gen 3 processor. It was a not-so-subtle snubbing of the upcoming Snapdragon X Elite PC processor that sports Qualcomm's new Oryon CPU core expected to launch in the middle of 2024.
The one benchmark statistic that stood out was the Core Ultra's 8 percent multi-threaded performance-per-watt improvement over Intel's 13th Gen i7-1370P (Raptor Lake) which seems modest considering the Core Ultra's new tile architecture and upgrade to Intel 4 which is Intel's first process node to use EUV lithography. That being said, Michelle Holthaus claimed that Core Ultra was "up to 30 percent lower on processing power" for general computing tasks and 2.5 times more efficient in processing AI workloads versus prior-generation Intel Core processors.
A reimagining of Intel PC compute
Core Ultra represents a huge architectural revamping of Intel’s mobile processor line. The Ultra Core uses a chiplet-based architecture (tiles in Intel's parlance) popularized by AMD years ago. The processor also uses Foveros advanced 3D packaging and implements Universal Chiplet Interconnect Express (UCIe) interconnects, a world's first.
The Core Ultra design is composed of five tiles of differing source and process nodes for optimal cost-performance. The main compute tile (CPU) is based on Intel 4 process node while the graphics tile is based on TSMC's N5. The SoC tile and I/O tile are produced on TSMC's N6. Finally, the Foveros base tile that integrates the functional tiles together is based on Intel 22FFL (Intel 16) process.
The Core Ultra family comes in two series upon launch, the 24W H series and the more power economical 9 to 15W U series of processors. Each of the eight SKUs in the Core Ultra family at launch comes in a range of CPU compute configurations of three performance tiers comprised of performance (P-Core), efficiency (E-Core), and low power efficiency cores (LP E-Core).
Intel decided to port ARC GPU to the Core Ultra H series bringing discrete GPU capabilities and features to systems with at least 16GB of system memory in a dual-channel configuration. The integrated ARC GPU delivers premium features such as hardware-accelerated ray tracing and up to twice the gaming performance of prior generation Core processors. The U series SKUs come outfitted with integrated Intel Graphics.
Tying all of this diversity of compute together is Intel's 3D Performance Hybrid Architecture that leverages an improved Thread Director to orchestrate and optimize the placement of AI workloads across three "AI engines," the CPU, the GPU, and the NPU which resides on the SoC tile along with two Low Power E-Cores. Intel cites this architecture as a key to the intelligent power efficiency of the Core Ultra family of processors.
What will AI PC mean for users?
The big question is what is new with this idea of AI PC? The reality is most major ISVs such as those Intel cited--Adobe, Zoom, and others--already have AI-infused features that have been running on desktops to mobile phones for years. In many ways, we have had a mature case study with three generations of Apple's M Series Macs to understand the impact of AI on the PC computing experience which borrows from a much longer legacy of AI computing on smartphones.
Generative AI is broadly seen as a demand catalyst for the AI PC. Time will tell if generative AI lives up to exaggerated expectations. At a minimum, it will be just another AI workload handled by the best accelerator for the job, whether it is a CPU, GPU, or NPU. Given growing concerns about confidentiality and privacy, there is no doubt that the AI PC will have an important role in shaping a hybrid generative AI future.
Early days for AI PC and Generative AI
The AI PC race is more than a one-horse race. A diverse range of players, design approaches and architectures are vying for mindshare, influence, and market share. Qualcomm put up some very impressive single threaded and multithreaded benchmark numbers for their X Elite SoC in Maui at Snapdragon Summit 2023, suggesting that the San Diego-based chip firm has achieved a highly competitive if not leadership threshold with their all-performance core design. AMD continues to press their advantage in the x86 lane with their newly announced Hawk Point Ryzen 8040 processor with devices available in Q1 2024. AI PC competition is already intense.
Despite established and emerging competition, Intel has the ecosystem scale and the integrated software across XPUs to accelerate developer adoption of the new hybrid/heterogeneous compute capabilities of Core Ultra in an era of AI-augmented PC computing. Intel touts that over 100 ISVs are onboard to develop AI PC applications. This is an advantage that cannot be trivialized along with Intel's scope, scale, and leadership legacy in client computing.
Intel's Core Ultra is a timely entry into the AI PC race. From the looks of it, Core Ultra is a competitive offering wrought from a host of profound product re-invention and the application of advanced Intel technologies.
Only time will tell if Core Ultra, the AI PC, and generative AI sum up to a Centrino moment.
Leonard Lee is founder and executive analyst at neXt Curve, a research advisory firm focused on Information and Communication industry and technology research. He has worked as an executive consultant and industry analyst at Gartner, IBM, PwC and EY and has advised leading companies globally on competitive strategy, product and service innovation and business transformation. Follow Leonard on LinkedIn: linkedin.com/in/leonard-lee-nextcurve