Intel readies its AI PC neural processor, a new take on AI Everywhere

AI was front and center at the kick off Tuesday of the third annual Intel Innovation conference, held in person in  San Jose, Calif.

Generative AI and LLM became a much bigger developer focus after ChatGPT rocked the world last year, and much of the attention has been on faster GPUs and CPUs for data center uses.  Nvidia’s GPUs have garnered most of the publicity and investor love, but Intel was out to show this week it cannot be overlooked by developers and investors for both its hardware and software prowess.

Intel clearly sees Nvidia as the competitor to go after--with its Intel CPUs, mainly--but Intel is also making software into a new line of business to generate revenues, largely around the AI craze.

CEO Pat Gelsinger pounced on an event theme of “AI Everywhere” in a keynote, surprising some who haven’t paid attention with a December 14 launch date for the AI PC running Intel Core Ultra processors with a relatively new abbreviation: NPU, for Neural Processing Unit. He had first described the coming AI PC era in July.

“AI will fundamentally transform, reshape and restructure the PC experience, unleashing personal productivity and creativity through the Power of the cloud and PC working together,” Gelsinger said in prepared remarks delivered Tuesday. “We are ushering in a new age of the AI PC.”

Upcoming Core Ultra processors, code-named Meteor Lake, will feature Intel’s NPU for power efficiency with AI acceleration and local inference on the PC.  They launch Dec. 14.

Intel considers Core Ultra an inflection point in client processing, since it is the first chiplet with Fovoros packaging, which jams together the NPU with Intel 4 process technology and graphics capabilities with Intel Arc graphics.

On stage at the event, Acer planned a sneak peek of an upcoming Acer laptop powered by Core Ultra. Jerry Kao, chief operating officer at Acer, said his company had been co-developing with Intel a suite of Acer AI apps using the Core Ultra.

The significance of AI on a PC is not to be overlooked, said Jack Gold, analyst at J. Gold Associates. “This is a big deal,” Gold said in an email. “Intel is the first to bring a true AI accelerator to a PC with Meteor Lake.” He noted that Microsoft is even adding capabilities in next-gen Windows to access the NPU.

Gartner analyst Avivah Litan said processing power on local devices will be “very important” to developers as more and more AI workloads run on local devices—whether PCs, mobile or edge devices.  Some sensors shown at Sensor Converge 2023 in June are being produced to support AI inference so that video data can be sorted locally to reduce the need to transfer to the cloud. 

Patrick Moorhead, analyst at Moor Insights and Strategy, sounded the same theme. 

"AI PCs are a very big deal. They mean we can do many of the generative AI magic tricks yet unconnected from the cloud, faster, and with higher levels of privacy and security," Moorhead said.

“Intel in AI is trying to show they are a true competitor in the AI space across the entire spectrum of compute needs,” Gold added.

To be completely accurate, Intel has already added Moviodius AI capability via add-on cards alongside its 13th-gen Core chips, which have appeared in the Samsung Galaxy Book3 Ultra and others. AMD has a rival Ryzen AI. 

Intel was also set to tackle a range of advancements  at the event, including a focus on an efficiency core alongside a productivity core for processing, something Arm has done for some time “but a major change for Intel,” Gold said. “You need to be able to power down for workloads that don’t need the highest compute needs.”  Running a 100 watt chip when only 20 watts is needed allows a developer to run a task on the e-cores and “save a huge amount of power,” Gold said.

Fifth generation Intel Xeon processors will also launch Dec. 14. Next year, an E-core processor named Sierra Forest will have a version with 288 cores.

In a briefing with reporters in advance of the event, Deepak Patil, general manager of data center AI, said Intel is “making AI more accessible at scale across the entire AI continuum: client, edge, network and cloud…Our customers must cover the complete model from the smallest to the largest with different use cases…Our customers are adapting and know the market is changing and they us to go through these changes.”

Intel also announced general availability of Intel Developer Cloud for testing high performance applications and said a  2023.1 release of Intel Distribution of its OpenVINO toolkit will help developers advance AI capabilities.

With OpenVINO, the “complexities with hardware are abstracted away,” said Pallavi Mahajan, general manager of the NEX group at Intel.

Gold said Intel won’t admit it directly, but it is trying to “break the stranglehold that Nvidia has on the CUDA proprietary dev platform” by pursuing a translation model to an open software platform with SYCL and others. “This is a big deal for many developers since developing in CUDA means your software only runs, or at least runs best, on Nvidia hardware, even though a developer may want to use others,” he said.

Also along the software dev theme, Intel said its Project Strata edge-native software platform will launch in 2024 with modular building blocks. The company described it as a horizontal approach  to scale the needed infrastructure for the intelligent edge and hybrid AI.

RELATED: PCs will decline- in ‘23 by the most in 40 years , IDC says