The AI PC, propelled by Intel’s Meteor Lake, is nearly right here. So why do you have to care?
It’s the billion-dollar query. Intel is constructing NPU AI inferencing engines into its processors starting with its 14th-gen Core chip, Meteor Lake, often known as the Core Extremely. Robert Hallock, an AMD veteran now overseeing technical CPU advertising for Intel’s microprocessors, mentioned Tuesday that Intel’s objective by 2025 is to ship 100 million “AI PCs,” a time period Intel CEO Pat Gelsinger started utilizing in July.
Intel introduced the structure behind the 14th-gen Core Extremely chip on Tuesday, the identical day as Intel’s Innovation convention started in San Jose. Intel used the convention to pitch builders each on what an AI-powered future will appear like, particularly one powered by Intel.
Intel’s Core Extremely is notable for a number of issues, together with its transfer to the Intel 4 course of expertise and the disaggregation of the standard two-die design for 4 separate “tiles” all mounted on an interposer. Transferring to Intel 4 helped reduce energy by half over the Thirteenth-gen Raptor Lake, assisted by new low-power E-cores. However the important thing addition is the NPU, the AI inferencing engine that may assist convey AI to the plenty.
Intel
“We see the AI PC as a sea change in tech innovation,” Gelsinger mentioned throughout his opening keynote.
That’s the hope, anyway.
A fourfold technique to win over shoppers
The implicit theme of Intel’s Innovation convention was AI, AI, AI — from Gelsinger’s opening keynote to late afternoon panel discussions. Intel is making an attempt to actually carve out a brand new marketplace for AI on the PC the place none has existed: “That is class creation, at its best,” Gelsinger mentioned in response to a PCWorld query on how it might do exactly that.
However AI has primarily existed within the cloud, led by ChatGPT, and on smartphones, with portrait mode and numerous filters. On PCs, it’s confined to a couple Home windows Studio Results applied sciences. Intel is making an attempt to ascertain native AI on the PC as one thing that eliminates the spherical journey to and from the cloud whereas being personal sufficient to speak to solely you. However what, precisely? Properly, one thing.
The technique seems to be fourfold. First, design the {hardware}. Intel unveiled the design of the NPU to press and analysts at an occasion in Malaysia, promising that the NPU would supply a extra energy-efficient method for processing AI duties than a standalone CPU or a GPU. Accomplished. Intel’s Gelsinger additionally confirmed off 2024’s Meteor Lake successor, Lunar Lake; Arrow Lake, a chip on the 20A course of node, and 2025’s Panther Lake, heading to the fab in 2024 for later manufacturing. The CEO additionally demoed an AI process working on Lunar Lake’s AI engine.
(There’s a slight twist. Whereas Intel talks concerning the NPU as an “AI engine,” it’s actually just one a part of it; a video codec, for instance, performs decoding and encoding features. There are two components to “AI;” the coaching, and inferencing (predictions) based mostly upon the educated fashions. For now, the AI engines in PC processors can solely carry out inferencing.)
Second, exhibit the AI apps that exist right now and tomorrow. Gelsinger’s keynote showcased a number of functions that would use AI in numerous capacities: Deep Render, which makes use of AI to compress information by 5X; Rewind.ai, a listening to assist that additionally transcribes what it hears for future reference, together with ChatGPT-like queries; and Fabletics, which along with accomplice Match:match AI creates a digital avatar of you to attempt on garments.
YouTube / Intel
Notably, a lot of the functions averted the “conventional” makes use of of AI: generative AI artwork and ChatGPT. As an alternative, they confirmed off how AI might be used to allow revolutionary new functions. (Take a look at these 12 AI companies that may enhance your life proper now.)
That’s the place the rubber will meet the highway. Sooner or later in time, shoppers are going to ask what they’ll truly do with AI. To some extent, that query has been answered with ChatGPT and associated apps. However the subsequent query is why a client would wish AI on their PC, and never simply through a connection to the cloud. It’s that query that Gelsinger’s demos tried to reply, to middling success.
The third a part of Intel’s AI technique seems to attempt to elevate the APIs which allow AI. Executives talked about how the Core Extremely’s NPU would speed up numerous APIs, together with OpenVINO, which Intel has helped develop as an AI-centric API. (If a developer helps it, it may doubtlessly give Intel a bonus, analysts at Innovation mentioned.)
Mark Hachman / IDG
Lastly, ship at scale — for those who construct it, they are going to come.
“I’m going to ship a cadence of merchandise yearly, I’m going to ship thousands and thousands of items subsequent yr, and I’m going to ship billions of TOPS (tera operations per second, a measure of computing velocity) that builders can design to, and that basically opens the ecosystem up,” mentioned Michelle Johnston Holthaus, an Intel govt vice chairman and normal supervisor of Intel’s Shopper Computing Group.
Early days for AI
Will all of it work? Intel executives are additionally practical. Gelsinger referred to being within the “first inning” of AI. Hallock mentioned that Intel would launch its first AI/NPU benchmarks in October — with out saying how Intel was anticipated to fare, and even what benchmarks can be used.
The message is that Intel supplies open AI, reliable AI, and aggressive AI — however there’s nonetheless rather a lot that Intel seems to be leaving to religion. The query of what “good AI” is, or how shoppers are anticipated to check it to AMD and Qualcomm, is one Intel hasn’t actually answered but.
“I imply, these are substantive positive factors that we’re seeing for the platform, the place we’re fairly enthusiastic about it, and our merchandise are comfortably forward of the competitors for it,” Gelsinger mentioned in response to PCWorld’s query on how it might promote AI to shoppers. “So we’re going to distinguish on the deserves of the merchandise and available on the market integration and the work that we do.”
Is that the reply you need? Most likely not. However shoppers most likely received’t run out to purchase a Core Extremely laptop computer simply because it has AI, both.
What AI’s speedy adoption has taught us, nevertheless, is that after a viral app like ChatGPT takes maintain, everybody rushes to attempt it out. Intel (and its rivals) are constructing out AI’s basis, hoping that subsequent viral app is ready within the wings.
Disclosure: Particulars of the Meteor Lake structure had been unveiled at an Intel occasion in Penang, Malaysia. Intel paid for PCWorld to fly there, plus meals and airfare. PCWorld retained editorial independence all through the method. This story was up to date at 2:45 PM on Sept. 20 with further particulars.
