Skip to main content

The next Intel desktop chips may muddy the AI waters

Intel CEO Pat Gelsinger presents Intel's roadmap including Arrow Lake, Lunar Lake, and Panther Lake.

Intel is hitting a big milestone with Arrow Lake-S, its upcoming desktop processors. After all, those chips will be the first ones to feature a neural processing unit (NPU) in a desktop PC. However, impressive as the milestone might be, the CPU itself will be far from an AI workhorse. In fact, the NPU may be nowhere near being able to qualify for Microsoft’s Copilot+ PC program. The good news? It really doesn’t matter.

Before we dive in, it’s worth noting that Intel is yet to confirm the specs for Arrow Lake-S. However, Jaykihn on X (formerly Twitter) shared a list of various specs that are rumored to appear in the processors. This includes the trillion operations per second (TOPS) counts for the GPU, the CPU, and the NPU. That’s where things get a bit dicey.

According to the leak, the NPU will feature a measly 13 TOPS of AI performance. For comparison, Intel Lunar Lake CPUs, found only in laptops, offer 45 TOPS, while AMD’s Ryzen AI 300 series bumps that number up to 50 TOPS. Both qualify for the Microsoft Copilot+ certification, which calls for at least 40 TOPS. Why, then, would Intel release a next-gen product with a little over a fourth of the AI performance it already introduced in laptops?

It’s simple: Those NPU TOPS aren’t really needed in a desktop — at least not now. As Tom’s Hardware points out, NPUs are meant to handle AI and machine learning workloads while having little impact on the battery. That’s why we’re seeing these high TOPS counts in NPUs mainly found in ultralight workstation laptops. Seeing as battery life is not a concern for a desktop user, one big reason to even have an NPU in the first place is out.

Collection of various small improvements/changes shipping with Arrow Lake -S.

All specifications are “up to”, except for TOPS, which may exceed the listed number on the retail product.

— Jaykihn (@jaykihn0) July 9, 2024

Then, there are other factors. Desktop PCs that are meant to handle AI workloads should just be equipped with a discrete GPU, which may be a better option for handling AI anyways. Nvidia’s RTX 4090 features up to 1,321 AI TOPS, which explains why it’s still so expensive nearly two years after launch. The current demand for AI drives both AI laptops and desktop GPUs.

Even if Intel were to equip the Arrow Lake-S with a more capable NPU, it’d be useless compared to a discrete GPU from Nvidia’s latest generation. It doesn’t even have to be the RTX 4090; a cheaper GPU will still run circles around any sort of NPU.

It feels like, if these specs are true, Intel will be able to slap a label on Arrow Lake-S and talk about having the first desktop NPU for consumers — even if it has little impact on actual AI workloads. Ultimately, if the CPUs had to feature an NPU, it’s for the best that the NPU isn’t great, as there would be very few scenarios where you could take advantage of a 45 TOPS NPU in a desktop setting anyway.

It does, however, leave a question mark over certain features, such as the now-delayed Recall. The feature has so far been exclusive to Copilot+ PCs, though you have to assume Microsoft will one day consider bringing it to desktop if it turns out to be a success.

Monica J. White
Monica is a UK-based freelance writer and self-proclaimed geek. A firm believer in the "PC building is just like expensive…
How to watch Intel’s big Computex 2024 keynote tonight
Intel CEO Pat Gelsinger delivers the Day 1 closing keynote at IAA Mobility

Intel is the next big name that will be talking up its upcoming products and technologies at Computex, with a big keynote address from CEO Pat Gelsinger scheduled for tonight. While we won't be talking about his jacket like with the Nvidia CEO's, we may well hear about the CPUs that are going up against AMD's Zen 5 later this year.

Alongside new chip announcements, AI will certainly be a major component of the announcement, as it has been with just about everyone else's so far.
How to watch Intel's Computex 2024 keynote
Intel will hold its keynote at 11 p.m. ET/8 p.m. PT on June 3 (that's 11 a.m. local time on June 4 for anyone in attendance in Taiwan). If you want to watch it live, Intel has a livestream planned for its own website where you can register your interest and watch along at home.

Read more
AMD’s new Ryzen AI 300 chips look like the real deal
AMD's CEO delivering the Computex 2024 presentation.

The pressure is on at Computex this year. With the May announcement of Copilot+ PCs and Microsoft's deep partnership with Qualcomm, the stakes were high for AMD coming into the show. But the company certainly didn't show up empty-handed.

Its announcements have all centered around Zen 5, the company's latest architecture, both on desktop and mobile. But at the moment, these mobile chips feels especially noteworthy in light of Copilot+. AMD calls it the Ryzen AI 300 series. It's a complete rebrand for AMD, not unlike Intel's move to "Core Ultra" in its most recent generation. But this time, it's all about AI.

Read more
Intel is ready for Copilot+ PCs with Lunar Lake
On-package memory on Intel Meteor Lake processors.

The talk of the town in the world of PCs is Snapdragon's new X Elite processor, but Intel wants you to know it's not down for the count in this new era of Copilot+ PCs. The company is previewing its next-generation Lunar Lake CPUs before it fully reveals them at Computex 2024, and they sound like a massive upgrade.

Although we saw a neural processing unit (NPU), which is used for AI tasks, in Intel's last-gen Meteor Lake chips, it wasn't that powerful. Snapdragon all but nullified Meteor Lake by announcing the X Elite, which has an NPU capable of 45 trillion operations per second (TOPS). That's more than four times what Meteor Lake's NPU was capable of.

Read more