Intel touts new Xeon chip's AI power in bid to fend off AMD, ARM advances

Intel on Monday revealed new versions of its Xeon 6 server processors, in a bid to proliferate AI processing throughout its data center product line as it fends off incursions on two fronts — from AMD and ARM Holdings.

The new processors, dubbed Xeon 6 6500 and 6700, extend the chip giant’s product lineup first announced in September 2024. Code-named “Granite Rapids,” the Xeon 6 chips feature what are called performance cores, dozens of individual computing elements designed to deliver the most powerful computing activity in the company’s chip lineup.

Also: The fall of Intel: How gen AI helped dethrone a giant and transform computing as we know it

Intel’s initial 6900 Xeon 6 chips, announced in September, offer 128 of the performance cores in each chip, whereas the 6500 and 6700 chips offer lower core counts, 16 to 86, at lower prices and lower power consumption. (A separate series of chips, the “E” series, is meant to focus on energy efficiency in Xeon 6.)

In a briefing with media prior to Monday’s announcement, Intel chief architect Ronak Singhal emphasized the abilities of the 6500 and 6700 against AMD’s latest x86-based CPUs, the 9005, code-named “Turin.”

The Xeon chips perform higher computation with fewer of the cores than AMD’s 9005, said Singhal, making for a lower total cost of operation (TCO), claimed Singhal. He showed a slide referencing “up to 50% higher AI performance with one third less cores.”

“Even at the lower core counts, we’re able to provide higher performance across the range of these workloads,” said Singhal, where workloads range from “general compute,” such as running databases, to “high-performance computing,” such as running scientific models, to AI broadly, including inference and training of AI models.

Singhal emphasized the company’s proliferation of special AI-friendly operations in the Granite Rapids chips, known as Advanced Matrix Extensions (AMX). Those chip instructions allow the cores to perform the matrix multiplications of linear algebra, which make up most of AI processing, alongside traditional computer instructions.

The AMX instructions are winning Xeon more usage at cloud computing giants, said Singhal. 

“For instance, go to your favorite cloud service provider, and see what they’re saying about using AMX with their end customers,” said Singhal. “You’ll see them talk about how their end customers are renting Xeon instances to take advantage of AMX, how they’re doing that in some confidential compute scenarios.”

Also: AI data centers are becoming ‘mind-blowingly large’

The announcement of the 6500 and 6700 is largely as expected, another step in Intel’s quest to keep its products at least comparable to competitors as its prominence in the chip world has drastically declined in recent years.

Intel has seen its financial results plummet as AMD and ARM have increased their sales in the data center. 

In the September quarter of last year, AMD’s quarterly revenue from the data center, $3.5 billion, surpassed Intel’s, $3.3 billion, for the first time ever. 

As a result, according to industry research firm TechInsights, Intel’s share of the x86 server chip market, a decades-long duopoly between Intel and AMD, has gone from 96% in the first quarter of 2020 to 65% at present, while AMD’s share has ballooned from four percent to 35%.

ARM, which doesn’t make chips but sells designs to chip makers as intellectual property, has a much smaller base of revenue, but it is seeing notable growth from data center uses of its designs. The cloud giants, Google and the rest, are relying on ARM’s designs to make their custom chips such as Google’s TPU and Amazon’s Graviton processor rather than Intel’s x86 technology.

In the past 12 months, roughly 10% of ARM’s estimated $3.6 billion in revenue came from networking and data center applications, according to analysts. ARM’s projected revenue growth of 26% for the year is due in no small part to increasing sales into the data center, the company has indicated.

Also: Google just made AI coding assistance free for everyone – with very generous limits

Exact figures for ARM’s share in total data center processor sales are unavailable, but any royalties ARM gets as Google and others deploy more non-x86 chips is only upside for ARM and a loss for Intel (and AMD, to a lesser extent).

Intel makes the case that the Xeon parts, especially the new 6500 and 6700, can be more generally useful for a broad set of enterprise applications, thereby lowering a company’s TCO versus going with either AMD or ARM chips.

“This is really focused on the broad enterprise, and all of the different use cases there, with a specific focus then on AI and security and how do we provide our customers with something that improves their investment in their infrastructure,” said Singhal. 

The idea, said Singhal, is to “allow [enterprises] to consolidate old infrastructure into this new infrastructure that has better capabilities for new workloads and can reduce their power footprint.”

Also: If Intel can’t come up with a Qualcomm-killer soon, it’s game over for x86 PCs

However, Intel — whose CEO Pat Gelsinger departed late last year — is still pressed to deliver the most cutting-edge advances, according to TechInsights. 

In the media briefing, Singhal emphasized that the company’s successor to Granite Rapids, the code-named Clearwater Forest Xeon chip, is “on track” for production in the first half of next year. “We have it in the lab today” with customer tests, said Singhal.

However, that is a delay versus an expected debut of Clearwater Forest later this year, according to TechInsights.

Also, Intel’s dedicated AI chip, the Gaudi 3, failed to meet Intel’s own internal sales targets last year, according to TechInsights. That means “Intel has effectively exited the AI accelerator race, leaving AMD and NVIDIA as the only major players.”

A successor chip, Falcon Shores, has since been set aside, and Intel looks to a new product, Jaguar Shores, expected sometime in 2027.