Intel doubles down on Enterprise AI with Xeon 6 P-cores and Gaudi 3

Intel has launched two new products designed to accelerate AI adoption in enterprises: the Intel Xeon 6 with P-cores and Gaudi 3 AI accelerators.

“Demand for AI is leading to a massive transformation in the data center, and the industry is asking for choice in hardware, software and developer tools,” said Intel’s General Manager of the Data Center and Artificial Intelligence Group, Hotard.

“With our launch of Xeon 6 with P-cores and Gaudi 3 AI accelerators, Intel is enabling an open ecosystem that allows our customers to implement all of their workloads with greater performance, efficiency and security.”

So that’s the promise. Here’s a little more detail.

Intel Xeon 6 with P-cores

Following the June 2024 release of Xeon 6 with E-cores, the 6700-series, Intel has announced the Xeon 6 with P-cores. Identified as the 6900-series, codenamed Granite Rapids, the new chips are designed for scalable performance.

Intel claims up to 5.5x higher AI inferencing performance versus AMD EPYC offerings, and up to 2.1x higher HPC performance. It also promises big increases compared to 5th Gen Intel Xeon processors, as per the slide below.

“Our performance cores are uniquely architected to deliver significant performance leads in critical growth spaces like AI and HPC, and databases, while also delivering lower power consumption so our customers can scale without impacting their power constraints,” said Ryan Tabrah, VP and GM of Intel Xeon, in a pre-briefing session.

“So as they build out new data centres or existing data centres, they can advance their compute.”

Ultimately, Intel reckons this new generation will reduce the cost of cloud computing, with 1.6x higher performance per watt than 5th Gen Xeons and 30% lower average total cost of ownership.

With up to 128 cores per CPU, increased memory bandwidth and LLC (last level cache), plus a bunch of new extensions focused on AI, Intel has declared the Intel Xeon 6 with P-cores as “the world’s best CPU for AI inference”.

However, we expect AMD to announce its latest generation of EPYC processors next month. It will no doubt have a similar set of graphs and claims to show that the new EPYC is, in fact, is fastest.

What matters to customers – whether data centres, enterprises or the likes of IBM and Google – is that the new generation of AI chips are faster and more efficient. That the improvements each year are so significant. And that both AMD and Intel are pursuing an open approach.

Intel Gaudi 3 AI accelerators

It feels as if Intel has been talking about Gaudi 3 AI accelerators for years. We’ve already covered Intel’s performance claims earlier this year and the fact that they will be available on IBM Cloud in early 2025.

So, what’s new? In short, there is now general availability for Gaudi 3, with both Dell and Supermicro announcing products. Both the Dell PowerEdge XE9680 and Supermicro X14 are due to ship in October, Intel claims.

Here’s Intel’s Gaudi 3 promise: to do battle against Nvidia’s H100 and H200 AI accelerators, and AMD’s Instinct MI300 series, to deliver large-scale generative AI in the cloud. And to do so on an open platform.

“The Intel Gaudi 3 accelerator was optimised for LLMs and designed to scale up and scale out efficiently,” said Justin Hotard, adding that enterprises could scale “from a single node to thousands of nodes”.

Intel is also emphasising its open platform, as opposed to Nvidia’s proprietary approach. It claims this makes it simpler to develop AI applications and choose the platform of your choice. In particular, enterprises should be able to easily switch between AMD and Intel platforms, with no extra coding required.

What Intel isn’t promising is equivalent performance to Nvidia’s H200 accelerator. Notably, its slides compare Gaudi 3 to the H100, which remains popular due to its lower cost and greater availability. Even then, its biggest wins come in terms of performance per dollar rather than outright performance.

Trending Articles