While acquisition rumors swirled, Intel launched two products that are vital to the success of Pat Gelsinger’s turnaround strategy. But are they competitive?
While the press is full of reports of Qualcomm’s potential acquisition of Intel, which we doubt will occur, and speculation swirls about other possibilities, the company launched two new products that could be game changers. That is if Intel can get customers to buy enough of them. The new Performance-core-equipped Xeon 6P is a worthy challenger to AMD’s current line of EPYC CPUs, and the Gaudi 3 is a viable alternative to Nvidia H100 and the AMD MI300 GPUs for large-scale AI. So, what are the gotchas?
Intel Xeon 6P
Xeon already launched the Xeon 6E back in June with Efficiency cores, similar to the AMD Bergamo, which is also designed for web applications and more cores. With Granite Rapids Intel has finally launched the Xeon 6P with Performance-cores, doubling the performance for AI and HPC workloads. If you want more details, I recommend you take a look at Timothy Prickett Morgan’s piece here:
An area I would point out is the AI performance on the Xeon 6P, which really sings. Since Xeon 5, Intel Xeon processors have been increasingly capable of handling generative AI workloads, particularly for inference tasks and smaller to medium-sized models (under 20 GB). If you are looking to integrate AI into your existing infrastructure, as opposed to building an AI factory, the 6P offers a cost-effective solution without requiring specialized hardware (e.g., GPUs or Gaudi).
The new Xeon 6 AI performance, which Intel claims is over 5 times faster than the current AMD EPYC CPU, is due to further exploitation of the matrix processors on the Xeon. Now, realize that AMD could copy this approach on their Turin launch next week, so be sure to watch what they announce. If they do finally add AI to EPYC, I’d expect AMD to be at least in the same ball park as Intel. If they do not, then Intel has an important differentiator the market will care a great deal about.
Gaudi 3 is Finally Here
I’m not saying Gaudi 3 is later than planned, but I am saying its about a year too late to make a huge difference to Intel, especially since it will be replaced in about a year with the next-gen Falcon Shores GPU. Intel has said they expect up to $500M in 1st year revenue for the Gaudi 3, and had already announced that IBM would use the Gaudi 3 in their Cloud services. But AMD has already announced they would sell 10 times that amount, or $5B. And THAT is still a distant second compared to Nvidia.
Intel realizes it can’t compete in generative AI with just chips; it needs a full stack approach with rack level designs and software. Consequently, Intel has released specification for server designs from a 1-node to 1024-node clusters, using Arista networking and systems from Dell Technologies, to Lenovo and Supermicro for server designs.
Since there is not much new info that hasn’t already been pre-disclosed about the Gaudi 3 AI accelerator, which I have covered a few times, Intel chose to focus the launch on their embrace of the Open Platform for Enterprise AI, or OPEA, an open software platform that eases enterprise adoption of AI, and RAG, or retrieval augmented generation. I don’t see much here that Nvidia hasn’t already delivered in spades, and then some, to help Enterprise adoption of AI, but it is a solid start to be sure. And OPEI’s openness and broad community support will help this ecosystem grow.
Conclusions
Both Xeon 6 and Gaudi 3 are solid steps forward for Intel that improve their competitive position, but neither fundamentally changes the data center calculus, in my humble opinion. Xeon 6 should stem the market share erosion somewhat to AMD, who should still grow CPU market share albeit perhaps more slowly. And Gaudi 3 is an excellent platform for AI, especially for inference processing of LLMs. But it still puts Intel into 3rd place, in what will likely be a two horse race.