sm:leading-[6px] sm:text-sm”>
You may like
-
At Nvidia’s GTC event, Pat Gelsinger reiterated that Jensen ‘got lucky with AI,’ Intel missed the boat with Larrabee
-
Intel’s AI PC chips aren’t selling well — instead, old Raptor Lake chips boom
Under the new leadership, Intel’s AI unit will focus on workload-specific solutions for AI, which probably includes a diverse lineup of AI products, starting from humble low-power solutions for inference on the edge to rack-scale solutions for inference and training in the data center space. Intel’s management sees the company’s silicon photonics capabilities as one of its advantages in the AI data center realms. At the same time, Intel acknowledges that it needs to work hard on its software. Ultimately, the company plans to offer full-stack solutions integrating hardware (including CPUs, GPUs, IPUs [which Intel calls DPUs], storage, silicon photonics, etc.) and software.
Despite canceling the codenamed Falcon Shores GPU (which used to be a hybrid processor with x86 and Xe cores before becoming just a data center GPU), Intel stressed that Jaguar Shores remains on the AI roadmap. At the same time, Intel is actively exploring partnerships for custom-built silicon and is evaluating new architecture options. In addition to high-performance x86 processors, IPUs, and decades of experience, one of the advantages that Intel has for custom silicon is that it already has experience with silicon photonics, leading-edge processors, multi-chiplet solutions, and advanced packaging.
“We also see optics as a critical element of that rack scale architecture as well,” said Michelle Johnston Holthaus, Intel’s products chief. “I think I would just remind everybody that Intel is the only foundry out there that has an optics-based foundry option for our customers. So we are very optimistic that those two things [IPUs and silicon photonics] add to the overall opportunity to build out that rack- scale architecture. Also, our open x86. What we are seeing is that customers do love the x86 ecosystem, the software around it. If they can build out an AI infrastructure with x86, they are very interested in doing that. We already have one large custom design win, and I would expect one [more].”
Follow Tom’s Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.