Intel abandons Xe-HPC GPU accelerator in favor of Gaudi, but still Gaudi 3 AI accelerator misses sales target. So Intel is pinning its artificial intelligence hopes on its next-generation Falcon Shores GPU platform. However, interim co-chief executive Michelle Johnston Holthaus said this week that instead of falcon coast As a first-generation breakthrough, it requires rapid iteration to become a competitive platform.
“We really need to think about how to move from Gaudi to our first-generation Falcon shores, which are GPUs,” Barclays said at the 22nd annual Global Technology Conference. “I’m telling you right now, is this going to be great? No, but it’s a good first step in completing the platform, learning from it, understanding how all the software is going to work, how the ecosystem is going to work in response so that we can be very Iterate quickly.
Intel admitted that its Gaudi 3 platform will not meet its 2024 sales goals, mainly due to imperfect software. At the Barclays conference, the company expanded on the situation and said that the Gaudi platform is complex to deploy, especially in large clusters used for training. That’s why it’s mostly used for edge reasoning.
“Gaudi doesn’t allow me to reach the public; it’s not a GPU that can be easily deployed in systems around the world,” the interim co-CEO said. “When you think about the enterprises deploying Gaudi, you’ll find everything from the largest hyperscalers to smaller customers deploying at the edge.”
Intel’s Gaudi platform is also beneficial because it allows the company to learn more about the platform and software design. While what’s learned from business hardware can be applied to next-generation AI platforms, it remains to be seen how the lessons learned from Gaudi will be applied to an entirely different next-generation platform.
Intel’s falcon coast Considered a multi-chip design with Xe-HPC (or at least something like Xe-HPC) and x86 chiplets with unified HBM memory. The company said in 2022 that integrating x86 CPUs and Xe-HPC GPUs into a single module with a unified memory architecture will enable Intel to achieve 5 times higher computing density and memory capacity than the February 2022 platform , bandwidth and performance per watt.
Considering the refined architecture and process technology the Falcon shores will feature, it’s reasonable to expect the device to be significantly faster than the company’s 2022 offerings (Xeon scalable “Ice Lake” processors and first-generation Gaudi accelerators) .
Logically, Falcon Shores will serve as a learning tool for Intel and its ISV partners. The company’s data center GPU Max “Ponte Vecchio” has yet to gain significant traction in the AI space, so ISVs will have to learn how to use Intel’s Xe-HPC (or rather Xe-AI?) architecture on Falcon Shores. Because it takes a long time to develop artificial intelligence software, Intel’s next-generation jaguar coast Possibly the first platform promising mass adoption.