Intel has shared some new details about its upcoming chip designed for artificial intelligence (AI) computing, which is scheduled for release in 2025. The chip, codenamed “Falcon Shores,” will feature 288 gigabytes of memory and support 8-bit floating point computation. These technical specifications are significant as AI models, such as those used in services like ChatGPT, have grown in size, driving the need for more powerful chips to run them.
These details mark one of the early disclosures as Intel shifts its strategy to compete against leading AI chip manufacturers Nvidia and AMD. Intel has struggled to gain market share in the AI chip market, with its Ponte Vecchio chip, originally aimed to rival Nvidia, facing multiple delays.
Intel claims that its Ponte Vecchio-based Aurora supercomputer, which is almost fully shipped to Argonne National Lab, outperforms Nvidia’s latest AI chip, the H100. However, Intel’s next-generation chip, Falcon Shores, won’t reach the market until 2025, by which time Nvidia is likely to have introduced another chip of its own.
Jeff McVeigh, corporate vice president of Intel’s super compute group, explained that Intel has abandoned its previous strategy of integrating graphics processing units (GPUs) with central processing units (CPUs). Instead, they are now focusing on discrete offerings, allowing users to choose the best combination of CPUs and GPUs from different vendors at the platform level.
Intel’s aim is to have both the best CPU and the best GPU in the market, and by offering discrete options, they hope to provide greater flexibility and choice to customers.
