Intel Announces Falcon Project: Building Super Chip for AI Data Centers by 2025

2023-05-23

Intel said its next AI-centric design will be faster than the current market leader. In recent months, the popularity of artificial intelligence has exploded, especially since the introduction of chatbots by Microsoft and Google. However, these rapid advancements in AI are not coincidental; they require unimaginable computing power to train and run these models. Intel hopes to gain a larger market share with its upcoming Falcon Shores chip and has revealed the first details of its design. Intel discussed its high-performance computing plans at the ISC 2023 conference in Germany. The company has confirmed that Falcon Shores will be ready for the next generation of large language models (LLMs), including OpenAI's GPT and Google's PaLM. These chips will have 288GB of high-speed memory and 8-bit floating-point processing. In comparison, the popular Nvidia H100 Hopper card has "only" 80GB of RAM. In terms of machine learning hardware, Nvidia is clearly in the lead, but AMD is expected to pose a greater challenge with the MI300 later this year. This will be AMD's first design specifically for high-performance computing, utilizing TSMC's 5nm process, 3D stacked chips, and memory with ultra-wide bandwidth. In contrast, Intel has just begun to focus on developing hardware for machine learning applications. Intel states that it is preparing to complete the work on the Aurora supercomputer at Argonne National Lab, which is based on the long-delayed Ponte Vecchio processor. Intel claims that Ponte Vecchio is even faster than Nvidia's H100, but only the researchers at Argonne National Lab know for sure. Afterward, Intel will strive to prepare Falcon Shores. Intel's current timeline calls for Falcon Shores to start shipping in 2025. This gives Nvidia more time to establish its leading position, with AMD close behind. There may even be new Nvidia ML processors by then. When Intel is ready to compete in the machine learning data center, it may find that many major players have already established their computing partners. Intel hopes that the popularity of artificial intelligence will translate into a long-term business opportunity. This is not a new idea - former cryptocurrency mining organizations that purchased GPUs during the 2021-2022 metaverse boom are now trying to change their operational strategies to run ML workloads. This technology is now profitable, but we cannot predict if there will be a new tech frenzy in 2025.

RECENT AI NEWS

RECENT AI TOOLS