Intel has unveiled details about its new artificial intelligence (AI) chip—the Gaudi 3—designed to enhance the performance of generative AI (GenAI) in enterprise settings.
The latest AI chip offers double the power efficiency and 1.5 times faster AI model processing than Nvidia's H100 GPU. Intel tested the chip on several models such as Meta's Llama and Falcon.
The Gaudi 3 delivers a fourfold enhancement in AI computation for BF16 and surpasses Nvidia's H100 in a head-to-head comparison, demonstrating an average 50% reduction in time required for training across different models. Additionally, the inference throughput of the system is expected to surpass that of the H100 by 50%.
The objective is to compete with Nvidia’s strong position in the AI chip industry and streamline the process of training and deploying artificial intelligence models, specifically those designed for language models.
The Intel Gaudi 3 is scheduled to be accessible to OEMs in the second quarter of 2024, including Dell Technologies, Hewlett Packard Enterprise, Lenovo, and Supermicro.
