Intel introduces Gaudi 3 AI accelerators manufactured by TSMC, escalating competition in data center accelerator chips.
Intel introduced the Gaudi 3, an AI accelerator developed with TSMC's 5nm process node, positioning it as an alternative to Nvidia's dominant market position. According to Intel's press release, the Gaudi 3 delivers an average of 50% better inference performance and 40% better power efficiency than the Nvidia H100, while maintaining significantly lower costs. The chip promises a 4x increase in AI computing for BF16 and a 1.5x boost in memory bandwidth compared to Gaudi 2, substantially enhancing AI training and inference capabilities for enterprises implementing generative AI at scale.
Designed with support from Alchip, the Gaudi 3 maintains the same architecture as Gaudi 2 but advances to TSMC's 5nm process from the 7nm process used in its predecessor. Intel CEO Pat Gelsinger stated that "the world needs more suppliers, and Intel is dedicated to providing the choice." While Gelsinger did not disclose specific pricing, he indicated that the company would provide an "extremely good" total cost of ownership. The Gaudi 3 will be available to OEMs including Dell, Hewlett Packard Enterprise, Lenovo, and Supermicro in the second quarter of 2024, with wider availability in the third quarter.
Intel has announced a significant lineup of customers and partners for its Gaudi accelerator, including Bharti Airtel, Bosch, CtrlS, IBM, IFF, Landing AI, Ola, Naver, NielsenIQ, Roboflow, and Seekr. The company plans to collaborate with industry leaders such as SAP, RedHat, and VMware to establish an open platform for enterprise AI, aiming to accelerate deployment of secure generative AI systems powered by retrieval-augmented generation (RAG) technology. Jeni Barovian, Intel's vice president for strategy and product management, stated that "customers are coming to Intel expecting Intel to deliver solutions that meet their needs."
Intel's Gaudi 3 launch represents a strategic challenge to Nvidia's dominant position, which controls over 80% of the AI chip market. However, Intel faces mounting competitive pressures as major cloud service providers including Amazon, Microsoft, and Google develop their own in-house data center chips. According to Intel's corporate filings, its Data Center and AI division has reported declining sales for two consecutive years, while its competitors Nvidia and AMD have experienced surging sales in the data center market.