Intel’s CEO expresses the company’s ambition to secure the second spot in global AI system production by the end of the decade.
Intel revealed its latest AI hardware on Tuesday, boasting superior performance and efficiency compared to industry rivals, including Nvidia’s highly sort after chip.
Unveiled at the Intel Vision 2024 conference in Phoenix, the Gaudi 3 AI accelerator represents a significant advancement over its predecessor, the Gaudi 2. Intel claims that the Gaudi 3 offers enhanced AI computing power, networking bandwidth, and memory bandwidth, making it ideal for scaling up AI training and inference tasks, particularly for large language models (LLMs) like ChatGPT. Additionally, the accelerator is designed to optimize training and inference for multimodal AI models, capable of processing diverse data types such as images and audio.
With a focus on accelerating parallel AI operations, the Gaudi 3 accelerator is poised to deliver impressive performance gains. Intel asserts that it can outperform Nvidia’s $40,000 H100 chip by 50% in training top LLMs and provide 30% faster inferencing for these models compared to Nvidia’s H200 chip. Inferencing, which involves making predictions using trained data, is a crucial aspect of AI model deployment. While awaiting Nvidia’s performance results for its Blackwell chip, Intel remains confident in the superiority of the Gaudi 3.
Intel CEO Pat Gelsinger emphasized the company’s commitment to AI innovation, noting that every enterprise is rapidly embracing AI technology. He highlighted Intel’s efforts to democratize AI across various domains, from personal computing to data centers and edge computing environments.
Moreover, Intel aims to enhance cost efficiency for data centers with the Gaudi 3’s improved memory capacity. By requiring fewer accelerators to process datasets from larger AI models, Intel anticipates significant savings for data center operators.
Intel’s ambitions in the AI hardware sector are bolstered by substantial government funding and strategic investments. The company recently secured $8.5 billion in direct government funding as part of its broader $100 billion plan to expand chip manufacturing in the United States. Gelsinger outlined plans to establish a state-of-the-art AI chip manufacturing facility near Columbus, Ohio, as part of this initiative. Additionally, Intel is eligible for up to $11 billion in federal loans to support its expansion efforts.
During his keynote address at Intel Vision, Gelsinger reaffirmed the company’s goal to emerge as a prominent player in the global AI systems foundry landscape by the end of the decade.
Intel’s Gaudi 3 accelerator is poised to address the existing gap in AI hardware offerings, particularly amidst a shortage of Nvidia’s chips, according to Intel executives.
Pat Gelsinger, Intel’s CEO, highlighted the growing momentum among industry partners and customers for the Gaudi 3 accelerator, emphasizing its positive reception within the market. Notable customers already leveraging the Gaudi accelerator include Bosch, LandingAI, and Seekr.
The Gaudi 3 accelerator is slated for availability to equipment manufacturers such as Dell, HPE, Lenovo, and Supermicro in the second quarter of this year, with broader availability expected later. Gelsinger emphasized that Intel’s new products and services are designed to provide flexible solutions tailored to meet evolving customer needs and capitalize on emerging opportunities.
Regarding pricing and performance, Gelsinger noted that Intel’s roadmap of innovation includes scaling up the Gaudi 3 accelerator in collaboration with customers and partners. He also highlighted the upcoming release of Intel’s next-generation Falcon Shores supercomputing chip, positioning Intel to compete effectively with Nvidia’s product offerings.
Gelsinger emphasized the superior performance of the Gaudi 3 accelerator compared to Nvidia’s H100 chip, citing total cost of ownership (TCO) as a significant differentiator between the two chipmakers. While specific pricing details for the Gaudi 3 were not disclosed, Gelsinger expressed confidence that Intel would offer competitive pricing below that of Nvidia’s H100 and Blackwell chips.
Christoph Schell, Intel’s chief commercial officer, underscored Intel’s competitive edge over Nvidia, emphasizing factors such as cost, availability, and access to data. According to Schell, Intel excels in each of these areas, positioning the company ahead of Nvidia in the AI hardware market.
