Intel announced that the Gaudi 3 boasts over twice the energy efficiency of Nvidia's H100 GPU chip, while running AI models at a speed 1.5 times faster than the H100 GPU. Furthermore, the Gaudi 3 offers various configuration options, such as the ability to integrate eight Gaudi 3 chips on a single board.
Intel also conducted chip performance tests on open-source artificial intelligence models developed by Meta, such as Llama. The company stated that Gaudi 3 can facilitate various models, including artificial intelligence image generation tools like Stable Diffusion and speech recognition models like OpenAI Whisper, for both training and deployment purposes.
Intel emphasized that compared to Nvidia's chips, Gaudi 3 boasts lower power consumption.
It is estimated that Nvidia currently holds an 80% share of the artificial intelligence chip market with its GPU chips. Over the past year, Nvidia's GPUs have remained the top choice for high-end chips among artificial intelligence developers.
Intel plans to ship Gaudi 3 to customers in the third quarter of this year. Companies like Dell, HP, and Supermicro will utilize this chip to build systems. Intel has not disclosed the pricing for Gaudi 3, but reports suggest it will be significantly lower than the H100.
During a conference call, Das Kamhout, Vice President of Software for Intel Xeon, stated, "We do expect it to be more competitive compared to Nvidia's latest chips." He added, "Given our competitive pricing, unique open-integrated chip fabric, and industry-standard Ethernet adoption, we are confident this will be a powerful product."
As cloud service providers and enterprises gradually build infrastructure for deploying artificial intelligence software, the artificial intelligence market in the data center sector is expected to continue growing. This means that even though Nvidia continues to dominate the majority of the artificial intelligence chip market, other competitors also have room to grow.
Furthermore, the high cost of running generative artificial intelligence and purchasing Nvidia GPU chips has led many companies to seek other vendors to reduce costs.
The boom in artificial intelligence has seen Nvidia's stock price more than double over the past year, while Intel's stock has only risen by 18% during the same period.
AMD is also seeking to expand its business by selling more artificial intelligence chips for servers. Last year, AMD launched a new data center GPU named MI300X, with Meta and Microsoft being clients for this product.
Earlier this year, Nvidia launched the B100 and B200 GPU chips. As successors to the H100, Nvidia promises improved performance with these chips, expected to begin shipping later this year.
Nvidia's success is also attributed to a powerful proprietary software suite called CUDA, allowing artificial intelligence scientists to access all hardware functionalities within the GPU. Intel is collaborating with other chip manufacturers and software giants like Google, Qualcomm, and ARM to develop non-proprietary open software, aiming to provide more convenience for software companies to switch chip suppliers.
During the conference call, Sachin Katti, Senior Vice President of Intel's Network Group, stated, "We are collaborating with the software ecosystem to develop open-source software and modules, allowing users to integrate the solutions they need without having to repurchase them."
Intel stated that Gaudi 3 adopts the 5-nanometer process technology, indicating that Intel is outsourcing chip production. CEO Patrick Gelsinger revealed last month that in addition to designing the Gaudi 3 chip, Intel plans to produce artificial intelligence chips at a new factory in Ohio expected to start operations in 2027 or 2028.
