Intel made waves on Tuesday with the introduction of its newest artificial intelligence chip, Gaudi 3, positioning itself at the forefront of the rapidly evolving AI landscape. This announcement comes amidst a flurry of activity in the semiconductor industry as chipmakers scramble to develop cutting-edge solutions capable of training and deploying large AI models, such as those powering OpenAI's ChatGPT.
The Gaudi 3 chip boasts impressive performance metrics, claiming to be over twice as power-efficient as Nvidia's H100 GPU while also running AI models 1.5 times faster. Available in multiple configurations, ranging from a bundle of eight chips on a single motherboard to standalone cards compatible with existing systems, Intel's latest offering promises versatility and scalability to meet diverse AI application needs.
During testing, Intel evaluated Gaudi 3's capabilities on various models, including Meta's open-source Llama and the Falcon project backed by Abu Dhabi. Results indicate that Gaudi 3 excels in both training and deploying AI models, demonstrating its potential across a spectrum of tasks, from Stable Diffusion to OpenAI's Whisper model for speech recognition.
As Intel positions itself to compete with Nvidia, which currently commands an estimated 80% share of the AI chip market with its GPUs, the company aims for Gaudi 3 to be available to customers in the third quarter. Strategic partnerships with industry stalwarts like Dell, Hewlett Packard Enterprise, and Supermicro ensure widespread adoption, although specific pricing details are yet to be disclosed.
In the rapidly expanding data center AI market, there is a growing demand for alternative chip suppliers to mitigate costs associated with running AI workloads. While Nvidia's stock has experienced exponential growth, Intel and AMD are intensifying their efforts to capture market share. AMD recently introduced the MI300X data center GPU, while Nvidia unveiled successors to the H100, the B100, and B200 GPUs.
Intel's approach to software integration sets it apart from Nvidia, with a focus on collaboration with industry leaders like Google, Qualcomm, and Arm to develop open software solutions. This strategy aims to provide customers with flexibility and choice, offering alternatives to Nvidia's proprietary CUDA software suite.
Moreover, Intel's decision to manufacture Gaudi 3 using a five-nanometer process suggests a reliance on external foundries for chip production. Looking ahead, Intel plans to expand its AI chip manufacturing capabilities, potentially catering to external companies, as CEO Patrick Gelsinger hinted at plans for a new Ohio factory expected to commence operations in 2027 or 2028.