Gaudi2 is Habana Labs‘ second-generation CPU, which Intel purchased for $2 billion in late 2019 from an Israeli AI semiconductor firm. As AI computing is one of the fastest-growing areas for data centers, AI chip startups have received significant funding in recent years.
Many AI researchers and companies are committed to using NVIDIA‘s software platform CUDA, thus taking market share away from them has been difficult. Intel noted that, in addition to new chips for AI computing, it has been focusing on software development.
“CUDA is not a moat that NVIDIA can really stand on for long,” said the chief business officer Eitan Medina at Habana Labs, adding that Intel’s software platform is an open standard, free to download, and use from GitHub, the software development site. “Now the question is who can do the work efficiently?”
The Gaudi2 was twice as fast as Habana‘s previous AI chip, according to Medina, and was built using Taiwan Semiconductor Manufacturing Co’s 7-nanometer transistor technology, as compared to 16-nanometers previously. The chip will be faster and more powerful if the transistor size is smaller.
Intel also unveiled Greco, a new chip for inferencing operations, which involves using an AI algorithm to make a prediction or identify an object.
Sandra Rivera, who heads Intel’s data center and AI, said that the AI chip industry will increase at a rate of 25% per year for the next five years, reaching roughly $50 billion. “We intend to invest and innovate to lead this market,” she added, adding that more investment will be put into software, both to increase Intel’s staff and to acquire other companies.