JERUSALEM, May 29 (Reuters) – Nvidia (NVDA.O) said on Monday it was building Israel’s most powerful artificial intelligence (AI) supercomputer to meet surging customer demand for AI applications. bottom.
Nvidia, the world’s most valuable publicly traded chip company, said the cloud-based system will cost hundreds of millions of dollars and will be partially operational by the end of 2023.
Nvidia senior vice president Gilad Shainer said Nvidia works with 800 Israeli startups and tens of thousands of software engineers.
Called Israel-1, the system is expected to deliver up to 8 exaflops of AI computing performance, making it one of the fastest AI supercomputers in the world. One exaflops is capable of performing 1 quintillion or 1,000,000,000,000,000,000 calculations per second.
Shiner said AI is “the most important technology in our lifetime” and that developing AI and generative AI applications requires large graphics processing units (GPUs).
“Generative AI is ubiquitous today. It needs to be able to run training on large datasets,” he told Reuters, adding that the Israeli company has access to supercomputers it does not currently have. pointed out that it would be possible.
“This system is a large system, and it really allows us to run training faster, build frameworks, and build solutions that can address more complex problems.”
For example, OpenAI’s ChatGPT was built using thousands of Nvidia GPUs.
This system was developed by the former Mellanox team. Nvidia surpassed Intel in 2019 with its acquisition of Israeli chip designer Mellanox Technologies for nearly $7 billion.
Shiner said Nvidia’s first priority when it comes to supercomputers is its partner in Israel. “In the future, we may use this system to work with partners outside Israel,” he said.
Nvidia announced last week that it was working with the University of Bristol in the UK to build a new supercomputer using new Nvidia chips that will compete with Intel and Advanced Micro Devices Inc.
Reporting by Stephen Shear Editing by Mark Potter
Our standards: Thomson Reuters Trust Principles.
