Nvidia taps into Israeli innovation to build generative AI cloud supercomputer

Chip giant says Israel-1 supercomputer valued at several hundred million dollars is a ‘major investment’ that will boost next-generation AI workloads

Sharon Wrobel is a tech reporter for The Times of Israel.

Nvidia's HGX supercomputing platform. (Courtesy)
Nvidia's HGX supercomputing platform. (Courtesy)

US gaming and computer graphics giant Nvidia said Monday that it will build the nation’s most powerful generative AI cloud supercomputer called Israel-1 which will be based on a new locally developed high-performance ethernet platform.

Valued at several hundred million dollars, Israel-1, which Nvidia said would be one of the world’s fastest AI supercomputers, is expected to start early production by the end of 2023.

“AI is the most important technology force in our lifetime,” said Gilad Shainer, Senior Vice President of high performance computing (HPC) and networking at Nvidia. “Israel-1 represents a major investment that will help us drive innovation in Israel and globally.”

AI processes analyze enormous datasets and require both ultra-fast computing performance and massive memory. The rise of generative AI applications and workloads like OpenAI’s ChatGPT present new challenges for networks inside data centers. As a result of the major changes AI cloud systems need to be trained using huge amounts of data.

Announced at the Computex tech exhibition starting this week, Israel-1 will be based on Nvidia’s newly launched Spectrum-X networking platform, a high-performance ethernet architecture purposely built for generative AI workloads. Developed in Israel the platform is tailored to enable data center around the world transition to AI and accelerated computing, using a new class of ethernet connection that is build from the ground up for AI.

“Transformative technologies such as generative AI are forcing every enterprise to push the boundaries of data center performance in pursuit of competitive advantage,” said Shainer. “NVIDIA Spectrum-X is a new class of Ethernet networking that removes barriers for next-generation AI workloads that have the potential to transform entire industries.”

Nvidia’s Spectrum-4 networking platform. (Courtesy)

Nvidia said that Israel-1 will serve as its ethernet-AI cloud reference platform for data centers around the world and will be used for internal R&D, as well as a testbed for joint development with partners.

Israel-1 will run at a performance of eight exaflops and will be one of the world’s fastest AI supercomputers, Nvidia said. An exaflop measures the ability to perform 1 quintillion – or 1,000,000,000,000,000,000 – calculations per second. Israel-1 is also expected to perform peak performance of more than 130 petaflops – the ability to carry out 100 trillion operations per second – for traditional scientific computing workloads. It will be equipped with data processing units (DPUs) called BlueField-3, which were developed in Israel.

“The delivery of end-to-end AI capabilities reduces run-times of massive transformer-based generative AI models, and allows network engineers, AI data scientists and cloud service providers to obtain accurate results and make informed decisions faster,” Nvidia said in a statement.

In 2020, Nvidia bought Israel’s Mellanox Technologies Ltd., a maker of high-speed servers and storage switching solutions used in supercomputers globally, for a massive $7 billion, adding about 1,000 employees to its Israel operations.

Nvidia’s R&D activities in Israel are already the firm’s largest outside of the US. The firm employs more than 3,000 workers, or 12% of its global workforce in seven R&D centers, from Yokne’am, the HQ of Mellanox, to Tel Aviv, Jerusalem, Ra’anana, and Beersheba in the south.

Alongside its R&D operations, Nvidia also runs the Nvidia Inception Program for Startups, an accelerator that works with hundreds of early-stage companies, including 800 Israeli startups, and the Nvidia Developer Program, which allows free access to Nvidia’s offerings for developers.

Most Popular
read more: