IBM, Nvidia Team Up on Converged Storage for AI Workloads

IBM and Nvidia joined forces on a converged storage system optimized for artificial intelligence (AI) workloads. The new IBM Spectrum AI with Nvidia DGX is a converged system that combines IBM’s NVMe flash storage and Spectrum Scale storage software with Mellanox networking on Nvidia GPU-based servers. It also uses Nvidia’s AI software stack.The storage can scale from a single 300 terabyte (TB) node to 8 exabytes and a yottabyte or 1 trillion TB — of files. It also supports any tiered storage such as tape and cloud. “While AI started in academia, the national labs, and universities, it has now spread to enterprises,” said Eric Herzog, vice president of product marketing and management at IBM Storage Systems. At this year’s Supercomputing Conference, which IBM has been attending for years, the event organizers claimed the attendee mix was more than 40 percent enterprises, Herzog said. “In the old days it was all guys from big labs, government, and the universities.”By 2019, some 40 percent of digital transformation initiatives will use AI services, and by 2021, 75 percent of commercial enterprise apps will use AI, according to IDC.“And IBM has a lot of experience in this,” Herzog added, pointing to IBM supercomputers. IBM’s Summit and Sierra are the fastest computers in the world that are also purpose-built for AI workloads. Washington University, St. Louis School of Medicine, and Wells Fargo are also IBM customers using software-defined storage platforms for AI workloads. The partnership with Nvidia is IBM’s fourth AI-specific infrastructure project. In June, it released a reference architecture for AI. In August, it launched an accelerated compute platform. And in October, it released a storage system for autonomous driving.