Nvidia and Foxconn team up to create Taiwan’s largest supercomputer

Nvidia and Foxconn, two giants in the technology industry, have partnered to construct Taiwan’s largest supercomputer. Announced at Hon Hai Tech Day, the Hon Hai Kaohsiung Super Computing Center will be the cornerstone of Nvidia’s Blackwell GPU architecture, featuring the state-of-the-art GB200 NVL72 platform. This advanced setup will consist of 64 racks and a staggering 4,608 Tensor Core GPUs, collectively offering more than 90 exaflops of artificial intelligence (AI) performance, positioning it as the fastest supercomputer in Taiwan upon completion.

The supercomputer is being built in Kaohsiung, a city that has become a focal point for technological advancement. The project has commenced, with the first phase slated for operational status by mid-2025, and full deployment anticipated by 2026. This substantial infrastructure is designed to enhance Taiwan’s standing as a leader in AI-driven sectors by enabling breakthroughs in areas such as cancer research, large language models, and smart city advancements. Central to these developments is Foxconn’s “three-platform strategy,” which integrates smart manufacturing, smart cities, and electric vehicle industries.

Foxconn, formally known as Hon Hai Precision Industry Co., is renowned as the world’s top electronics manufacturer. The company produces a diverse range of high-tech products including smartphones and servers for leading global technology brands. With this new supercomputer, Foxconn aims to deliver significant advancements in digital twin technologies, robotic automation, and sophisticated urban infrastructures, thus facilitating enhanced AI-assisted services in metropolitan areas like Kaohsiung.

Nvidia’s Blackwell AI chips form the backbone of this initiative, marking a significant leap in computational power and efficiency. The GB200 NVL72 platform is optimised for AI and accelerated computing, integrating 36 Nvidia Grace CPUs and 72 Nvidia Blackwell GPUs per rack, interconnected via Nvidia’s NVLink technology. This configuration provides a staggering 130TB/s of bandwidth, making it ideal for tasks that demand substantial computational resources, such as training large AI models and executing real-time complex inference tasks on trillion-parameter models.

The innovative Nvidia NVLink Switch technology enables these 72-GPU systems to operate as a single, unified GPU. This advanced capability supports the training and processing of complex AI models, enhancing real-time decision-making processes and contributing to digital twin development. Foxconn itself has been pioneering in this arena, having integrated Nvidia Omniverse to create digital twins of its manufacturing facilities, thus optimizing its operations for smart manufacturing and smart city applications.

As the promise of cutting-edge AI technology continues to grow, this collaboration between Nvidia and Foxconn underscores a significant milestone in Taiwan’s technological advancement landscape. With plans to integrate Nvidia’s latest technologies such as the Omniverse and Isaac robotics platforms, the supercomputer is set to revolutionize manufacturing processes and support the development of domain-specific large language models. These models are progressively being embedded into Foxconn’s AI factories, aimed at driving innovation in smart manufacturing, smart electric vehicles, and smart cities, further cementing Taiwan’s reputation as a hub for technological innovation.

Source: Noah Wire Services