Be a part of our each day and weekly newsletters for the newest updates and unique content material on industry-leading AI protection. Be taught Extra
Nvidia and Foxconn are constructing Taiwan’s largest supercomputer utilizing Nvidia Blackwell chips.
The challenge, Hon Hai Kaohsiung Tremendous Computing Middle, revealed Tuesday at Hon Hai Tech Day, can be constructed round Nvidia’s Blackwell graphics processing unit (GPU) structure and have the GB200 NVL72 platform, which features a whole of 64 racks and 4,608 Tensor Core GPUs.
With an anticipated efficiency of over 90 exaflops of AI efficiency, the machine would simply be thought-about the quickest in Taiwan.
Foxconn plans to make use of the supercomputer, as soon as operational, to energy breakthroughs in most cancers analysis, massive language mannequin growth and good metropolis improvements, positioning Taiwan as a world chief in AI-driven industries.
Foxconn’s “three-platform technique” focuses on good manufacturing, good cities and electrical autos. The brand new supercomputer will play a pivotal position in supporting Foxconn’s ongoing efforts in digital twins, robotic automation and good city infrastructure, bringing AI-assisted companies to city areas like Kaohsiung.
Development has began on the brand new supercomputer housed in Kaohsiung, Taiwan. The primary section is predicted to be operational by mid-2025. Full deployment is focused for 2026.
The challenge will combine with Nvidia applied sciences, resembling Nvidia Omniverse and Isaac robotics platforms for AI and digital twins applied sciences to assist rework manufacturing processes.
“Powered by Nvidia’s Blackwell platform, Foxconn’s new AI supercomputer is likely one of the strongest on the planet, representing a big leap ahead in AI computing and effectivity,” mentioned Foxconn vice chairman James Wu, in an announcement.
The GB200 NVL72 is a state-of-the-art information middle platform optimized for AI and accelerated computing.
Every rack options 36 Nvidia Grace CPUs and 72 Nvidia Blackwell GPUs related by way of Nvidia’s NVLink know-how, delivering 130TB/s of bandwidth.
Nvidia NVLink Swap permits the 72-GPU system to operate as a single, unified GPU. This makes it best for coaching massive AI fashions and executing complicated inference duties in actual time on trillion-parameter fashions.
Taiwan-based Foxconn, formally often known as Hon Hai Precision Trade Co., is the world’s largest electronics producer, identified for producing a variety of merchandise, from smartphones to servers, for the world’s high know-how manufacturers. Foxconn is constructing digital twins of its factories utilizing Nvidia Omniverse, and Foxconn was additionally one of many first firms to make use of Nvidia NIM microservices in
the event of domain-specific massive language fashions, or LLMs, embedded into quite a lot of inner techniques and processes in its AI factories for good manufacturing, good electrical autos and good cities.
