#9 Supercomputer Unveiled: Meet Nvidia Eos

Recently, NVIDIA unveiled a video showcasing for the first time its latest data center-scale supercomputer, Eos, providing a glimpse into the architecture powering the advanced AI factories. Named after the Greek goddess who opens the gates of dawn each day, Eos symbolizes NVIDIA’s commitment to advancing artificial intelligence (AI) technologies.

NVIDIA announced in March 2022 that its Hopper architecture GPUs would be featured in Eos, with installations completed by the end of that year. In the high-performance LINPACK benchmark certification of November 2023, Eos ranked 9th in the global supercomputing Top500 list, boasting a total of 18.4 Exaflops of FP8 AI performance. Beyond handling AI workloads, Eos is also capable of conducting large language model training, recommendation systems, quantum simulations, and more.

Eos is described as a massive NVIDIA DGX SuperPOD, offering developers an accelerated computing infrastructure and fully optimized software for groundbreaking achievements in AI. It is equipped with 576 DGX H100 systems and the Quantum-2 InfiniBand networking platform, with each system containing 8 H100 compute cards for a total of 4608 cards, in addition to 1152 Intel Xeon Platinum 8480C processors (each with 56 cores). Furthermore, Eos incorporates sophisticated software products like NVIDIA Base Command and NVIDIA AI Enterprise.

Officials have stated that Eos’s architecture is optimized for AI workloads requiring ultra-low latency and high-throughput interconnect across large accelerated computing node clusters, making it an ideal solution for enterprises seeking to expand their AI capabilities.