If you’ve ever wanted to take a peek inside a supercomputer, Nvidia has an absolute treat for you.
The AI tech behemoth, which recently overtook both Amazon and Google’s parent company Alphabet to become the third most valuable US company, has pulled back the curtain on Eos, its data-center-scale supercomputer – ranked one of the fastest in the world.
Eos was first unveiled at the Supercomputing 2023 trade show and is currently ranked ninth in the TOP500 list of the world’s fastest supercomputers.
Optimized for AI workloads
Named after the Greek goddess who opens the gates of dawn each day, Eos is an expansive Nvidia DGX SuperPOD, that serves as Nvidia’s development hub for AI breakthroughs, leveraging accelerated computing infrastructure and fully optimized software.
The Eos supercomputer is built with 576 Nvidia DGX H100 systems, Nvidia Quantum-2 InfiniBand networking, plus software, and is capable of delivering a whopping 18.4 exaflops of FP8 AI performance. Each DGX H100 system is equipped with eight Nvidia H100 Tensor Core GPUs, resulting in a total of 4,608 H100 GPUs.
As you’d expect from the world’s largest supplier of AI hardware, Eos’s architecture is optimized for AI workloads that demand ultra-low-latency and high-throughput interconnectivity across a large cluster of accelerated computing nodes.
Eos’s network architecture supports data transfer speeds of up to 400Gb/s, resulting in the rapid movement of large datasets essential for handling substantial AI workloads, including training LLMs, recommender systems, and quantum simulations.
To see what it’s like inside Nvidia’s Eos supercomputer, and to get an understanding of what it’s capable of, watch the short video below.
