Nvidia has taken the wraps off Eos, its groundbreaking data center-scale supercomputer, at the Supercomputing 2023 trade show. Dubbed as an “AI factory,” Eos is designed to push the boundaries of artificial intelligence development. Named after the Greek goddess of dawn, this supercomputer symbolizes a new era in AI acceleration.

The supercomputer can achieve 18.4 exaflops of AI performance

Eos is powered by an impressive setup of 576 Nvidia DGX H100 systems, integrated with Quantum-2 InfiniBand networking and specialized software, achieving a staggering 18.4 exaflops of FP8 AI performance. This setup marks an evolution from Nvidia’s previous supercomputing projects, SaturnV and Selene, showcasing an advanced DGX SuperPOD architecture. This design enables rapid scaling of AI data center solutions to meet high-performance demands.

NVIDIA

At the heart of Eos are 4,608 H100 GPUs, spread across each DGX H100 system’s eight H100 Tensor Core APUs. This hardware configuration is tailored to manage extensive workloads, such as training large language models, running AI recommenders, conducting large-scale analytics, performing quantum simulations, and more.

Nvidia emphasizes that Eos’s architecture is finely tuned for AI tasks, requiring ultra-low latency and high throughput in massive computing clusters. The supercomputer’s networking capabilities, reaching speeds up to 400GB/s, are critical for handling large datasets necessary for training AI models.

Eos also integrates specialized software to enhance AI development and deployment. Base Command facilitates AI workflow, cluster management, and provides libraries for compute, storage, and network acceleration. AI Enterprise, a cloud-native platform, aims to speed up AI application development, positioning itself as the “operating system” for enterprise-level AI. In recognition of its capabilities, Eos secured the ninth position on the TOP500 list of the world’s fastest supercomputers.

RELATED:

(VIA)

RELATED ARTICLESMORE FROM AUTHOR