NVIDIA demonstrates its Eos supercomputer powered by 4,608 H100 GPUs, optimized for generative AI.

Nvidia on Thursday published a video offering the first public glimpse into the architecture of Eos, its newest enterprise-oriented supercomputer designed for advanced AI development at the datacenter scale and the company's fastest AI supercomputer. The Eos machine, currently being used by Nvidia itself, is ranked as the world's No. 9 highest performing supercomputer in the latest Top 500 list, which is measured in FP64; in pure AI tasks, it is likely among the fastest. The blueprint of Eos can be used to build enterprise-oriented supercomputers for other companies as well.

Eos is equipped with 576 DGX H100 systems, each containing eight Nvidia H100 GPUs for artificial intelligence and high-performance computing workloads. In total, the system packs 1,152 Intel Xeon Platinum 8480C processors, each with 56 cores, as well as 4,608 H100 GPUs. This configuration enables Eos to achieve an impressive Rmax of 121.4 FP64 PetaFLOPS as well as 18.4 FP8 ExaFLOPS performance for HPC and AI workloads respectively.

The design of Eos relies on the DGX SuperPOD architecture and is purpose-built for AI workloads and scalability. The system uses Nvidia's Mellanox Quantum-2 InfiniBand with In-Network Computing technology, which features data transfer speeds of up to 400 Gb/s—crucial for training large AI models effectively and scaling out. According to Nvidia, "Eos has an integrated software stack that includes AI development and deployment software, [including] orchestration and cluster management, accelerated compute storage and network libraries, and an operating system optimized for AI workloads."

Built from the knowledge gained with prior Nvidia DGX supercomputers such as Saturn 5 and Selene, Eos represents the latest example of Nvidia AI expertise. As Nvidia stated in the video, "Every day EOS rises to meet the challenges of thousands of Nvidia's in-house developers doing AI research, helping them solve the previously unsolvable." The company further noted that "by creating an AI factory like Eos, enterprises can take on their most demanding projects and achieve their AI aspirations today and into the future." The system can address a variety of applications, ranging from ChatGPT-like generative AI to AI factories.

The exact cost of Eos remains undisclosed, as pricing for Nvidia's DGX H100 systems is confidential and dependent on factors such as volumes. However, with individual Nvidia H100 GPUs costing between $30,000 and $40,000 depending on volume, the total system cost is likely substantial.