NVIDIA Unveils Video Showcasing Architecture of Latest Enterprise Supercomputer Eos NVIDIA has released a video showcasing the architecture of its latest enterprise-grade supercomputer, Eos. This supercomputer is designed for advanced artificial intelligence (AI) development at a data center scale and is considered the company's fastest AI supercomputer. Currently, NVIDIA's self-used Eos machine ranks 9th in the latest Top 500 supercomputer performance ranking based on FP64 measurement. It is likely one of the fastest in pure AI tasks. Additionally, the blueprint of Eos can be used to build enterprise-grade supercomputers for other companies. In the video, NVIDIA states that Eos faces challenges every day and assists thousands of internal developers engaged in AI research to solve previously unsolvable problems. Eos is equipped with 576 sets of DGX H100 systems, each containing eight NVIDIA H100 GPUs for AI and high-performance computing (HPC) workloads. Overall, the system integrates 1152 Intel Xeon Platinum 8480C processors (56 cores per CPU) and 4608 H100 GPUs, achieving impressive HPC and AI performance of Rmax 121.4 FP64 PetaFLOPS and 18.4 FP8 ExaFLOPS. The design of Eos relies on the DGX SuperPOD architecture, built specifically for AI workloads and scalability. It incorporates NVIDIA's Mellanox Quantum-2 InfiniBand and its in-network computing technology, enabling data transfer speeds of up to 400 Gb/s, which is crucial for efficient training of large AI models and scalability. In addition to powerful hardware, Eos is equipped with software specifically built for AI development and deployment. Therefore, it can handle various applications, from generative AI like ChatGPT to AI factories. NVIDIA states in the video that Eos has an integrated software stack, including AI development and deployment software such as orchestration and cluster management, accelerated computing storage and networking libraries, and an operating system optimized for AI workloads. Eos is the latest testament to NVIDIA's expertise in AI, and by creating such an AI factory, enterprises can take on their most challenging projects and achieve their AI visions for today and the future. The cost of Eos is currently unclear, and the pricing of NVIDIA's DGX H100 systems is confidential and depends on factors such as quantity. Considering the cost of each NVIDIA H100 may range from $30,000 to $40,000, we can speculate that the total cost of Eos will be very high.