NVIDIA has unveiled its Eos data-centre-scale supercomputer, which has clinched the No. 9 spot in the TOP500 list of the world’s fastest supercomputers.
Named after the Greek goddess believed to open the gates of dawn each day, Eos is a colossal NVIDIA DGX SuperPOD that showcases the cutting-edge architecture empowering advanced AI factories. It consists of 576 NVIDIA DGX H100 systems and NVIDIA Quantum-2 InfiniBand networking and software, boasting an impressive 18.4 exaflops of FP8 AI performance.
Eos signifies NVIDIA’s commitment to pushing the boundaries of AI technology. It is equipped with 4,608 H100 GPUs, each DGX H100 system featuring eight NVIDIA H100 Tensor Core GPUs.
Eos is tailor-made for handling the most demanding AI workloads, including training large language models, recommender systems, and quantum simulations.
The timing of Eos’ arrival is opportune, as the world witnesses transformative applications of generative AI in fields ranging from drug discovery to chatbots and autonomous machines. To achieve such breakthroughs, developers require more than just expertise — they need an AI factory that can handle large-scale AI model building effectively and consistently.
Eos provides a purpose-built AI engine that is always available and capable of scaling AI capabilities. It incorporates NVIDIA’s advanced accelerated computing and networking technologies, complemented by sophisticated software offerings such as NVIDIA Base Command and NVIDIA AI Enterprise.
The architecture of Eos is optimised for AI workloads, ensuring ultra-low-latency and high-throughput interconnectivity across a vast cluster of accelerated computing nodes. This makes Eos an ideal solution for enterprises seeking to scale their AI capabilities.
At its heart lies the DGX SuperPOD architecture, driven by NVIDIA’s DGX H100 systems. This architecture is designed to provide tightly integrated full-stack systems capable of computing at an enormous scale.
Based on NVIDIA Quantum-2 InfiniBand with In-Network Computing technology, Eos supports data transfer speeds of up to 400Gb/s, facilitating the rapid movement of large datasets crucial for training complex AI models.
Entelechy Asia 