NVIDIA has formally launched its supercomputer, Eos, specifically designed for artificial intelligence (AI) applications. Eos boasts 576 NVIDIA DGX H100 systems, with each system consisting of 8 H100 accelerators, adding up to a total of 4608 units. The systems are all connected through NVIDIA’s Quantum-2 InfiniBand architecture, operating at 400 Gbps.
The NVIDIA Eos supercomputer offers AI application performance of 18.4 exaflops (FP8). The company initially unveiled it at the Supercomputing 2023 exhibition in November of last year and it is now ready to be operational. According to the developer, Eos is capable of handling the most significant AI workloads, including the training of massive language models, recommendation algorithms, quantum simulations, and more. It ranks ninth in the TOP500 supercomputer list.
Eos’s high performance is facilitated by NVIDIA’s robust accelerators and proprietary interfaces, along with its software, such as NVIDIA Base Command and NVIDIA AI Enterprise. “The Eos architecture is optimized for AI workloads that require ultra-low latency and high-throughput connections within the accelerator cluster, making it the perfect solution for enterprises looking to scale their AI infrastructures. As corporations and developers globally seek to leverage AI capabilities, Eos serves as a key resource promising to expedite the journey to AI-based applications for every organization,” NVIDIA stated.
