Nvidia released a video on Thursday that offers the first public glimpse of the architecture of Eos, its latest enterprise-oriented supercomputer designed for cutting-edge AI development at data center scale, and the company’s fast The ultimate AI supercomputer.
The Eos machine, which is currently being used by Nvidia itself, is ranked as the world’s No. 9 highest-performing supercomputer in the latest Top 500 list, measured at FP64. has been done Among pure AI tasks, this is possibly the fastest. Meanwhile, its blueprint could be used by other companies to build enterprise-oriented supercomputers.
“Every day EOS rises to meet the challenges of thousands of Nvidia’s in-house developers doing AI research, and helps them solve previously unsolvable problems,” Nvidia said in the video. said
Nvidia’s Eos 576 is equipped with DGX H100 systems, each containing eight Nvidia H100 GPUs for artificial intelligence (AI) and high-performance computing (HPC) workloads. In total, the system packs 1,152 Intel Xeon Platinum 8480C (with 56 cores per CPU) processors as well as 4,608 H100 GPUs, giving the Eos an impressive Rmax of 121.4 FP64 PetaFLOPS as well as 18.4 FPAC and Exceive FPAC performance. It helps to do.
Eos (which relies on the DGX SuperPOD architecture) is designed for AI workloads as well as scalability, so it uses Nvidia’s Mellanox Quantum-2 InfiniBand in-network computing technology. In which the data transfer speed is up to 400 Gb/s. s, which is important for efficiently training large AI models as well as scaling.
In addition to powerful hardware, Nvidia’s Eos also comes with powerful software, again, designed for AI development and deployment, the company says. As a result, Nvidia’s Eos can address a wide variety of applications, from generative AI like ChatGPT to AI Factory.
“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.” Nvidia said in the video. “Eos — derived from earlier Nvidia DGX supercomputers such as the Saturn 5 and Selene. Made from knowledge. The latest example of Nvidia AI expertise in action. […] By building an AI factory like Eos, enterprises can work on their most demanding projects and realize their AI ambitions today and in the future.”
We don’t know how much the Eos will cost, and it doesn’t help that pricing for Nvidia’s DGX H100 systems is confidential and depends on many factors, such as volume. Meanwhile, considering the fact that each Nvidia H100 can cost $30,000 – $40,000 depending on volume, one can start to wonder how many numbers we get here.