NVIDIA Unveils DGX H100 Systems to Power Next-Generation Enterprise AI
NVIDIA introduces its fourth-generation DGX H100 systems featuring powerful new H100 GPUs that deliver massive leaps in AI performance. The new hardware forms the foundation for exaflop-scale supercomputers designed to train massive language models.
NVIDIA announces the fourth-generation DGX H100 system as the world’s first AI platform built with the new NVIDIA H100 Tensor Core GPUs. Each system packs eight H100 GPUs connected by NVIDIA NVLink to deliver 32 petaflops of AI performance at new FP8 precision, providing six times more power than the previous generation.
These DGX H100 systems serve as the foundational building blocks for the next-generation NVIDIA DGX POD and DGX SuperPOD AI infrastructure platforms. By utilizing a new NVLink Switch System to connect up to 32 nodes and 256 H100 GPUs, a single DGX SuperPOD delivers one exaflop of FP8 AI performance to handle massive language models with trillions of parameters.
NVIDIA also reveals "Eos," a new supercomputer built with this architecture that features 576 DGX H100 systems and 4,608 H100 GPUs. Expected to be the world’s fastest AI system upon launch later this year, Eos provides 18.4 exaflops of AI computing performance and acts as a blueprint for enterprise customers to build their own scalable AI infrastructure.