All terms
Hardware & Systems
NVIDIA H100
NVIDIA's Hopper-architecture data-center GPU, widely used to train and serve large models.
Definition
The NVIDIA H100 is a data-center GPU built on the Hopper architecture and used heavily for training and serving large models. It pairs many compute cores with high-bandwidth memory and fast interconnects, and added hardware support for FP8 to speed up large-model workloads. Newer Blackwell-generation chips have begun to succeed it in fresh deployments, though large fleets of H100s remain in active use.