The NVIDIA H100 is a data centre GPU, built on NVIDIA's Hopper architecture, designed for large-scale AI and high-performance computing.
In plain words
It was introduced in 2022 as the successor to the A100. Its Tensor Cores added FP8, a smaller number format that speeds up AI maths, and a Transformer Engine for transformer language models. It was also the first GPU with HBM3 memory, giving over 3 TB per second of memory bandwidth. Many GPUs can be linked with NVLink to work as one large system.
A quick example
Imagine a lab training a language model. Instead of one computer, it rents a cluster of servers each holding several H100s, linked so the model's work is split across all of them at once.
Where you'll meet it
- Training and serving large language models in data centres.
- Cloud GPU rentals for AI research.
Related
Sources
- DocsH100 GPU, NVIDIA · read 29 Sept 2026
- DocsNVIDIA Hopper Architecture In-Depth, NVIDIA Technical Blog · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.