Glossary · Term

A100

← all terms

Definition

Plain language

A high-end NVIDIA data-center chip used to train and run large AI models, a generation before the H100.

As stated in the literature

NVIDIA's Ampere-architecture data-center GPU, widely used for large-scale training and serving prior to the Hopper-generation H100; eight of them form a common single-node training cluster.

Why it matters: It represents the kind of expensive, specialized hardware that determines whether a team can afford to train big AI models at all.

For example, a research lab might wire together eight A100 chips in a single machine to train a large language model.

Heard on the show

“Reproducing all three applications takes about forty-six GPU-hours on a single A100.”
Episode 173 — The Free Step-Level Grader Hiding in Every RL Training Run

Mentioned in 3 episodes

  1. 173
    The Free Step-Level Grader Hiding in Every RL Training Run
  2. 172
    One Bad Token Can Sink a Model's Math, And You Can Delete It
  3. 154
    How a 7B Model Out-Investigates a 72B One by Choosing What to Look At

Related terms