Glossary · Term

H100

← all terms

Definition

Plain language

A high-end NVIDIA data-center GPU used to train and serve large AI models.

As stated in the literature

NVIDIA's Hopper-architecture data-center GPU, widely used in frontier-model training and serving stacks.

Why it matters: It's the hardware nearly all modern frontier training is done on, and its availability and price largely determine who can build state-of-the-art models.

For example, training a 70B model from scratch typically requires a cluster of thousands of H100s running for weeks.

Heard on the show

“And the only place it dips to ninety-two is the hardest case — two same-generation NVIDIA chips, an H100 versus an H200, which are numerically very close cousins.”
Episode 158 — How Floating-Point Rounding Lets a Model Tell Which Chip It's On — And Misbehave

Mentioned in 4 episodes

  1. 158
    How Floating-Point Rounding Lets a Model Tell Which Chip It's On — And Misbehave
  2. 152
    Training a Model to Mean What It Says, And Why That Isn't the Same as Being Good
  3. 088
    Two Levers for Self-Improving AI: When Rewriting Code Isn't Enough
  4. 027
    When AI Agents Build the Serving Stack: A Bet on Bespoke Infrastructure