Draft page

This page is an outline. It describes what will be covered and is not yet complete technical documentation.

Running out of accelerator memory in production is one of the more common and more avoidable inference failures. A memory budget computed in advance turns it into a capacity decision rather than an incident.

This page sets out the components of that budget and how each one scales.

What you will learn

  • The components that make up accelerator memory consumption.
  • How to estimate weight memory from parameter count and precision.
  • How KV cache size scales with batch size and context length.
  • How much headroom to reserve for fragmentation and runtime overhead.

This page is an outline. The subsections below are the planned structure; they are filled in as the handbook is written.

The components of a memory budget

Draft

Model weights

Draft

KV cache

Draft

Activations and workspace

Draft

Framework and runtime overhead

Draft

Worked estimate

Draft

Validating the estimate against reality

Draft