float16-two-bytes-per-parameter-one-b-approx-2gb
IN premise — summaries/2026/08/24/wiki-Large_language_model-chunk-1.md
Created 2026-08-24T17:11:15+00:00
Float16 uses 2 bytes per parameter, so a 1B-parameter model occupies approximately 2 GB of memory in float16 precision.
Summary
This is the basic math behind fitting a model into hardware: a 1-billion-parameter model stored in 16-bit floating point takes up roughly 2 gigabytes just for its weights. That number is the starting point for any capacity planning, since it tells you whether a model will fit in a single GPU or requires distributed memory across machines.