AI Glossary

GPU memory

Also called VRAM

What does “GPU memory” mean?

GPU memory: Memory attached directly to a GPU holds the model's weights and working data while it runs. If a model does not fit in that memory, it must be split across several GPUs, compressed, or run more slowly. Capacity is often what decides which models a given machine can run.

GPU memory visual guide: Memory attached directly to a GPU holds the model's weights and working data while it runs. If a model does not fit in that memory, it must be split across several GPUs, compressed, or run more slowly. Capacity is often what decides which models a given machine can run.
Visual guide Open full size (new tab)

What is an example of “GPU memory”?

Running a large open-weight model on a home computer usually depends on how many gigabytes of GPU memory the graphics card has.

Keep learning

See also

High-bandwidth memoryTermGPUTermQuantizationTermModel compressionTerm

Same topic

Hugging FaceTermFLOPsTermHyperscalerTermEdge AITermInference serverTerm