AI Glossary

FP8

Also called 8-bit floating point

What does “FP8” mean?

FP8: A family of eight-bit floating-point formats used in some accelerator workloads.

FP8 visual guide: A family of eight-bit floating-point formats used in some accelerator workloads.
Visual guide Open full size (new tab)

What is an example of “FP8”?

A serving system uses FP8 weights or activations where accuracy remains acceptable.

Keep learning

See also

GPUTermComputeTermQuantizationTermMixed-precision trainingTermBfloat16Term

Same topic

FLOPsTermGPU memoryTermExpert parallelismTermHierarchical navigable small worldTermEuclidean distanceTerm