Glossary comparison
FP8 vs INT8: what is the difference?
The difference
INT8 holds evenly spaced integers and relies entirely on scale factors for range. FP8 has its own exponent, so its values are spaced more finely near zero and more widely at large magnitudes.
FP8 is also 8 bits but stores an exponent, so it covers a wider range with uneven spacing. INT8 values are evenly spaced and depend entirely on the scale factor for range.
The two terms
FP8
An 8-bit float defined by OCP in two variants, E4M3 favouring precision and E5M2 favouring range, used with per-tensor scaling or, in the OCP MX formats, a shared scale per block of 32 values, for training and inference on recent accelerators.
INT8
An 8-bit signed integer format used with per-tensor or per-channel scale factors for neural network inference. It is the default target of most NPUs and the precision behind many quoted TOPS figures.
Side by side
| Field | FP8 | INT8 |
|---|---|---|
| Stands for | 8-bit floating point | 8-bit integer |
| Type | Term | Term |
| Domain | Compute & AI Hardware | Compute & AI Hardware |
| Group | Numerics and precision | Numerics and precision |