Single-precision 32-bit floating point, the long-standing default for general-purpose computing and the baseline most GPU specifications quote. Largely superseded by lower precisions for AI, where FP32 throughput is now a small fraction of a chip's tensor performance.
TensorFloat-32 - A 19-bit precision format introduced with NVIDIA Ampere archite...
16-bit floating-point precision format (half precision). Commonly used in AI tra...
Double-precision 64-bit floating point, required for scientific and engineering ...
Trillion Floating-Point Operations Per Second - A measure of computational perfo...
Browse the full glossary or dive into architecture guides
