TensorFloat-32

This article needs additional citations for verification. Please help improve this article by adding citations to reliable sources. Unsourced material may be challenged and removed. Find sources: "TensorFloat-32" – news · newspapers · books · scholar · JSTOR (April 2025) (Learn how and when to remove this message)

TensorFloat-32 (TF32) is a numeric floating point format designed for Tensor Core running on certain Nvidia GPUs. It was first implemented in the Ampere architecture ^[1]. TensorFloat-32 combines the 8-bit exponent size of IEEE single precision with the 10-bit mantissa size of half precision for a total of 19 bits per number. It is comparable to the bfloat16 format, which uses a 7-bit mantissa.

The binary format is:

• 1 sign bit
• 8 exponent bits
• 10 significand bits (also called mantissa, or precision bits)

The 19-significant-bit format fits within a double word (32 bits), and while it lacks precision compared with a normal 32-bit IEEE 754 floating-point number, it provides much faster computation, up to 8 times on a A100 (compared to a V100 using FP32).^[2]

Stored in the same space as FP32, it is not a distinct storage format, but a specification for reduced-precision FP32 multiply–accumulate operations. FP32 inputs are rounded to TF32, multiplied to produce a 21-bit product (including the implicit msbit, this is an 11×11→22-bit multiply), and summed into a standard FP32 accumulator.^[3]

• IEEE 754