8-bit floating-point formats: E4M3 (4 exponent, 3 significand bits) and E5M2 (5 exponent, 2 significand bits), chosen to trade precision against range. Used to make AI training and inference faster and more power-efficient.
Field: Computer Arithmetic
It is mostly used with low precisions such as FP16, BF16, FP8 and INT8, greatly speeding up the matrix products that dominate neural network training and inference.
What is a characteristic of the tensor cores (matrix units)…Significand Tensor core INT8 Arithmetic shift bfloat16 Booth's algorithm Carry-lookahead adder Carry-save adder Catastrophic cancellation Exponent field