Tensor Core
Tensor Core: An NVIDIA hardware unit specialised in reduced-precision matrix operations used for accelerated training and inference. It uses formats such as FP16, BF16 or INT8 depending on generation. Benefits require compatible shapes and kernels and are not automatic for every program.
In italiano: Tensor CoreLegge IV · Loss, gradiente e ottimizzatoriIt uses formats such as FP16, BF16 or INT8 depending on generation. Benefits require compatible shapes and kernels and are not automatic for every program.