To extract maximum performance from generative AI training and inference, we push both hardware and software to their limits. Low-precision training is one of the directions we work on—learn more about FP8 and efficient AI training in our NVIDIA Technical Blog post.