“Floating point precision standard, indicating efficiency in AI training”

“Floating Point 8, a precision used for quantizing Orpheus's KV cache to save GPU memory and improve concurrency.”

“If you go to FP8”

“Every time we went from FP32 to FP16 to FP8, we're effectively quadrupling the amount of computation or reducing the amount of energy by a factor of four.”

“Floating point precision used in Blackwell for faster computation.”