hard20 pts · first answer countsgpu-computingperformance

Arithmetic intensity of a big matmul

Multiplying two N × N FP16 matrices takes 2N³ FLOPs and moves at least 3N² × 2 bytes (two inputs, one output). For N = 4,096, what is the arithmetic intensity?

A.About 4 FLOPs per byteB.About 1,365 FLOPs per byteC.About 136 FLOPs per byteD.About 13,650 FLOPs per byte

Discussion 0

Join free to post in the discussion.

Arithmetic intensity of a big matmul

Discussion 0

Related questions