If a program consists of 10 dynamic instructions, and each C…
If a program consists of 10 dynamic instructions, and each CUDA Block has 128 threads, and the width of a warp is 32 threads, how many times does that SM fetch an instruction? Assume there is only one CUDA block for this question.
Read DetailsLet’s assume that an SM has an execution width of 16 threads…
Let’s assume that an SM has an execution width of 16 threads and can accommodate 128 threads. Each instruction involves at most 2 read and 1 write register operations. How many minimum register read ports are needed to execute each instruction in one cycle, given that a register has 4 banks and 1 register read takes 1 cycle?
Read Details