Single-Precision Floating-Point Basic Arithmetic Throughput

Accelerated Computing CUDA CUDA Programming and Performance
1

hi all i was reading the programing guide chapter 5.

there is

Single-Precision Floating-Point Basic Arithmetic
Throughput of single-precision floating-point add, multiply, and multiply-add is 8
operations per clock cycle.

Why means it is “8”, why a constant.

can any body elaborate.

Thanks

–Biebo
NUST Pakistan

2

In the current generation, each multiprocessor has 8 ALU’s. So for every clock, 8 operations are one in parallel. addition, multiplication or madd, can all be done in 1 clock cycle on a ALU, so 8 of these per clock cycle.

3

thnx :)