The above chart graphically compares performance of the four Random Number Generators we discussed above. GPU substantially outperforms any of the CPU based implementations. However, so far the time to transfer the data from GPU memory to system memory has not been included in our measurements.
The above graph shows how much faster the GPU based Random Number Generation is. For example, the GPU method is nearly 250X faster than C++ latest standard libraries for generating floating-point numbers in 0.0 to 1.0 range, and nearly 150X faster for double precision floating-point.
The table below represents the values shown in the graph above, showing how many times the GPU is faster at generating random numbers than various methods on the CPU:
|C++ STD||Intel IPP||Intel MKL|