Dear Mohamed,

In general, theoretical results exist for optimization algorithms concerning the rate of convergence (linear, superlinear, quadratic etc...). Nevertheless, from a practical point of view, the CPU time will depend on the problem you are solving and on the implementation of the algorithm. What is exactly the problem you want tro solve ?

You can evaluate the performance of an optimization algorithm in different ways. For example,

1. you can look for the rate of convergence (as mentioned by Mr. Richard)

2. you may try to find out the level of minima or maxima achieved (depends on your problem type)

3.  you may look for the average CPU time needed for convergence (for multiple runs)

4. you may look into the statistical significance of the algorithm by running different statistical tests.

For more info on the statistical test results you may look into the following:

"A practical tutorial on the use of non-parametric statistical tests as a methodology

for comparing evolutionary and swarm intelligence algorithms" - by  Joaquín Derrac et al. 

Dear Mohamad,

You can compare the event log of CPU and Memory before and after in windows environment.

 Thanks for all, the problem was addressed is image segemntaion, the proposed is evaluated based on PSNR, FM, NRM, and so on.

I need to compute GPU, how