Hi, I want to check my runtime about 100,000 * 100,000 size matrix in different threads. Anyone can help me compare mine with yours? Scheduling mode: static Probability: 0.05 jwan*3@s*t*n*x*03:/scratch/courses0101/jwang3/project1> cat slurm-15659059.out Testing with 1 threads: -------------------------- Program Execution time is 516.236606 seconds -------------------------- Testing with 4 threads: -------------------------- Program Execution time is 531.101039 seconds -------------------------- Testing with 8 threads: -------------------------- Program Execution time is 528.720953 seconds -------------------------- Testing with 16 threads: -------------------------- Program Execution time is 518.605728 seconds -------------------------- Testing with 32 threads: -------------------------- Program Execution time is 527.469895 seconds -------------------------- Testing with 64 threads: -------------------------- Program Execution time is 516.329616 seconds -------------------------- Testing with 128 threads: -------------------------- Program Execution time is 529.702909 seconds -------------------------- Thanks, Joey

Your run-times seem significantly faster than mine are -> mine are closer to 7500 seconds for 0.05 and 128 threads. As for run time measurements, are you using clock() to measure the time? I found when using that it seemed to actually accumulate total run time across all cores, so was not a good measure of overall time. I have been using the double omp_get_wtime() function built into openMP instead, and that gave me values closer to what I expected

I mentioned in another thread that a multiplication runtime of around 600 seconds is quite reasonable. BTW, I noticed that you are using `sbatch` which is supposed to be 10 times slower than `srun`. I have posted about this before, but there is no official response. [help5507]⬈ According to my recent test, this problem seems to have been fixed. Good news for everyone!

Maybe because I only calculate the running time for matrices multiplication.