[lostking18]

Hi, there

I am comparing the calculation speed between two steady state simulations by reading the .out file. I have the following questions:
1) What is the relationship between the wall clock second and CPU second? I know the wall clock second is the ‘real-world’ time consumption. But if I know the CPU second and I am using one node with 16 cores of CPU, can I calculate the wall clock seconds from the CPU second?
2) From the .out file I only know roughly the accumulated CPU second at the current iteration from the value in the parenthesis, how can we get the CPU seconds with higher precision?
3) Is there any way we can get the time consumption in each iteration? I tried to calculate the difference of the CPU second between two sequential iterations. But the precision of the CPU second is too low.

Thank you in advance

[Opaque]

What are you trying to achieve out of this calculation?

There is only enough information to determine the cost per iteration for the model. The precise value of the iteration cost is (to me) irrelevant, the average value is more meaningful if you are assessing performance.

Take the wall clock time (CFD time) before the solution starts, and when it ends, divide by the number of iterations and that is your average to estimate future runs of the model.

There is no explicit relationship between wall clock time and cpu time since it depends on the machine load at a given time. It may take 1 CPU second for a task and 20 min wall clock to be completed if the machine is heavily loaded.

[Dinar]

Wall clock time = CPU seconds / partitions

I began to notice on some tasks, the wall clock time doesn't match with the "job finished - job started" time.
Considering that the solver starts quickly.
It's a mystery to me