I have a question about the results according to tile number

[kobl1201]

Hi everyone~

I have some question about the result

I worked two experiments which are the same condition except tile number.

For example, at the Exp 1, i used NtileI == 8, Ntilej == 10 and at the Exp 2 i set Ntile == 10, Ntilej == 16.

And then when i compared two results, I found the different results which showed different circulation patterns.

The results are simulated, even though the conditions are same.

Is that true or I havea mistake??

Please tell me the your opinion.

Thank you.

[kate]

You are supposed to get the same answer no matter the tiling, but Kaitlin Alexander has found slight variations when using large numbers of cores due to how MPI reduce operations work in ROMS.

[kobl1201]

Thank you Kate~

Do you know the paper of Kaitlin Alexander or other people about the this problem??

[kate]

Something like this? Not that I've read it: http://cacs.usc.edu/papers/Small-OrderI ... PDPS16.pdf