NDTW problem with different number of procs

[marchdf]

I am seeing a problem with the NDTW calculation using the latest nalu-wind.

Left is NDTW from a run with 4 procs, right is with 8 procs. About halfway up you can see NDTW drop to 0 for the run with 8 procs. It only drops at the corners of the domain. A slices through there shows that the rest of the field in that plane is fine.

I think it's happening at a MPI grid boundary. I increased the tolerance on NDTW and that didn't change anything. Does anyone have any ideas? Here's the input file: ndtw-test.txt. I can provide the mesh (120MB) if it helps.

Zoom:

MPI boundaries

[psakievich]

@marchdf do you see this at the corners of every processor boundary or just that one instance?

[marchdf]

Just at that processor boundary... All others are fine.

…

On Tue, Nov 15, 2022, 16:57 psakievich ***@***.***> wrote: @marchdf <https://github.com/marchdf> do you see this at the corners of every processor boundary or just that one instance? — Reply to this email directly, view it on GitHub <#1080 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADSXPT3VE44SZHY645DJO5DWIQPMTANCNFSM6AAAAAASBNO6I4> . You are receiving this because you were mentioned.Message ID: ***@***.***>