Hello,
I am trying to run OpenIFS 48r1 on LUMI EuroHPC using initial data with TCo319 truncation. We have been successfully running the model with the linear reduced grids, but the cubic oct grids somehow produce errors. I have the correct climate data files for this resolution (319_4) and I believe I reserved enough memory for the model. I am using all initial files as well as the fort.4 provided by the Data Hub (ERA5 reforecast). I attached the error log as a screenshot. When I launch the job, I do not get the +00000000 grib files and the NODE.001_01 file is empty. Any tips what I could possibly do?
Thanks,
Sandor
Hi Sandor,
Thanks for posting here. Unfortunately I haven’t got much information to go by.
But something has caught my eye: The last line of your screenshot shows the directory OIFS_master in the path. Can I just check whether you have built OpenIFS from the 48r1 release branch or from the main branch?
This is relevant because at the current date the main branch on the GitHub repo has meanwhile progressed further and should not be worked from, always work from release branches.
If this isn’t the problem then I will need more information to help you. Could you please attach the entire stdout/stderr output and also your fort.4 to this thread? I presume that you have checked that in exp-config.h the variable OIFS_GRIDTYPE is set to “o” for the cubic octahedral grid?
Best wishes,
Marcus
As I understand, the problem has now been solved, it was related to specific FFT namelist settings that are required for the Cray compiler on the LUMI HPC system.
Best wishes,
Marcus
