VASP errors when running the simulation

[manoj_wijesingha]

Hi Colleagues,
I have two problems regarding the errors running the VASP simulations.
1. When running the VASP, this error codes shows "ib_mlx5_dv.c:435 UCX WARN mlx5dv_devx_obj_destroy(MKEY, ATOMIC) failed: Operation canceled
ucp_mm.c:355 UCX WARN failed to dereg from md[3]=mlx5_2: Input/output error"

2. For the frequency runs, this error happens "ML_FF is missing" on VASP 6.5.1 (this version has compiled successfully) before the IDM calculation. But when use the version of VASP 6.3.2, this works (this has only running errors).

I have attached the files you need to solve these problems, and could you please consider this request and provide the information.

Thank you,

Best Regards,
Manoj.

[manoj_wijesingha]

Regarding with the previous message for the error of frequency simulation "ML_FF missing", I could run successfully with VASP.6.3.2 version. But end of this simulation it shows some codes of "Program received signal SIGABRT" (attached the files_freq). This simulation is not working on VASP.6.5.1, showing "ERROR: ML_FF is missing, exiting...". Currently, I use the latest version of VASP for my all simulations. Could please consider the previous two problems and this matter as well.
Thank you.

Best Regards,
Manoj.

[henrique_miranda]

This error

Code: Select all

ib_mlx5_dv.c:435 UCX WARN mlx5dv_devx_obj_destroy(MKEY, ATOMIC) failed: Operation canceled
ucp_mm.c:355 UCX WARN failed to dereg from md[3]=mlx5_2: Input/output error

is probably related to your MPI installation. This is really hard for us to help you with because it depends a lot on your installation and local environment.
Does this error occur in all VASP jobs? Does it also occur when you run a single-node jobs? On a smaller system?
Does it occur in any of the tests in our testsuite?
If you are using an HPC facility this might be something to ask to your HPC sys admin

PS: I jut notices that you also attached some testsuite logs where this issue did not occur. Can you try the previous job (or one with a smaller number of atoms) on a single node and see if the error keeps appearing?

[henrique_miranda]

Regarding the second issue, note that ML_MODE is only available since VASP.6.4.0
So the calculation running in VASP.6.3.2 is different from the one that is trying to run with VASP.6.5.1.