internal error in: vhdf5.F

[jonathan_lahnsteiner2]

Dear md_ashraf_ali,

Regarding your first picture. The error output tells you that there are no projectors available.
It also tells you to rerun vasp by setting LORBIT>= 10 in your INCAR file.
For more information on LORBIT, please check the vasp wiki
https://www.vasp.at/wiki/index.php/LORBIT
So you try to read something from the vaspout.h5 file which was not computed in your vasp calculation.

Regarding your second picture. The vaspout.h5 file is written in binary format. This file format is not meant to be human-readable
or opened in a text editor. If you want to check the contents of the vaspout.h5 file you can use py4vasp or some other tools for dealing with h5 format.
For example, with h5ls you can display the objects in the root group. Or the h5dump tool dumps or displays the contents of an HDF5 file (textually).
For more information on these tools please check https://docs.hdfgroup.org/hdf5/develop/ ... _view.html

All the best Jonathan

[md_ashraf_ali]

sir, when I run my test code with mpirun -np 2 it takes few times to execute but whenever I try it with mpirun -np 4 it is taking a lot of time. Why is this happening when I want to use all of my 4 cores?

[jonathan_lahnsteiner2]

Dear md_ashraf_ali,

When I look at the graph you sent I can see that all cores on this node are running with 100% utilization.
If you are only using 4 cores then there is another job running on this compute node. So it could also very well
be that you share the cores you are using with another job. I would try again running your job with 2 and 4 cores again and
while verifying there is no other job running on the same computation node.

Another hint, when using a parallel version of vasp you should
try to optimize the NPAR tag to get maximal performance.
https://www.vasp.at/wiki/index.php/NPAR


All the best

Jonathan

[md_ashraf_ali]

There is no other job running on the same node.Only the code was running in vasp.

[jonathan_lahnsteiner2]

Dear md_ashraf_ali,

You said you were running vasp with mpirun -np 4. This means you use 4 processors. But in the
graphic above you sent, 36 processors are running. So either you submitted the job to 36 cores with mpirun -np 36
or there is another job running on this node. If there was another job running on this node it would be best to try
again when no other jobs are running.
Please try this. You can also send the input files for your job and I will try and execute them on 2 or 4 cores. Maybe
you did some unlucky settings.
Also if you want to determine timings, you should not use a single value but average over multiple runs. You could
even specially compile vasp to obtain detailed timings in the OUTCAR file. Therefore you can visit
https://www.vasp.at/wiki/index.php/Profiling

Best Jonathan

[md_ashraf_ali]

Please try and execute them on 2 or 4 cores.I cannot find out the problem.As far I execute, neither there is another job running in the node nor I run the code with mpirun -np 36.
Is there any problem related to the attached part? Should I add "export OMP_NUM_THREADS=4 or 36" in my /.bashrc ?

[jonathan_lahnsteiner2]

Dear md_ashraf_ali,

I executed you jobs on 2 and 4 cores.
With 2 cores I get the following timings (LOOP+ time from OUTCAR file):

Code: Select all

4.0196
1.5477
1.3246
0.8803

For the execution with 4 cores I get

Code: Select all

2.6864
0.9527
0.8122
0.5476

So the execution on 4 cores is faster. I was using your job files without any modifications.
I recognized the line

Code: Select all

running     4-mpi ranks with 36 threads per rank, on 1 nodes

in your job submission. So somehow your command

Code: Select all

export OMP_NUM_THREADS=1

is overwritten or ignored. This is the reason why your job becomes slower.
You could try to recompile vasp without openmp support and only use openmpi.
You can do this by removing the following tags from your makefile.include:

Code: Select all

-D_OPENMP
-fopenmp
-lfftw3_omp

and then recompile vasp. Or you could use the makefile.include.gnu for compilation


All the best Jonathan

[md_ashraf_ali]

I have found the tags in my makefile.include at line 11,16,17 and 71. Should I remove all of these lines?

[jonathan_lahnsteiner2]

Dear Dear md_ashraf_ali,

Sorry for my late reply.
No, you should not remove the entire lines where the flags occur, but only the flags themselves.
If there is only one specified flag in the line, then you can remove the entire line.

All the best Jonathan