My Community

Hi VASP Community,
I am running a hybrid functional calculation for a cell with 160 atoms. Currently I am in the process of optimizing the performance and ran some tests to estimate memory usage and performance with different KPAR and NCORE tags.
The hardware specification that I used is
#Nodes = 2
#Cores/node = 32
#Memory = 125GB/node

The estimated memory usage per node in the OUTCAR file is = 0.8GB (mpi-rank0)
This leads to the total memory usage of around 51.2GB for all the 64 cores. But when I monitor the code performance, I see the each of the nodes has used 51.2GB rather than total memory usage of 51.2GB.
Am I missing something?

Any help in this regard will be greatly appreciated.

Please post your calculation with all the necessary files (INCAR, KPOINTS, POSCAR, POTCAR).

Hi,
Attached are the input files for the calculation that I used for benchmarking.
Please let me know if any more information is required.

Thank You,
Ankit

I tried to reproduce your problem, but I can't. The memory assumption always agrees with the actually used amount of memory.
I tried intel/impi and gnu/ompi compilers VASP 6.3.0 and also VASP 6.1.2 that you used. I also tried with and without shared memory and with 2 to 8 nodes with 8 processors each node.

Which compilers/toolchain did you use?
Is the memory consumption constant or does it ramp up over time? It's important to know because some not to old compiler versions for both open mpi and intel mpi show memory leaks.

You used VASP 6.1.2, but please try again with VASP 6.3.0.

Hi,
Update:
The VASP memory issue is solved now.

As a follow up question, I am using the fake SC approach to do hybrid calculations, but I need to obtain the wavefunctions at the kpoints which have 0 weights.
Is there a way to obtain these wavefunctions?

Thank You,
Ankit

Dear Ankit,

Using the fake SC approach and writing the WAVECAR, gives you the wavefunction at each k-point listed in KPOINTS, including those with weight 0.

Dear Ankit,

I am facing the same memory problem. Could you please share how you solved the memory issue?

Divanshu