My Community

Hello,

I wonder if there is anyone in the community who could please advise us on the installation of Vasp 5.3.2 on a computational cluster. I have a user who would like to do hybrid DFT calculations using vasp, however our vasp crashes with a segmentation fault just as the simulation gets started. I have installed vasp 5.3.2 on a RHELS 5.2 cluster and I'm using the Intel compilers v12.0 Update 4, MKL v10.3 Update 4 and OpenMPI v1.4.5. MKL provides the BLAS, LAPACK and Scalapack libraries.I've tried using the standalone scalapack, fftw libraries and vasp LAPACK libraries to no avail. I note that we have set the stack size to unlimited on each compute node.

Actually, to be more precise. Our vasp does appear to work on one compute node. So using 8 processor cores on one node does work, however multiple node simlations always fail.

If you have vasp installed and working on your cluster then I wondered if you could please advise us -- either by sharing your makefile or by advising us on the options that we should us. We're quite new to vasp and don't really know/understand how and when to use options like -DNGZhalf, for example.

Any advice that you could give us would be apprecaited, please.

Best regards -- David.

Dear David,

I have the same problem with VASP 5.3.2, it runs without problems in one computing node, but it crashes when using more than one computing node.

Since I didn't realize about your post, I made a post in this forum about this issue (http://cms.mpi.univie.ac.at/vasp-forum/ ... hp?3.12037).

You can find there the Makefile I am using to compile VASP and other details of my cluster. In short, I am also using Intel Compilers and Intel MKL.

I can tell you that using ScaLAPACK I observed problems related to memory overloaping in the MPI calls, so I didn't use it.

I hope that you find something useful in my post.

Regards,

Iv?n

Dear David,

I found a solution to the problem. Check this link:
http://cms.mpi.univie.ac.at/vasp-forum/ ... hp?3.12037

Best regards,

Iv