My Community

Hello,

I am running VASP on a SGI Altix 3700 with Intel Itanium 2 Processors. I recently compiled a parallel VASP executable on this system. It runs my spin polarized Fe-Bench correctly for as many nodes I choose (I tested up to 6-nodes).

The problem is when I run a NEB simulation. The calculation hangs up on the 1st electronic step. The first image, no matter how many images I choose to use, never completes the 1st electronic step. The rest of the images complete the first electronic step. This strange behavior occurs no matter what transition state simulation I run. I have successfully ran NEB simulations on our groups other machines.

For the compilation of parallel VASP, I used the SGI makefile template for compiling libdmy.a in the libraries folder with the following changes to the makefile:

SUFFIX=.f90
# C-preprocessor
CPP = ecc -EP -C $*.F >$*$(SUFFIX)
FC=f90
CFLAGS = -O
FFLAGS = -FR -cm -w95 -lowercase

For the compilation of the parallel VASP execuatable I used the Altix makefile template with the following changes to the makefile:

FFLAGS = -FR -w -ftz -r8
OFLAG = -O2
LIB = -L../vasp.4.lib -ldmy \
../vasp.4.lib/linpack_double.o $(LAPACK) \
-lsdsm -lscs -lmpi
FFT3D = fftmpi.o fftmpi_map.o fft3dlib.o

Note that I am using the SGI Scientific Computing Software Library. Thanks for any help

modey3

Dear colleague, just to make sure: if possible, please check if -by bad luck- the first image always runs on the same node. If so, maybe this machine is defective, or some libraries are not installed properly on this particular node?
are there any hints for the reason of the behaviour in the job.stdout or parallelization.stdout files ?