I compiled VASP 5.3 using INTEL MPI and MKL. The resulting executable will run just fine when using less than 20 cores for mpirun. But as soon as I specify more than 20 (say 24 with 4 nodes in my .hosts file) I obtain the warning
"Sub-Space-Matrix is not hermitian"
and the job will never complete.
I DID search the forums and found some older posts on this but none of the suggested solutions did the trick for me .
My lib-versions are consistent AFAICS:
Code: Select all
$?ldd?vasp
linux-vdso.so.1?=>??(0x00007fff343f8000)
libdl.so.2?=>?/lib/libdl.so.2?(0x00007f6b6ed1c000)
libmkl_intel_lp64.so?=>?/opt/intel/composer_xe_2013.2.146/mkl/lib/intel64/libmkl_intel_lp64.so?(0x00007f6b6e5f5000)
libmkl_sequential.so?=>?/opt/intel/composer_xe_2013.2.146/mkl/lib/intel64/libmkl_sequential.so?(0x00007f6b6df4f000)
libmkl_core.so?=>?/opt/intel/composer_xe_2013.2.146/mkl/lib/intel64/libmkl_core.so?(0x00007f6b6ccec000)
libmkl_scalapack_lp64.so?=>?/opt/intel/composer_xe_2013.2.146/mkl/lib/intel64/libmkl_scalapack_lp64.so?(0x00007f6b6c43d000)
libmkl_blacs_intelmpi_lp64.so?=>?/opt/intel/composer_xe_2013.2.146/mkl/lib/intel64/libmkl_blacs_intelmpi_lp64.so?(0x00007f6b6c202000)
libmpigf.so.4?=>?/opt/intel/impi/4.1.0.030/intel64/lib/libmpigf.so.4?(0x00007f6b6bfd2000)
libmpi.so.4?=>?/opt/intel/impi/4.1.0.030/intel64/lib/libmpi.so.4?(0x00007f6b6b9a6000)
librt.so.1?=>?/lib/librt.so.1?(0x00007f6b6b79e000)
libpthread.so.0?=>?/lib/libpthread.so.0?(0x00007f6b6b582000)
libm.so.6?=>?/lib/libm.so.6?(0x00007f6b6b2ff000)
libc.so.6?=>?/lib/libc.so.6?(0x00007f6b6af9d000)
libgcc_s.so.1?=>?/lib/libgcc_s.so.1?(0x00007f6b6ad87000)
/lib64/ld-linux-x86-64.so.2?(0x00007f6b6ef34000)
Thanks for any help!