I noticed that both versions give approximately the same energy but off by an order of 1E-3:
vasp 5.2: E0 = -.44152118E+03
vasp 5.3: E0 = -.44152538E+03
In the OUTCAR, 5.3 finds constrained configuration C_1 while 5.2 finds group C_1h.
Is this a bug?
I see that using -Doldsym in the makefile will keep the old behavior for k-point generation but I'm insure if the new behavior is better.
I found the "benchmark" on this site: http://docs.notur.no/uit/misc/benchmarks2011/vasp
I'm also using 32 cores for the calculation. 5.2 took 381 minutes while 5.3 took 1231 minutes.
My makefiles are essentially the same for both versions: I'm using gcc 4.5.
Code: Select all
FC=mpif90?-ffree-form?-ffree-line-length-none?-fopenmp?
FCL=$(FC)?
CPP_?=?./preprocess?<$*.F?|?/usr/bin/cpp?-P?-C?-traditional?>$*$(SUFFIX)?
FFLAGS?=?
OFLAG?=?-O2
OFLAG_HIGH?=?$(OFLAG)?
OBJ_HIGH?=?
OBJ_NOOPT?=?
DEBUG?=?-FR?-O0?
INLINE?=?$(OFLAG)?
BLAS=?-L$(ACMLMP_HOME)/lib?-lacml_mp
LAPACK=?../vasp.5.lib/lapack_double.o?
SCA=-L$(SCA_HOME)/lib?-lscalapack?$(SCA_HOME)/lib/blacsC.a?$(SCA_HOME)/lib/blacsF77.a?$(SCA_HOME)/lib/blacs.a
LIB?=?-L../vasp.5.lib?-ldmy?../vasp.5.lib/linpack_double.o?$(LAPACK)?$(BLAS)?$(SCA)
LINK?=?-L$(FFTW_HOME)/lib?-lfftw3_threads?-lfftw3
CPP?=?$(CPP_)?-DMPI?-DHOST=\"GFortran\"?\
-Dkind8?-DRPROMU_DGEMV?-DRPROMU_DGEMV?-DLONGCHAR?\
-DCACHE_SIZE=12000?-Davoidalloc?-DNGZhalf?\
-DMPI_BLOCK=8000?-DscaLAPACK?-Doldsym
FFT3D?=?fftmpiw.o?fftmpi_map.o?fftw3d.o?fft3dlib.o
INCS=-I$(FFTW_HOME)/include