Arkansas High Performace Computing Center [hpcwiki]

User Tools

Site Tools

Trace:
vasp

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revisionPrevious revision
Next revision		Previous revision
vasp [2020/02/17 21:22] – rootvasp [2026/08/07 18:16] (current) – pwolinsk
Line 29: Line 29:
 installs$ installs$
 </code> </code>
 +
 ''getshmem.o'' and ''linpack_double.o'' are made in build/lib, and modules and ''makefile.include'' are: ''getshmem.o'' and ''linpack_double.o'' are made in build/lib, and modules and ''makefile.include'' are:
 +
 <code> <code>
 installs$ module load intel/19.0.5 mkl/19.0.5 impi/19.0.5 installs$ module load intel/19.0.5 mkl/19.0.5 impi/19.0.5
Line 115: Line 117:
 MPI_INC    = $(I_MPI_ROOT)/intel64/include/ MPI_INC    = $(I_MPI_ROOT)/intel64/include/
 installs$  installs$ 
 +
 </code> </code>
 +
 Some run examples with an OpenMP, MKL sequential build: Some run examples with an OpenMP, MKL sequential build:
 +
 <code> <code>
 module load intel/19.0.5 mkl/19.0.5 impi/19.0.5 module load intel/19.0.5 mkl/19.0.5 impi/19.0.5
 mpirun -np 16 -genv OMP_NUM_THREADS=2 /scrfs/apps/vasp/vasp.5.4.4/build/std/vasp.O2.smkl mpirun -np 16 -genv OMP_NUM_THREADS=2 /scrfs/apps/vasp/vasp.5.4.4/build/std/vasp.O2.smkl
-took 75 seconds on Trestles, 35 seconds on Razor 16-core (with OMP_NUM_THREADS=1)**, 24 seconds on Pinnacle.+took 75 seconds on Trestles, 35 seconds on Razor 16-core (with OMP_NUM_THREADS=1)**, 27.3 seconds on Pinnacle. 
 +vasp.o2.smkl took 26.1 seconds on Pinnacle.
 </code> </code>
 +
 **For performance MPI threads x OpenMP threads should not exceed physical cores. **For performance MPI threads x OpenMP threads should not exceed physical cores.
-Parallel MKL and -O3 didn't help performance. 
  
  
  
vasp.1581974549.txt.gz · Last modified: (external edit)