1
These are input scripts used to run GPU versions of several of the
2
benchmarks in the top-level bench directory. The results of running
3
these scripts on different machines are shown on the GPU section of
4
the Benchmark page of the LAMMPS WWW site (lammps.sandia.gov/bench).
6
Examples are shown below of how to run these scripts. This assumes
7
you have built 3 executables with both the GPU and USER-CUDA packages
14
The precision (single, mixed, double) refers to the GPU and USER-CUDA
15
pacakge precision. See the README files in the lib/gpu and lib/cuda
16
directories for instructions on how to build the packages with
17
different precisions. The doc/Section_accelerate.html file also has a
20
------------------------------------------------------------------------
22
If the script has "cpu" in its name, it is meant to be run in CPU-only
25
mpirun -np 1 ../lmp_linux_double -c off -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cpu
26
mpirun -np 12 ../lmp_linux_double -c off -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cpu
28
The "xyz" settings determine the problem size. The "t" setting
29
determines the number of timesteps.
31
------------------------------------------------------------------------
33
If the script has "gpu" in its name, it is meant to be run using
34
the GPU package. For example:
36
mpirun -np 12 ../lmp_linux_single -sf gpu -c off -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
38
mpirun -np 8 ../lmp_linux_mixed -sf gpu -c off -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
40
The "xyz" settings determine the problem size. The "t" setting
41
determines the number of timesteps. The "np" setting determines how
42
many CPUs the problem will be run on, and the "g" settings determines
43
how many GPUs the problem will run on, i.e. 1 or 2 in this case. You
44
can use more CPUs than GPUs with the GPU package.
46
------------------------------------------------------------------------
48
If the script has "cuda" in its name, it is meant to be run using
49
the USER-CUDA package. For example:
51
mpirun -np 1 ../lmp_linux_single -sf cuda -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cuda
53
mpirun -np 2 ../lmp_linux_double -sf cuda -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.eam.cuda
55
The "xyz" settings determine the problem size. The "t" setting
56
determines the number of timesteps. The "np" setting determines how
57
many CPUs the problem will be run on, and the "g" setting determines
58
how many GPUs the problem will run on, i.e. 1 or 2 in this case. You
59
should make the number of CPUs and number of GPUs equal for the