These are input scripts used to run versions of several of the benchmarks in the top-level bench directory using the GPU and USER-CUDA accelerator packages. The results of running these scripts on two different machines (a desktop with 2 Tesla GPUs and the ORNL Titan supercomputer) are shown on the "GPU (Fermi)" section of the Benchmark page of the LAMMPS WWW site: lammps.sandia.gov/bench.

Examples are shown below of how to run these scripts. This assumes you have built 3 executables with both the GPU and USER-CUDA packages installed, e.g.

lmp_linux_single lmp_linux_mixed lmp_linux_double

The precision (single, mixed, double) refers to the GPU and USER-CUDA pacakge precision. See the README files in the lib/gpu and lib/cuda directories for instructions on how to build the packages with different precisions. The GPU and USER-CUDA sub-sections of the doc/Section_accelerate.html file also describes this process.

Make.py -d ~/lammps -j 16 -p #all orig -m linux -o cpu exe Make.py -d ~/lammps -j 16 -p #all opt orig -m linux -o opt exe Make.py -d ~/lammps -j 16 -p #all omp orig -m linux -o omp exe Make.py -d ~/lammps -j 16 -p #all gpu orig -m linux \

-gpu mode=double arch=20 -o gpu_double libs exe

Make.py -d ~/lammps -j 16 -p #all gpu orig -m linux \

-gpu mode=mixed arch=20 -o gpu_mixed libs exe

Make.py -d ~/lammps -j 16 -p #all gpu orig -m linux \

-gpu mode=single arch=20 -o gpu_single libs exe

Make.py -d ~/lammps -j 16 -p #all cuda orig -m linux \

-cuda mode=double arch=20 -o cuda_double libs exe

Make.py -d ~/lammps -j 16 -p #all cuda orig -m linux \

-cuda mode=mixed arch=20 -o cuda_mixed libs exe

Make.py -d ~/lammps -j 16 -p #all cuda orig -m linux \

-cuda mode=single arch=20 -o cuda_single libs exe

Make.py -d ~/lammps -j 16 -p #all intel orig -m linux -o intel_cpu exe Make.py -d ~/lammps -j 16 -p #all kokkos orig -m linux -o kokkos_omp exe Make.py -d ~/lammps -j 16 -p #all kokkos orig -kokkos cuda arch=20 \

-m cuda -o kokkos_cuda exe

Make.py -d ~/lammps -j 16 -p #all opt omp gpu cuda intel kokkos orig \

-gpu mode=double arch=20 -cuda mode=double arch=20 -m linux \
-o all libs exe

Make.py -d ~/lammps -j 16 -p #all opt omp gpu cuda intel kokkos orig \

-kokkos cuda arch=20 -gpu mode=double arch=20 \
-cuda mode=double arch=20 -m cuda -o all_cuda libs exe

To run on just CPUs (without using the GPU or USER-CUDA styles), do something like the following:

mpirun -np 1 lmp_linux_double -v x 8 -v y 8 -v z 8 -v t 100 < in.lj mpirun -np 12 lmp_linux_double -v x 16 -v y 16 -v z 16 -v t 100 < in.eam

The "xyz" settings determine the problem size. The "t" setting determines the number of timesteps.

These mpirun commands run on a single node. To run on multiple nodes, scale up the "-np" setting.

To run with the GPU package, do something like the following:

mpirun -np 12 lmp_linux_single -sf gpu -v x 32 -v y 32 -v z 64 -v t 100 < in.lj mpirun -np 8 lmp_linux_mixed -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.eam

The "xyz" settings determine the problem size. The "t" setting determines the number of timesteps. The "np" setting determines how many MPI tasks (per node) the problem will run on. The numeric argument to the "-pk" setting is the number of GPUs (per node); 1 GPU is the default. Note that you can use more MPI tasks than GPUs (per node) with the GPU package.

These mpirun commands run on a single node. To run on multiple nodes, scale up the "-np" setting, and control the number of MPI tasks per node via a "-ppn" setting.

To run with the USER-CUDA package, do something like the following:

mpirun -np 1 lmp_linux_single -c on -sf cuda -v x 16 -v y 16 -v z 16 -v t 100 < in.lj mpirun -np 2 lmp_linux_double -c on -sf cuda -pk cuda 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.eam

The "xyz" settings determine the problem size. The "t" setting determines the number of timesteps. The "np" setting determines how many MPI tasks (per node) the problem will run on. The numeric argument to the "-pk" setting is the number of GPUs (per node); 1 GPU is the default. Note that the number of MPI tasks must equal the number of GPUs (both per node) with the USER-CUDA package.

These mpirun commands run on a single node. To run on multiple nodes, scale up the "-np" setting, and control the number of MPI tasks per node via a "-ppn" setting.

If the script has "titan" in its name, it was run on the Titan supercomputer at ORNL.