LAMMPS on GPU does not use more than one CPU core

LAMMPS LAMMPS Mailing List Mirror
1

Hello, dear developers!

We have decided to use LAMMPS on GPU cores. We succeed to compile lammps (includes gpu library), which works on our computer.

But there is a problem: built lammps always uses only one CPU core. Can you explain how to use more CPU cores (as shown in example "log.gpu.melt.5.0.9Jan12.linux.4").

More information is below.
1. To compile LAMMPS we modified only /src/MAKE/Makefile.linux as follows
CC = mpicc
MPI_LIB = -lmpich -lpthread
2. The command to run the executable lammps-file is
srun -p tesla -N 1 --gres=gpu:8 -t 18 lmp_linux <in.gpu.melt.5.0
In this case only one CPU is used.
This problem is not solved by adding a key "-c4" (forced using 4 CPU)
srun -c4 -p tesla -N 1 --gres=gpu:8 -t 18 lmp_linux <in.gpu.melt.5.0
3. As before the log-files contains the line
"1 by 1 by 1 MPI processor grid"

Also we have enclosed the files desribed above.

Please, give us any idea how to solve our problem!

Sincerely, Azat O. Tipeev.

Makefile.linux (2.42 KB)

in.gpu.melt.5.0 (557 Bytes)

log.lammps (2.33 KB)

2

Hello, dear developers!

We have decided to use LAMMPS on GPU cores. We succeed to compile lammps (includes gpu library), which works on our computer.

But there is a problem: built lammps always uses only one CPU core. Can you explain how to use more CPU cores (as shown in example "log.gpu.melt.5.0.9Jan12.linux.4").

More information is below.
1. To compile LAMMPS we modified only /src/MAKE/Makefile.linux as follows
CC = mpicc
MPI_LIB = -lmpich -lpthread
2. The command to run the executable lammps-file is
srun -p tesla -N 1 --gres=gpu:8 -t 18 lmp_linux <in.gpu.melt.5.0
In this case only one CPU is used.
This problem is not solved by adding a key "-c4" (forced using 4 CPU)
srun -c4 -p tesla -N 1 --gres=gpu:8 -t 18 lmp_linux <in.gpu.melt.5.0
3. As before the log-files contains the line
"1 by 1 by 1 MPI processor grid"

how do you run other MPI based parallel programs.
normally, there has to be some "mpirun" in there
somewhere. it looks like you are using slurm, so
you have to look up details in the slurm docs or
the machine docs or talk to your admin.

this is not a lammps issue.

axel.