Hi, lammps-users,
I tried the lammps/examples/accelerate using CUDA and GPU packages.
It’s okay for “lammps -sf gpu < in.lj” & “mpirun -np 8 lammps -sf gpu < in.lj”.
However, error occurred, when typing “mpirun -np 12 lmp_gpu_double -sf gpu -pk gpu 2 < in.lj” & “mpirun -np 4 lmp_gpu_double -sf gpu -pk gpu 2 tpa 8 < in.lj.5.0”, messages following:
Hi, lammps-users,
I tried the lammps/examples/accelerate using CUDA and GPU packages.
It's okay for "lammps -sf gpu < in.lj" & "mpirun -np 8 lammps -sf gpu <
in.lj".
the common denominator of all the failures below is that you try to
access two GPUs.
all of those commands obviously work for a machine that is correctly
set up and has the requested hardware available and activated via its
driver.
the obvious conclusion is that this has to do with your machine setup
and that is something that we cannot correct or easily advise on from
remote.
axel.
Thanks a lot for Pro. Axel
GPU package can be used correctly with pair_style eam/alloy/gpu now. However, following test results do not show any accelerate:
190.65 seconds was cost per 10,000 steps with GPU;
177.5 seconds was cost per 10,000 steps without GPU;
For CUDA package, it still cannot be used correctly.
I have a question. Can I say that, the successfully run of GPU package indicates the driver and CUDA-7.5 tool kit have been correctly compile?
Thanks for lammps-users.