Hi, all,
I was running lammps with cuda package (lammps version 5Mar12, cuda library is 4.0) and encountered an unexpected problem, named “unspecified launch failure” from cuda. Tests by disabling other compute had been done and seems the problem is caused by compute stress/atom here. This command is needed prior to compute heat/flux. I am using a tersoff potential for a less than 10000 atoms carbon nanotube with fixed boundary condition.
For this case the error comes from cuda_nve.cu. However, I have other longer input script which had the exactly error message but comes from cuda_wrapper.cu. Also need to note is that it is perfectly ok to run for 1 step however any longer steps will generate the problem I described.
Here below is a shorter version script I had.
log SWNT1010_400A_equ.log
clear
suffix cuda
units metal
boundary f f f
atom_style atomic
read_data SWNT1010_400A.lammps05
pair_style tersoff
pair_coeff * * C_thermal.tersoff C
region hot block INF INF INF INF 30 60 units box
group hot region hot
compute hotMyKE hot ke/atom
compute hotMyPE hot pe/atom
compute hotMyStress hot stress/atom virial
compute hotFlux hot heat/flux hotMyKE hotMyPE hotMyStress
timestep 0.001
thermo 1000
thermo_style custom step pe ke etotal enthalpy vol press temp c_hotFlux[1] c_hotFlux[2] c_hotFlux[3]
fix 1 all nve
dump 1 all custom 1 SWNT1010_400A_equ.lammpstrj id type x y z vx vy vz
run 10
unfix 1
undump 1
write_restart SWNT1010_400A_equ.restart