Hello ,
I am trying to get gpu support for the machine mentioned in the subject.
I successfully installed the cuda toolkit 4.0.13 , the devdrier 4.0 the gpu computing sdk 4.0.13, and managed to compile and run the example cuda
codes from the Nvidia cuda website.
It seems to me that my CUDA installation is ok.
When I try to compile lammps (svn version 07 May 2011), everything works smoothly (although I do not succeed to compile reax, meam, or peri).
I use mpicxx to compile parallel code.
The produced binary does work as it should, only the gpu part does not work.
In a nutshell, I followed these steps:
cd lib/gpu
make -f Makefile.linux
./nvc_get_devices
No errors, nvc_get_devices lists my graphics card.
cd lammps/src
make clean-all
make yes-asphere
make yes-class2
make yes-colloid
make yes-dipole
make yes-dsmc
make yes-gpu
make yes-granular
make yes-kspace
make yes-manybody
make no-meam
make yes-molecule
make yes-opt
make yes-peri
make no-poems
make no-reax
make yes-replica
make yes-shock
make no-srd
make yes-xtc
make no-user-ackland
make no-user-atc
make no-user-cd-eam
make no-user-cg-cmm
make no-user-eff
make no-user-ewaldn
make no-user-imd
make no-user-reaxc
make no-user-smd
make package-status
make -j4 ubuntu_parallel
No errors, everything compiles.
However, when I run a program with
pair_style lj/cut/gpu one/node 0 2.5
I get an error:
Illegal pair_style command.
I tried to compile gpu lammps, but there were so many missing libraries/link errors when I tried to include USER-CUDA that I gave up.
Thanks for any help!