This shows you the differences between two versions of the page.
Both sides previous revision Previous revision Next revision | Previous revision Next revision Both sides next revision | ||
cluster:73 [2021/04/12 19:17] hmeij07 [LAMMPS CPU/GPU] |
cluster:73 [2021/05/10 19:45] hmeij07 [OpenHPC] |
||
---|---|---|---|
Line 18: | Line 18: | ||
on either cottontail2 or node n33 located at / | on either cottontail2 or node n33 located at / | ||
Some compilations require 7.x and are in / | Some compilations require 7.x and are in / | ||
+ | |||
+ | ====== Ovito ====== | ||
+ | |||
+ | * https:// | ||
+ | * / | ||
+ | * on nodes '' | ||
+ | * uses freeglut/ | ||
====== OpenHPC ====== | ====== OpenHPC ====== | ||
Line 1341: | Line 1348: | ||
This is a version of lammps (29Oct20, last stable) that contains all packages with user-quip. | This is a version of lammps (29Oct20, last stable) that contains all packages with user-quip. | ||
- | It does not compile with MPI, nor ifort in CentOS7, so there is only a " | + | It does compile with MPI (mpif90), not ifort, in CentOS7 ... was tricky, solution for MPI was the '' |
- | * Requires CentOS 7 so queues: mwgpu, amber128, exx96 | + | * Requires CentOS 7 so queues: mwgpu, amber128, exx96 for cpu-only binaries |
+ | * Rwequires CentOS7 plus Cuda 10.2 so queue: exx96 for cpu+gpu binaries | ||
Read, if you wish | Read, if you wish | ||
Line 1350: | Line 1358: | ||
* / | * / | ||
- | Environments | + | **Environments** |
< | < | ||
# serial - cpu only (although quippy compiled, it refused to install) | # serial - cpu only (although quippy compiled, it refused to install) | ||
+ | # MPI STUBS is compiled in | ||
export PATH=/ | export PATH=/ | ||
export LD_LIBRARY_PATH=/ | export LD_LIBRARY_PATH=/ | ||
- | # serial | + | / |
+ | |||
+ | # mpi - cpu only | ||
+ | # gfortran_openmpi | ||
+ | |||
+ | export QUIP_LIBRARY=/ | ||
+ | export PATH=/ | ||
+ | export LD_LIBRARY_PATH=/ | ||
+ | export PATH=/ | ||
+ | export LD_LIBRARY_PATH=/ | ||
+ | |||
+ | / | ||
+ | |||
+ | |||
+ | # mpi - cpu+gpu | ||
+ | # gfortran_openmpi arch | ||
+ | |||
+ | export PATH=/ | ||
+ | export LD_LIBRARY_PATH=/ | ||
+ | export QUIP_LIBRARY=/ | ||
+ | export PATH=/ | ||
+ | export LD_LIBRARY_PATH=/ | ||
+ | export PATH=/ | ||
+ | export LD_LIBRARY_PATH=/ | ||
+ | export CUDA_HOME=/ | ||
+ | export PATH=/ | ||
+ | export LD_LIBRARY_PATH=/ | ||
+ | export LD_LIBRARY_PATH=/ | ||
+ | |||
+ | # the " | ||
+ | # did find on nvdia dev forum reference problem disappears when using cmake (vs make) | ||
+ | # these lmp binaries do work | ||
+ | / | ||
+ | / | ||
+ | / | ||
+ | |||
+ | # however I had to load non-mpi version of libquid.a into lmp | ||
+ | # witch seems ok given parallel work is driven by -suffix | ||
- | (to come, hopefully) | + | # sample script: ~hmeij/ |
</ | </ |