The AMD Interlagos chip has only one FPU per two integer unit cores. As a result for almost all scientific software it can only hurt you to run every core on a node. However only running one core of each pair is a bit of a challenge in the OIC environment. Please see the method in the PBS examples. This results in a 50-100% speed up in VASP and should have a similar effect whenever the code is primarily doing floating point operations.
## Scratch Space
$GLOBAL_SCRATCH is available within jobs. It is immediately deleted when the job completes. You do have access to the directory while the job runs.
## Specific Codes
### LAMMPS
Has undergone only minimal testing via the examples. Additionally I may have not built important extensions, I don't know which user written extensions are usually used. Happy to build these.
