[LA-SiGMA-gpu] Philip GPU Nodes

[Alexander Pacheco]

Dear All,

GPU nodes on Philip are now open for testing. To submit a job using cuda compiled executable, use the newly created "gpu" queue

gpu queue characteristics
Max Walltime: 24 hrs
Max Nodes: 2

GPU Nodes:
2 Intel Xeon X5650 2.66GHz dual hexa-core with 48GB RAM and 12MB L3 cache
3 Tesla M2070 with 448 CUDA cores, 1.15 GHz clock speed and 5.2GB RAM on each node
Compilers: cuda 4.0 and PGI accelerator 11.5
Interconnect: infiniband for nodes and gpudirect for gpus

cuda and pgi compilers can be accessed using the softenv keys.
softenv -k 'cuda|portland'

The following MD packages have been compiled with cuda support and can be accessed using softenv keys
LAMMPS version 17Sep11
NAMD version 2.8
Gromacs version 4.5.4 (mdrun-gpu can only use one cpu and one gpu)

As always, please pass on any comments/questions to sys-help at loni.org

Cheers,
Alex

-- 
Alexander Pacheco, Ph. D.
IT Analyst 3
LSU - HPC &  LONI
Louisiana State University
Baton Rouge, LA 70803
http://www.cct.lsu.edu/~apacheco