Print
There are 3 computing nodes “compute-9-0”, “compute-9-1” and “compute-9-2” installed with 12 NVIDIA® Tesla® V100 CPU cards, and 2 computing nodes “compute-9-10” and “compute-9-11” installed with 16 NVIDIA® Tesla® V100 CPU cards.

You can submit your job to “gpu-short” or “gpu-normal” queue for using GPU acceleration.

Here is an example of submission script for running in a single node:

#!/bin/sh
#$ -S /bin/bash
#$ -cwd
#$ -V
#$ -N p1
#$ -j y
#$ -A username
#$ -q gpu.q
#$ -pe 12_orte 12
#$ -m bes
#$ -M email

# Load standard environment
. /opt/gridengine/default/common/settings.sh

# Load module environmnet
. /etc/profile.d/modules.sh

# Add GCC module
module add gcc/4.8.4

# Add CUDA module
module add cuda/6.5.12

# Add OpenMPI module
module add ompi-gnu/1.8.5-gcc-4.8.4-cuda-6.5.12

MV2_USE_CUDA=1 mpirun -np 1 ./jacobi_cuda_aware_mpi -t 1 1

Here is another example of submission script for running parallel program in both GPU computing nodes:

#!/bin/sh
#$ -S /bin/bash
#$ -cwd
#$ -V
#$ -N p1
#$ -j y
#$ -A username
#$ -q gpu.q
#$ -pe 12_orte 24
#$ -m bes
#$ -M email

# Load standard environment
. /opt/gridengine/default/common/settings.sh

# Load module environmnet
. /etc/profile.d/modules.sh

# Add GCC module
module add gcc/4.8.4

# Add CUDA module
module add cuda/6.5.12

# Add OpenMPI module
module add ompi-gnu/1.8.5-gcc-4.8.4-cuda-6.5.12

MV2_USE_CUDA=1 mpirun -np 2 ./jacobi_cuda_aware_mpi -t 2 1

After that, you can use qsub to submit your program to cluster.

sbatch gpu_jobs.sh