Use scratch gpfs on Tigergpu - SSDs

ASvyatkovskiy · ASvyatkovskiy · commit 151738c743d1 · 2018-02-23T15:53:40.000-05:00
diff --git a/examples/slurm.cmd b/examples/slurm.cmd
@@ -1,27 +1,25 @@
 #!/bin/bash
-#SBATCH -t 01:30:00
+#SBATCH -t 01:00:00
 #SBATCH -N 3
 #SBATCH --ntasks-per-node=4
 #SBATCH --ntasks-per-socket=2
 #SBATCH --gres=gpu:4
 #SBATCH -c 4
 #SBATCH --mem-per-cpu=0
 
-export PYTHONHASHSEED=0
-module load anaconda
-source activate pppl
+module load anaconda/4.4.0
+source activate PPPL
 module load cudatoolkit/8.0
 module load cudnn/cuda-8.0/6.0
 module load openmpi/cuda-8.0/intel-17.0/2.1.0/64
-module load intel/17.0/64/17.0.4.196 intel-mkl/2017.3/4/64
+module load intel/17.0/64/17.0.4.196
 
 #remove checkpoints for a benchmark run
-rm /tigress/$USER/model_checkpoints/*
-rm /tigress/$USER/results/*
-rm /tigress/$USER/csv_logs/*
-rm /tigress/$USER/Graph/*
-rm /tigress/$USER/normalization/*
+rm /scratch/gpfs/$USER/model_checkpoints/*
+rm /scratch/gpfs/$USER/results/*
+rm /scratch/gpfs/$USER/csv_logs/*
+rm /scratch/gpfs/$USER/Graph/*
+rm /scratch/gpfs/$USER/normalization/*
 
 export OMPI_MCA_btl="tcp,self,sm"
-
 srun python mpi_learn.py
