Merge branch 'master' of gitlab.unige.ch:hpc/softs

c/cuda/cuda_visible_devices_9.1.85.sbatch

+#!/bin/sh
+#
+# <http://baobabmaster.unige.ch/enduser/src/enduser/applications.html#nvida-cuda>
+
+#SBATCH --partition=shared-gpu-EL7
+#SBATCH --time=00:15:00
+#SBATCH --gres=gpu:1
+
+echo "I: full hostname: $(hostname -f)"
+
+module load CUDA/9.1.85
+
+# see here for more samples:
+# /opt/cudasample/NVIDIA_CUDA-8.0_Samples/bin/x86_64/linux/release/
+
+# if you need to know the allocated CUDA device, you can obtain it here:
+echo "I: CUDA_VISIBLE_DEVICES: ${CUDA_VISIBLE_DEVICES}"
+
+echo "====="
+
+srun /opt/ebsofts/Compiler/GCC/6.4.0-2.28/CUDA/9.1.85/extras/demo_suite/deviceQuery

c/cuda/cuda_visible_devices_9.2.88.sbatch

+#!/bin/sh
+#
+# <http://baobabmaster.unige.ch/enduser/src/enduser/applications.html#nvida-cuda>
+
+#SBATCH --partition=shared-gpu-EL7
+#SBATCH --time=00:15:00
+#SBATCH --gres=gpu:1
+
+echo "I: full hostname: $(hostname -f)"
+
+VERSION_COMPILER='7.3.0-2.30'
+module load GCC/${VERSION_COMPILER}
+
+VERSION_CUDA='9.2.88'
+module load CUDA/${VERSION_CUDA}
+
+# see here for more samples:
+# /opt/cudasample/NVIDIA_CUDA-8.0_Samples/bin/x86_64/linux/release/
+
+# if you need to know the allocated CUDA device, you can obtain it here:
+echo "I: CUDA_VISIBLE_DEVICES: ${CUDA_VISIBLE_DEVICES}"
+
+echo "====="
+
+srun /opt/ebsofts/Compiler/GCC/${VERSION_COMPILER}/CUDA/${VERSION_CUDA}/extras/demo_suite/deviceQuery

p/pytorch/cuda_-_device_count.py

+#!/usr/bin/env python
+#
+# <https://pytorch.org/docs/stable/cuda.html>
+
+
+### ATTENTION, print behavior changed between Python 2.x and 3.x!
+from __future__ import print_function
+
+import torch
+device_count = torch.cuda.device_count()
+print("torch.cuda.device_count:", device_count)

p/pytorch/cuda_-_matrix_zeros.py

+#!/usr/bin/env python
+#
+# Create a 10x10 matrix filled with zero via the CUDA device
+#   <https://hpc-community.unige.ch/t/gpu010-cuda-singularity-cuda-runtime-error/306>
+
+
+import torch
+cuda = torch.device('cuda')
+a = torch.zeros(10, device=cuda)
+print(a)

p/pytorch/cuda_9.1.85_-_device_count.sbatch

+#!/bin/sh
+#
+# <https://hpc-community.unige.ch/t/gpu010-cuda-singularity-cuda-runtime-error/306>
+
+#SBATCH --partition=shared-gpu-EL7
+#SBATCH --time=00:15:00
+#SBATCH --gres=gpu:1
+
+echo "I: full hostname: $(hostname -f)"
+
+VERSION_PYTHON='3.6.4'
+module load GCC/6.4.0-2.28  OpenMPI/2.1.2
+module load Python/${VERSION_PYTHON}
+
+VERSION_CUDA='9.1.85'
+module load CUDA/${VERSION_CUDA}
+
+module load PyTorch/0.3.0-Python-${VERSION_PYTHON}
+
+# if you need to know the allocated CUDA device, you can obtain it here:
+echo "I: CUDA_VISIBLE_DEVICES: ${CUDA_VISIBLE_DEVICES}"
+
+echo "====="
+
+srun ./cuda_-_device_count.py

p/pytorch/cuda_9.2.148.1_-_matrix_zeros_-_singularity.sbatch

+#!/bin/sh
+#
+# <https://hpc-community.unige.ch/t/gpu010-cuda-singularity-cuda-runtime-error/306>
+
+#SBATCH --partition=shared-gpu-EL7
+#SBATCH --time=00:15:00
+#SBATCH --gres=gpu:1
+
+echo "I: full hostname: $(hostname -f)"
+
+module load GCC/5.4.0-2.26
+module load Singularity/2.4.5
+
+VERSION_CUDA='9.2.148.1'
+module load CUDA/${VERSION_CUDA}
+
+# if you need to know the allocated CUDA device, you can obtain it here:
+echo "I: CUDA_VISIBLE_DEVICES: ${CUDA_VISIBLE_DEVICES}"
+
+echo "====="
+
+srun singularity \
+    exec \
+    --bind $(readlink ${HOME}/scratch) \
+    --nv \
+    ${PWD}/pytorch.simg \
+        python \
+        ${PWD}/cuda_-_matrix_zeros.py