OpenHPC Container Recipe

singularity exec ohpc-recipe4.simg python /benchmark.py

note: scale down the memory usage in Vagrantfile if your system can't support the specified amount (4096)

This is a container recipe for NVIDIA's HPC container maker. The base image is OpenHPC's development environment, with added Python, TensorFlow, and Keras support

---

[share]$ nvcc --version nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2015 NVIDIA Corporation Built on Tue_Aug_11_14:27:32_CDT_2015 Cuda compilation tools, release 7.5, V7.5.17

module: loading 'cuda/7.5.18'

---

tensorflow-2.6.0 cuDNN 8.1 cuda 11.2

in sbatch script, module load cuda/11.3.1 module load cudnn/8.1.0

https://www.tensorflow.org/install/source

Usage examples

Docker

hpccm --recipe ohpc-recipe.py --format docker > Dockerfile
docker build -t ohpc-recipe -f Dockerfile .
docker run -v $PWD/python_scripts/:/mnt/python_scripts/ -it --rm ohpc-recipe python3.7 /mnt/python_scripts/test.py

Singularity

Note: For singularity builds, root access is required. If you are on MacOS or Windows, please check out the instructions here on how to use Vagrant to build a Singularity virtual machine

hpccm --recipe ohpc-recipe.py --singularity-version=3.8 --format singularity > Singularity.def version 3.8 for multistage

hpccm --recipe ohpc-recipe.py --format singularity > Singularity.def
sudo singularity build ohpc-recipe.simg Singularity.def
singularity exec --nv ohpc-recipe.simg python3 $PWD/python_scripts/benchmark.py

An alternate solution is to build using Docker, then rebuild as singularity singularity build ohpc-recipe.simg docker://kevinhsuk/ohpc-recipe