Devito images prepared to be used on HPC Cluster Mintrop
645
https://ndf.poli.usp.br/recursos-computacionais/
Allocate a GPU node with one of the following partitions
srun -p gpu_debug --pty bash -y
srun -p gpu_small --pty bash -y
srun -p gpu_large --pty bash -y
Load Singularity, CUDA and CUDNN modules:
module purge
module load singularity-3.5.3-gcc-8.3.0-3hixhf5 \
cudnn-8.1.1.33-11.2-gcc-8.3.0-cuda-puavfnj \
cuda-11.2.2-gcc-8.3.0-zuguy2i
Pull one of the nvidia tags:
singularity pull docker://mogekag/devito:nvidia-4.8.7 devito.sif
Then, start the container with the --nv flag to have access to GPU cards and -e flag to avoid inheriting module environment variables.
singularity shell --nv -e devito.sif bash
If you need a different devito version, you can either build it using the following Dockerfile, or repeat the commands on an interactive shell or script, prior to running your code.
Remember to replace the version to whatever devito version you want to install. The version must also be a github tag on devito repository, which is usually the case for releases.
FROM devitocodes/devito:nvidia-nvc-v4.8.0
USER root
RUN apt update -y && apt install -y git
RUN git clone https://github.com/devitocodes/devito.git /opt/devito
WORKDIR /opt/devito
RUN git checkout v4.8.7 && \
python3 -m pip install --upgrade -e .[extras,nvidia,tests]
Content type
Image
Digest
sha256:0f749a063…
Size
5.6 GB
Last updated
over 2 years ago
docker pull mogekag/devito