# GPU on HUNT Cloud

You can add GPU cards to your On-demand and BLUE machine for your AI/ML workloads.

# GPU Models

See our GPU accelerator machine types for more details on the specific GPU models we offer.

# Software

We pre-install the machines with with the following software to get you started:

  • Ubuntu 22.04 LTS
  • Docker
  • NVIDIA 550 drivers or later
  • NVIDIA Container Toolkit (also known as nvidia-docker)

You may install other tools and versions.

# How to order

You can order your GPU machines in our administer science service desk.

# How to use

Log into your GPU lab machine and run the following command to call the NVIDIA System Management Interface (nvidia-smi) to manage and verify that things are working:

nvidia-smi

# Specify CUDA version

If you wish to use a specific version of CUDA, we recommend that you use the NVIDIA Container Toolkit to run a container of your choosing. Here is an example for CUDA version 12.0:

sudo docker run --rm --runtime=nvidia nvidia/cuda:12.0.0-base-ubuntu22.04 nvidia-smi

More versions of CUDA base image can be found in Docker Hub (opens new window).

If you do not want to use Docker, you can also try our CUDA installation guides.

# Jupyter in Nvidia Docker

If you want to use Jupyterlab on your GPU machine we recommend the iot-salzburg/gpu-jupyter (opens new window) docker image.

This project uses the NVIDIA CUDA image as the base image and installs their toolstack on top of it to enable GPU calculations in the Jupyter notebooks. Python packages Tensorflow and Pytorch are preinstalled to match the Cuda version (GPU drivers).

docker pull cschranz/gpu-jupyter:v1.4_cuda-11.6_ubuntu-20.04_python-only

docker run --rm -d --runtime=nvidia -v ${HOME}:${HOME} --workdir ${HOME} -e HOME=${HOME} -e GRANT_SUDO=yes -e JUPYTER_ENABLE_LAB=yes -p 8888:8888 --user root --name gpu-jupyter cschranz/gpu-jupyter:v1.4_cuda-11.6_ubuntu-20.04_python-only

docker ps

The commands above starts an instance of GPU-Jupyter with the tag v1.4_cuda-11.6_ubuntu-20.04_python-only at http://localhost:8888 (port 8888). The default password is gpu-jupyter.

It is also possible to start an interactive bash session and run scripts inside:

docker run --rm --runtime=nvidia -v ${HOME}:${HOME} --workdir ${HOME} -e HOME=${HOME} -v /etc/passwd:/etc/passwd -v /etc/group:/etc/group --user "1000" --entrypoint bash -ti cschranz/gpu-jupyter:v1.4_cuda-11.6_ubuntu-20.04_python-only

More details about this docker image can be found on docker hub (opens new window).

# More info

Last Updated: 12/9/2024

← Introduction →