Dawn is a supercomputer hosted at the University of Cambridge, and is part of the AI Resource Research (AIRR). It makes available Intel Data Centre GPU Max 1550 GPUs. To install and run the workshop software on Dawn, you will need to have been granted access. For further information, see: Access to Dawn.
If you haven't done this already, and have an account, enable access to Dawn via the login-web interface.
Login at: https://login-web.hpc.cam.ac.uk/
At the top of the dashboard page presented after login, select:
Interactive Apps -> Jupyter Notebook
On the Jupyter Notebook form enter:
- Project account: training-dawn-gpu
- Partition: pvc9
- Reservation [leave blank]:
- Number of hours: 2
- Number of cores: 1
- Number of GPUs: 2
- Modules: intel-oneapi-mkl intel-oneapi-compilers jupyterlab
- Number of nodes: 1
The above request the resources needed for this workshop, for a period of 2 hours. The number of hours may be decreased or increased as needed.
Click the Launch button.
Your request for a Jupyter Notebook will progress through the states: Queued, Starting, Running. Once the Running state is reached, click the Connect to Jupyter button.
Once connected to the Jupyter server, you will initially be shown the File Browser, in the Jupyter Home tab.
On the Jupyter page, select:
File -> New -> Terminal
The Terminal window that opens will be in the home directory of your account on Dawn.
Clone this repository to your home directory:
git clone https://github.com/RSE-Cambridge/RSECon25-Dawn-Workshop
Set up the environement:
source ~/RSECon25-Dawn-Workshop/scripts/workshop_setup.sh
This will clone another repository with examples (practical-ml-with-pytorch)
to your home directory, will create account-specific setup files in
~/RSECon25-Dawn-Workshop/install and will create kernel-definition files
in ~/.local/shared/jupyter/kernels.
Move to the workshop examples directory, and submit the two jobs that perform multi-node processing:
cd ~/RSECon25-Dawn-Workshop/examples
sbatch --account=training-dawn-gpu run_mnist_classify_ddp.sh
sbatch --account=training-dawn-gpu run_lightning_toy_example.sh
On the Jupyter page, select:
File -> View -> Open JupyterLab
In the left panel, navigate to RSECon25-Dawn-Workshop/examples. Open,
and experiment with: check_devices.ipynb. Before running the notebook,
set the kernel to ai:
Kernel -> Change Kernel... -> ai
In the left panel, navigate to practical-ml-with-pytorch/worked-solutions.
Open and experiment with the four notebooks. Before running each notebook,
set the kernel to practical-ml-with-pytorch:
Kernel -> Change Kernel... -> practical-ml-with-pytorch