Skip to content

Smart Medicial Imaging CDT

Smart Medical Imaging (SMI) CDT PhD students have access to a dedicated pool of GPU resources on CREATE HPC for the purposes of real-time development of models. Each student with access to this pool can schedule a single GPU instance in real-time with no limit on job length. This resource is akin to having a GPU in a local development machine with the benefit that once code is developed via this provision it is immeadiately ready to be queued up to run on CREATE HPC's larger A100 based GPU pool. The setup is based on servers with 8 x A30 GPUs installed. These have currently been split into 4 x 24GB instances and 8 x 12GB instances. The splitting of the A30 GPUs is acheived using the 'Multi Instance GPU' feature.

Student Access

Info

Students are limited to 1 GPU, 120GB RAM, 12 CPU cores each

  1. Register for CREATE HPC access
  2. Review our documentation for accessing the cluster via SSH
  3. If you are working interactively it is advisable to start a tmux or screen session on the login nodes so that you can reconnect to your session if you lose connectivity (this beginners guide to tmux might help)
  4. Depending on the size of your model access 12GB or 24GB instances in real-time as follows:
1
2
srun -n 12 --mem 122880 -p smi_cdt_gpu --gres=gpu:2g.12gb:1 --pty /bin/bash
srun -n 12 --mem 122880 -p smi_cdt_gpu --gres=gpu:4g.24gb:1 --pty /bin/bash

For more general information about using CREATE HPC, please start by reviewing our 'Running Jobs' doc.

Administration

CDT management can allocate students access to this resource as follows:

  1. Browse to https://portal.er.kcl.ac.uk/groups/1441#access
  2. Click 'Add user'
  3. Enter the email address of the student you wish to add and click the search button
  4. Confirm the details of the search are as expected and click 'Add user' to confirm