Installed in Feb 2019 as part of the University fleet refresh.


  • i7-8700 @ 3.2GHz, 64GB

  • Nvidia Quadro P2000 5GB

  • 64GB RAM (upgraded from 32GB)

  • 1TB SSD system drive

  • 1TB SSD dataset drive (mounted at /mnt/datasets)


  • Ubuntu 20.04.3; CUDA 11.4; cuDNN 8.2.2

  • R 4.1.2; R packages: MRAN 2022-01-18

  • Python 3.8.10; Tensorflow 2.5.0 (GPU enabled)

Note: Configuring CUDA/cuDNN/Tensorflow is fiddly; this Medium post by Harish Masand from August 2021 was helpful.


/mnt/datasets/objectnet (190GB) - The ObjectNet image set.

Logging on

  1. Be on campus, or connect to the campus network via the VPN.

  2. Log on via SSH: user@ If you want to be able to e.g. view images, you can ssh -X user@ and use the eog command.

Using tensorflow

In order to use Tensforflow, you must edit your .bashrc file. Each user must do this once. Specifically, add this to the end of your file:

export LD_LIBRARY_PATH=/usr/local/cuda-11.4/lib64:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH

save, exit, and then source ~/.bashrc

Remote reboot

Remote reboot via SSH was set up following these instructions concerning dropbear-initramfs. Currently, only AW has the SSH key required to gain access to this functionality, and only AW knows the password to decrypt the drive once access is granted. Send an SMS to AW in case of emergency.

There are three steps

  1. Reboot: sudo reboot

  2. Unlock drive: ssh root@ -p 8022 -i ~/.ssh/id_rsa, wait until you see the “please unlock” dialog including the final colon, type in the password that was used to encrypt the drive.

  3. Now you can log in normally, as per above.