We have 2 Linux GPU compute servers available in COM for staff use.

These are currently experimental systems that were built mostly from spare parts and should not be expected to provide rock-solid service, so it is suggested they mainly be used for testing/developement at this stage.

ford and arthur have NVIDIA GeForce RTX 3060 cards. The OS is Ubuntu Server, and the CUDA libraries are available in /usr/local/cuda. You can see the current usage status of the GPUs and which version of the NVIDIA CUDA drivers are available by running the gpustat command.
You can access the server remotely from on-campus or from VPN, via SSH with your username/password.
ford uses the DCS account authentication scheme, so you need a DCS account to use this one.
arthur uses ITS (university) accounts so you use your uni credentials, but you will need to request access first.

Requesting access to these servers
Any staff/PhD in COM can have access to these but they must first request it using the COM SR system so they can be added to the access list on the servers. See Contacting Technical Support for how to.

Resource sharing / Job submission
No job submission system is currently installed on the systems. If we do find that usage is high then we might have to re-consider this in future. For the moment, users should be aware that others may be using the system at the same time and therefore should not necessarily expect 100% access to the processing capability.

Support for GPU computing
COM Tech Support cannot provide help for programming the GPU cards. The system is provided for those who already know how to use the GPU capability, or can find out how to do it.

Data storage
There is some Terabytes of storage in /data available on both servers, which should be considered a temporary working area or scratch space.

ford mounts the home directory (20GB quota) to $HOME.
arthur has only a small space in $HOME to store dotfiles. If you have a DCS account with some data stored there (sometimes known as the H-drive) you can mount that by first creating a local directory (mkdir $HOME/dcs) and then running sshfs your_dcs_username@staffstore.dcs.shef.ac.uk: $HOME/dcs