The pat cluster is a set of GPU workstations belonging to David Wales which are configured with shared filesystems and a batch queueing system. It is intended for running GPU work within a single compute node. It should not be used for MPI between compute nodes as it does not have a high speed network for message passing.
It contains the following types of compute node
- Two nodes each with four Nvidia Tesla K20 GPUs
- Twenty-six nodes with varying numbers of Nvidia Titan Black GPUs; mostly four per node.
Cluster user documentation:
Cluster admin documentation: