There are some applications that benefit from GPU computing, typically using the CUDA libraries or applications that are able to take advantage of GPU computing. There are additional servers in the BlueBEAR service to meet these needs.
These servers are considered as a scarce resource and are controlled as follows:
- registration on the general BlueBEAR service is a prerequisite for access to these nodes. See the BlueBEAR registration help page for details about registering on the general BlueBEAR service.
- experience of running work using the standard resources is required prior to requesting access to the GPGPU service.
- to request access to this service follow the instructions in the Registering a username for specialist BEAR services section of the BlueBEAR registration help page.
- you will be invited in by IT Services to discuss your work and identify the best way of meeting your needs
- approved users will be added to the bbgpu QoS in SLURM and will then be able to submit jobs to these nodes. See the help page for job submission for more details about submitting jobs.
- only one job will be allowed on these nodes at any time, since GPU sharing between jobs is not recommended nor supported.
- a job may only make use of one of the GPU-enabled nodes; it is not possible to run jobs across them.
Please contact the IT Service Desk with any questions about this service.
Last modified: 2 August 2017