Access to GPUs is restricted and you must be in the appropriate Unix group (and research group in some cases) in order to use them. Please contact its-ri-team@manchester.ac.uk to enquire about GPU access.
31st May 2017 – the nvidia 2050/2070s have been retired due to hardware faults.
If you have been given access to the GPUs the following resource flags should be added to your jobscript to use the GPUs:
Nvidia C2050 or C2070 GPUs (system will choose any free node, note that only one 2050 and two 2070s are available)
#$ -l nvidia
Nvidia C2070 GPUs only
#$ -l nvidia -l 2070
Nvidia C2070 GPUs on InfiniBand-connected hosts (2 GPUs per host). Currently unavailable.
#$ -l nvidia_ib
- Nvidia K20 GPUs (2 GPUs per host). Available only to specific research groups.
#$ -l nvidia_k20
If starting an interactive session on a GPU node using qrsh
then you must specify the following flag in addition to the flags above on the qrsh
command-line:
-l inter
For example:
qrsh -cwd -V -l inter -l nvidia_k20 myapp
#
# replace myapp with your
# application/software executable
Please see the Nvidia GPUs on CSF documentation for how to write CUDA/OpenCL code and/or run GPU jobs.