Special Servers

epldt001

This server is our GPU chassis and has 2 Tesla P100 cards available for use. It has CUDA 11.0 available and the appropriate drivers active. If you login using ssh as usual and run nvidia-smi you will see the specs of the cards. You should be able (and it is recommended) to use the latest TensorFlow releases ( > 2.1.0 ) with this version of CUDA. It also best to use the CentOS 7 image rather than Fedora and in fact, in the future MWS will reinstall without the Fedora base as Nvidia seems to have stopped supporting it. When in the CentOS 7 image, the CUDA dev kit can be found at /usr/local/cuda.

epldt002

This is a powerful server to be used for software and jobs that can't be parallelised and will require signficant computing power. It has Intel Xeon Gold 6152 CPUs providing 88 cores, 128GB RAM and 3.6TB HDD space. Though this is very useful to run large scale compute tasks and is intended to stop people using the login nodes, please use the batch where possible as there is no fairshare or rationing of the resources and you could easily cause others problems if you blindly run jobs on it.