We have 4 GPU nodes – 2 that have 4 Nvidia H100’s each and 2 that hold 4 Nvidia A30’s each.  Currently these nodes, called the ‘chewy’ nodes are separate from the Wooki cluster because we have run out of network ports available on the internal network switches.  This will be temporary.

Like Wooki the machines are only available from within the Unversity’s network and when off-campus one must use the University’s VPN.

To access these machines, you need to use the IP number/address as the hostname has not been registered with the University.  However, you can access them from Wooki using the hostname shortcuts.

IP address:  10.40.33.59         hostname:  chewy1           Hardware:  4x A30
IP address:  10.40.33.60         hostname:  chewy2           Hardware:  4x A30
IP address:  10.40.33.61         hostname:  chewy3           Hardware:  4x H100
IP address:  10.40.33.62         hostname:  chewy4           Hardware:  4x H100

As of Jan 20, usernames and passwords from Wooki have been copied over to these nodes, and the home directory from Wooki is mounted on each of these nodes. ‘shared_scratch’ is not currently available on the chewy nodes.

The chewy nodes have the latest version of Alma Linux installed on them (this is a Linux distro that is a CentOs replacement) with the latest Nvidia drivers installed as well.

For now, we have set up the nodes quickly to allow users to figure out how to install their software on the machines and run test jobs.  For now, one can run their jobs directly without using the queue (we will set this up a slurm queue on each node later).  One can see what other jobs are running on the queue using the ‘top’ command.  The GPU status can be examined with the ‘nvidia-smi’ command.