Casper cluster¶

The Casper cluster is a system of specialized data analysis and visualization resources; large-memory, multi-GPU nodes; and high-throughput computing nodes.

Casper is composed of over 100 nodes featuring a mixture of Intel and AMD processors, with a variety of NVIDIA General Purpose Graphical Processing Units.

Please refer to the hardware summary table below for detailed specifications.

Quick Start¶

Logging in¶

Once you have an account, have reviewed the Casper Use Policies, and have a Casper resource allocation you can log in and run jobs on the Casper data analysis and visualization cluster.

To log in, start your terminal or Secure Shell client and run an ssh command as shown here:

ssh -X username@casper.hpc.ucar.edu

Some users (particularly on Macs) need to use -Y instead of -X when calling SSH to enable X11 forwarding.

You can omit username in the command above if your Casper username is the same as your username on your local computer.

After running the ssh command, you will be asked to authenticate to finish logging in.

Casper has full access to NSF NCAR storage resources, including GLADE. Users can transfer data to and from Casper.

To run data analysis and visualization jobs on the Casper system's nodes, follow the procedures described here. There is no need to transfer output files from Derecho for this since Derecho and Casper mount the same GLADE file systems.

Environment¶

The Casper HPC system uses OpenSUSE Linux Version 15 and supports widely used shells on its login and compute nodes. Users also have several compiler and MPI library choices.

Shells¶

The default login shell for new Casper users is bash. You can change the default after logging in to the Systems Accounting Manager (SAM). It may take several hours for a change you make to take effect. You can confirm which shell is set as your default by entering echo $SHELL on your Casper command line.

Environment modules¶

The Casper module utility enables users to easily load and unload compilers and compatible software packages as needed, and to create multiple customized environments for various tasks. See the Environment modules page for a general discussion of module usage. Casper's default module environment is listed here.

Accessing software and compiling code¶

Casper users have access to Intel, NVIDIA, and GNU compilers. The Intel compiler and OpenMPI modules are loaded by default and provide access to pre-compiled HPC Software and Data Analysis and Visualization Resources.

See this page for a full discussion of compiling on Casper.

Many Casper data analysis and AI/ML workflows benefit instead from using Conda, especially NSF NCAR's Python Library (NPL) or to gain access to several Machine Learning Frameworks.

Running jobs on Casper¶

Users can run a variety of types of jobs on Casper, including both traditional batch jobs submitted through PBS and also interactive and/or graphics-intensive analysis, often through remote desktops on Casper.

Job scripts¶

Job scripts are discussed broadly here. Users already familiar with PBS and batch submission may find Casper-specific PBS job scripts helpful in porting their work.

Casper hardware¶

Status¶

Nodes¶

GPU Usage¶

Queues¶