Steps to Get an Account for a Princeton Cluster
1. Decide Which Cluster to Use
Princeton University has a range of high-performance computing resources that are available to faculty and students.
Broadly speaking, there are two types of systems, small and large.
The smaller systems (Nobel, Adroit, Tigressdata) are meant for exploration, small production runs, and visualization. The larger systems (Della, Tiger, Stellar, Traverse) are meant for solving computationally demanding research.
To find which system is best suited to your work, we recommend exploring the Systems Overview table on the Princeton Research Computing website. There is also a more technical Hardware Overview table for those interested.
An important note is that the Stellar and Traverse clusters are largely dedicated to astrophysics, plasma physics research, chemical and biological engineering and atmospheric and oceanic sciences. Researchers outside those fields who need larger clusters are encouraged to use Della or Tiger.
2. Get An Account for Your Chosen Cluster
If you joined the university before July 2021, first make sure your NetID is enabled for Unix.
Although you can visit each cluster's page for instructions on getting an account and accessing that particular cluster, below we provide a general overview of getting access to each cluster, in addition to detailed guidelines on getting access to the larger clusters.
For Nobel, no further action is needed to get an account. All Princeton faculty, students, and staff have automatic access.
For Adroit, fill out the online registration form (Adroit registration). You will be notified once your account is approved. A sponsor is not needed for Adroit.
Faculty Member Sponsors a User
A faculty member with an already approved project on one of Princeton's clusters can sponsor additional users on that cluster by sending a request to firstname.lastname@example.org. Requests must come directly from the faculty member and not from members their research group. If the new account should have access to certain group directories or other resources then please state that in the request.
We can only provide accounts for individuals with a Princeton NetID. Any non-Princeton user must be sponsored by a Princeton faculty or staff member for a Research Computer User (RCU) account.
If a research group wants to start working on a cluster, a faculty member can submit a brief proposal. Proposals for the large cluster systems should be submitted as PDF or MS Word documents not to exceed 3 pages.
The proposal, which can be submitted through an online form, should include:
- Which system or systems you need to use (e.g., Della)
- A list of researchers who will need accounts
- The faculty member(s) who is sponsoring the project
- The scientific background for your project including scientific merit of the proposed work
- The programming approach for your project:
- Programming language(s)
- Parallelization mechanism (e.g., MPI or OpenMP)
- Required libraries
- The resource requirements for your project:
- Number of concurrent CPU-cores
- Total CPU time
- RAM per task
- Total disk space
- A few references or citations
- Faculty Member Contributes to the Purchase or Expansion of a Cluster
There are no fees to use the existing cluster systems; however, faculty can contribute to the purchase of our systems. By contributing to a system, a faculty member guarantees that averaged over some time period (two weeks or 30 days depending on the system) members of that group will be able to utilize the proportion of the system corresponding to the contribution. Unused system capacity is available to all other researchers on the system.
If you are interested in contributing to the purchase of a new cluster or an expansion of an existing system, please contact email@example.com.
Access to Tigressdata is automatically granted with access to the larger systems, such as Della, or Tiger.
There is no charge for storage. The large HPC clusters provide a variety of filesystems for storage. Nobel uses the OIT home filesystem. Adroit has its own filesystem.