Allocations

Overview of Allocations for the HPC clusters

You may submit allocation requests through the CHPC Portal for allocations on all CHPC clusters. You can log in by clicking the "Login" button on the CHPC Portal, then choosing "University of Utah" as the Identity Provider and clicking "Log On" on the CILogon page.

Submit an allocation request on the CHPC Portal

CHPC awards allocations on the general CPU resources of the notchpeak and redwood clusters based on allocation request proposals. Allocations are required for CPU and GPU resources on the granite cluster (as of January 1, 2025). An allocation applies to all allocated clusters within an environment: Allocation in the General Environment can be used on both notchpeak and granite, and there is no need to apply separately for each cluster. Allocations are awarded to CHPC PIs, and access to the allocation is made available to all members of the PI's group. Jobs from groups with allocation run at a higher priority and can preempt jobs running without allocation on CHPC-owned (general) resources. Note that the kingspeak and lonepeak general nodes are run in an unallocated manner and are available for use by all CHPC account holders.

In addition, the general GPU resources on notchpeak, kingspeak, and lonepeak are also run in an unallocated manner, and all CHPC users have access. Please note that the GPU resources should only be used for jobs that are making use of the GPUs. On the granite cluster, jobs on general nodes with GPUs are charged GPU hours, even if they do not use GPUs. Jobs that do not use GPUs should be run on CPU-only nodes.

All allocation requests must be submitted by a CHPC PI or designated delegate. To specify a delegate, the CHPC PI should email helpdesk@chpc.utah.edu and provide the name and uNID of the user they wish to have allocation delegate rights. 

Other resources on the clusters are "owned" by other PIs and your jobs will not be able to run on those nodes at priority unless you are a member of that group. You may run on the "owner nodes" by changing the SBATCH directives for the account and partition as described on the Slurm documentation page. Your job can thus be run on any idle owner nodes, at low priority, but your job can be preempted (killed) if a member of the owner's group submits a job.

Allocations are measured in wall clock CPU core or GPU hours, which are 1 hour of wall clock time for each CPU core or GPU allocated, respectively. In other words, the unit of an allocation is the product of the number of resources consumed (CPU cores or GPUs) and the time for which they were consumed. For more details, see our HPC Allocation Policy. Allocations are managed on a calendar quarter and any allocation remaining at the end of a given allocation quarter is not carried over to the next quarter.

Note that allocation of the GPUs includes the associated CPUs on that GPU node, so one does not need to request CPU allocation when only requesting GPU allocation.

There are two kinds of allocation requests:

• Small allocation request (up to 30,000 CPU core hours and/or 300 GPU hours for up to upcoming 4 quarters, plus the current quarter)
• Regular allocation request (submitted at most quarterly for up to 4 quarters at a time)

Allocations are not cluster-specific; they are shared across the allocated resources. That is, a single allocation is used for both notchpeak and granite.

Small Allocation Process

Small allocations are for PIs who are new to having a CHPC allocation, or don't need a large amount of resources. PIs may submit a small allocation request through the CHPC Portal for up to 30,000 CPU core hours and/or 300 GPU hours per quarter, up to four quarters plus the current quarter. Small allocations are immediately reviewed by CHPC staff and awarded at CHPC's discretion. Note that every allocation request that is within the Small Allocation limit will be automatically considered as such.

Regular Allocation Process

You can submit a regular allocation request through the CHPC Portal. Requests for allocations are accepted throughout the year, with deadlines according to the following schedule:

• December 1 for allocation for the period of January 1 through March 31
• March 1 for allocations for the period of April 1 through June 30
• June 1 for allocations for the period of July 1 through September 30
• September 1 for allocations for the period of October 1 through December 31

The current maximum award is 300,000 CPU core hours per quarter. Additionally, GPU allocations on granite are currently limited to 3,000 GPU hours per quarter. This is currently independent of the GPU type, though we are exploring the possibility of weighting time on GPUs by specifications such as the global memory or throughput.

PIs are asked to submit a request for the four upcoming calendar quarters. Your allocation request will then need to be renewed the following year by submitting another request. You may ask for fewer quarters, but if you expect to be using our systems long term we request that you submit a full year request. If you find your allocation amount needs to be adjusted, you can submit a new request to replace your existing request at any allocation cycle during the year.

Please fill out the allocation request form completely! The committee particularly needs to see your justification for the number of core wall clock hours you have requested. If you are new to the allocation process, feel free to take a look at this Sample Application. It is good to have a track record using our systems via the small allocation. Be sure to list your sources of funding, including granting agency and grant/contract numbers. It is also important to let us know of publications resulting from work supported by CHPC. This information is critical to us as it helps us continue to support the computation research community. The more you succeed, the more we succeed. Our mission to to help you get your research done!

For more information, see the HPC Allocation Policy.