Model for HPC Hardware Support

The High Performance Computing Core Facility (HPC@UCD) seeks to provide the best possible support for HPC on campus. To ensure that we can continue to support the campus most efficiently, we are pleased to announce the creation of a new campus-wide cluster, Hive. A centrally managed cluster, with standardized hardware and connectivity and a defined support lifecycle, facilitates greater access to HPC for a larger number of users on campus while maintaining support for college-level needs. HPC@UCD offers two tiers of support moving forward and incentives to merge existing hardware with the new cluster when possible. Below, we describe our model, including a description of two tiers of support – a campus-funded Priority Tier for hardware on Hive, and a PI-supported Maintenance Tier for hardware not on Hive nor under maintenance contract.  Details about HPC@UCD Priority (Tier 1) and Maintenance (Tier 2) support can be found at https://hpc.ucdavis.edu/supported-services

Existing HPC Hardware

We will merge HPC hardware that is less than 5 years old into the Hive cluster. Clients will receive priority support as outlined above until the hardware reaches 5 years of age. Clients can opt to maintain this hardware, paying the rack fees and purchasing Maintenance support from HPC@UCD at an hourly rate for up to 2 additional years.

Existing hardware older than 7 years will not be supported. Hardware in HPC@UCD racks will be decommissioned. Should the Client wish to retain possession of decommissioned hardware outside of HPC@UCD, they may do so at their expense, with the approval of their Dean’s office and the college IT.

New Purchases

Starting January 2025, new purchases of HPC hardware will only be supported if approved by the High-Performance Computing Core Facility (HPC@UCD). Most new purchases will be expected to be part of the university cluster Hive. Hive will function under the “condo” model, where a client (such as a PI or a unit such as a college or center) can purchase any number of cores for 5 years. During those 5 years, HPC@UCD will maintain hardware and provide priority support for users in the client’s group.

After 5 years, HPC@UCD will no longer support the hardware. The client can maintain the hardware with maintenance support at the established labor rate for up to two additional years. The client will be expected to pay for rack fees and HPC@UCD-approved supplies necessary for physical maintenance. See here for current rates and fees.

After 7 years, HPC@UCD will no longer offer hardware support. 

Hardware that is not supported will be decommissioned and removed from HPC@UCD racks at the discretion of HPC@UCD. Should the Client wish to retain possession of their hardware outside of HPC@UCD, they may do so at their expense, with the approval of their Dean’s office and the college IT department.

Buyout Proposal

Under coordination with the START Research Computing task force, HPC@UCD has put together a proposal to campus to buy out some portion of existing hardware older than 5 years and replace it with new hardware on Hive under the model above. Funds from this proposal would be managed by each unit’s IT in conjunction with HPC@UCD.