Enhancing MSI Allocations for Expanded Computational Needs
Dedicated computing options are available for purchase when an MSI allocation is not enough to meet your computational needs. Multiple service configurations are available for a one-time fee.
Service Configurations and Cost
MSI staff are responsible for all hardware and software maintenance issues over the terms of the dedicated computing service. Systems included in this service are all connected via Infiniband and Ethernet, as well as MSI's primary storage. The OS image and software stack available on these dedicated computing options are the same as MSI's HPC Agate system. Dedicated computing services are subject to the same maintenance period as other MSI systems.
The following five dedicated computing configuration options on the Agate cluster are available for purchase. There is one additional option with older GPU hardware on the Mangi/Mesabi cluster. The smallest unit of service for purchase corresponds to a single compute node. Some older nodes with 2 x NVIDIA V100 GPUs may also be available, you may request one via the Form.
Configuration options and costs for FY24 are:
| CPU | Memory | GPUs | 1 Year Cost | 5 Year Cost | |
| Option 1 | 128 cores AMD | 512 GB | none | $2632 | $12269 |
| Option 2 | 128 cores AMD | 2048 GB | none | $4082 | $19519 |
| Option 3 | 64 cores AMD | 512 GB | 4 x NVIDIA A100 | $9321 | $45715 |
| Option 4 | 128 cores AMD | 1024 GB | 8 x NVIDIA A100 | $16320 | $80713 |
| Option 5 | 128 cores AMD | 512 GB | 8 x NVIDIA A40 | $10109 | $49659 |
Note: Dedicated Computing is exempt from Facilities and Administrative costs. The rationale for this approach follows a national trend toward incentivizing best practices related to computing and data storage. There may be delays in provisioning some solutions based on immediate supply.
Service Agreements
MSI will offer dedicated computing through two options. A one-year agreement and a five-year agreement.
1-Year Service
Researchers can purchase dedicated computer service, which is made available for a 1-year term. The service is made available using the same resource manager as other compute nodes at MSI. The purchased service can be accessed by any MSI user on a list determined by the PI. A common configuration will be to make the purchased service available to anyone in the PI group. The dedicated service would be available for batch-submitted jobs, as well as for interactive command-line access. The node(s) will be made available through the MSI scheduling system in its own partition. The partition (queue) can be configured to allow a maximum runtime of 30 days. Please note that jobs will only start if they can be completed before the next system reservation. The system reservation can be viewed with the command "scontrol show reservation". The node(s) will have the same cleanup scripts run between jobs to clean the scratch disk and users will not have sudo or admin privileges on the system. In the event of permanent system failure, the dedicated node will be replaced with a new one.
5-Year Service
The 5-year service option is similar to the 1-year, but lasts for 5 years. The annual cost of 5-year reservations is about 5% less, because the setup efforts are spread over 5 years.
Computer hardware at the University is depreciated over 5 years, so purchases that coincide with a new acquisition will generally remain unchanged for the duration of the service. If a PI purchases the 5-year service at a time that does not coincide with a new acquisition, the dedicated nodes that support this service will be supported by hardware of equal or greater cost after the initial hardware reaches the end of its service life. However, researchers will not be able to request changes in hardware type. MSI only plans to change hardware when it is no longer supported.
The equivalent hardware will be determined by cost, not performance, so the researchers can expect an increase in performance when the nodes are exchanged for new nodes partway through the 5-year term.
Purchasing
Log a purchase request for Dedicating Computing on our helpdesk.