Condo Cluster Program Purchase Request Model
The by-request system purchase model is a useful option when research groups need to make a bulk purchase using a research grant or departmental budget, and it is also useful for grant proposal budget planning purposes.
Like the annual subscription model, research groups will have exclusive access to the resources they purchase. Principal Investigators (PIs) only need to purchase compute nodes (CPUs and/or GPUs), cables, and a 5-year manufacturer’s warranty service. Other equipment and services, including racks, network switches, power and cooling, system support, and machine room operation in our data center, is paid for by the university.
Previously, CCP purchases could only be made once a year during a limited timeframe. We are no longer limiting when PIs can request purchases. A purchase request can be submitted to the CARC team at any time of the year. Additionally, CARC is no longer placing the order on PI’s behalf. We source the best quote and your business administrator pays the vendor directly.
1 Compute resources available for purchase
CARC can source quotes for compute resources on your behalf for your research needs. The most up-to-date CPU models are available for purchase. Most CPU models have dual processors, 32 or 64 cores, and can be configured with many different GPU combinations.
To ensure competitive pricing, high-quality service, and an efficient procurement process, CARC will only work with a select group of trusted vendors. Because we have established strong rapport with leading computing manufacturers, computing resources purchased from outside vendors will not be accepted.
Email us at carc-condo@usc.edu and we will work to accommodate your needs and budget.
2 Pricing
Pricing may vary based on market condition and CARC’s equipment purchase price. In addition to the cost of the selected CPU and/or GPU nodes, PIs will be charged for cables, chassis (if necessary), and sales tax. Pricing does not include interconnect hardware, secondary enclosure (racks), or utility costs for power and cooling. ITS/CARC will cover the costs of this necessary service and equipment. The PI will not be charged any setup or recurring fees as part of this arrangement for the duration of this agreement, including CARC administrative support costs.
3 How to request a condo purchase
Previously, CCP purchases could only be made once a year during a limited timeframe. We are no longer limiting when PIs can request purchases. A purchase request can be submitted to the CARC team at any time of the year. Additionally, CARC is no longer placing the order on PI’s behalf. We source the best quote and your business administrator pays the vendor directly.
A condo purchase request can be made at any time. The following outlines the steps for requesting a condo purchase:
- Email the CARC team at carc-condo@usc.edu to start the process.
- A member of the CARC team will help you find the right node configuration for your research needs.
- The CARC team contacts our select group of trusted vendors to obtain a competitively-priced quote.
- Your department’s business administrator is responsible for using the quote for the procurement process and paying the vendor directly. Nodes must be shipped directly to the Carol Little (CAL) Building for setup
- Once the nodes arrive, the CARC team handles system installation, deployment, and maintenance.
The whole process of purchase request, system configuration, vendor selection, purchasing, system installation, and deployment typically takes three months. The purchased resources will be in service for 5 years, and after this service period, CARC decommissions these condo nodes.
4 Suggested compute resources
The following tables summarize some suggested CPU and GPU node configurations. If your group requires different configurations, we will work to accommodate your needs.
4.1 CPU nodes
| CPU | CPU quantity | Memory |
|---|---|---|
| Dual AMD EPYC 9555 64-core 3.20GHz | 2 | 384 GB, 768 GB, or 1.5 TB |
| Dual AMD EPYC 9355 32-core 3.55GHz | 2 | 384 GB, 768 GB, or 1.5 TB |
4.2 GPU nodes
| CPU | CPU quantity | Memory | GPU | GPU quantity |
|---|---|---|---|---|
| Dual AMD EPYC 9355 Processor 32-core 3.55GHz | 2 | 768 GB | L40S | 2, 4, or 8 |
| Dual AMD EPYC 9355 Processor 32-core 3.55GHz | 2 | 768 GB | H200 | 2, 4, or 8 |
Final configurations of system specs will depend on system availability from the vendor.