The Secure Enclave Service (SES) provides GPU resources to support computational workloads for restricted data within the university research community. These GPU nodes are hosted in a dedicated cluster, within the SES infrastructure, ensuring secure and efficient access for research computing needs.
Configuration
The SES GPU cluster consists of five (5) HPE DL385 Gen10 Plus nodes with the following specifications:
- Processors: Dual AMD EPYC 7542 32-Core CPUs
- Memory: 1024GB RAM
- GPUs: Two (2) NVIDIA A40 GPUs per node (each GPU includes 48GB GPU RAM and 10,752 CUDA cores)
- Networking: Dual-port 100GB NVIDIA (Mellanox) Ethernet, configured for redundant paths
Virtual GPU (vGPU) Allocation
GPU access is managed through an NVIDIA licensed server, providing flexible vGPU allocations to meet diverse research requirements:
- vGPUs can be allocated in 1/4, 1/2, or full (whole) increments.
- Access is time-sliced, meaning customers have full use of the allocated memory fraction and all GPU cores.
- vGPUs can be configured for either computation or rendering (display), but not both simultaneously (this is a static configuration).
- A single Principal Investigator (PI) may be allocated a maximum of 2 whole GPU equivalents.
- NVLink (bonding multiple GPUs/vGPUs) is not supported.
Cost
The cost of using a GPU is $3.00 per GB of RAM per month. With respect to the allocation sizes, rates are as follows:
| Service Option | Rates |
|---|---|
| 12 GB of GPU RAM | $36.00 per month |
| 24 GB of GPU RAM | $72.00 per month |
| 48 GB of GPU RAM | $144.00 per month |
Impact of Data Classification on GPU Allocation
SES manages multiple data classifications by assigning separate tenant environments to ensure data segregation. Consequently, GPUs are allocated to the tenant hosting a specific data classification. In cases involving Controlled Unclassified Information (CUI), a GPU may be dedicated to a single-occupant tenant, eliminating the possibility of GPU failover to a standby resource. However, for multi-occupant tenants, such as the HIPAA-aligned general GPU tenant, a dedicated GPU is reserved for failover, ensuring continued availability in case of failure.
Please contact the SES support team for further information or to request access.

