As of January 2025, the RIT-HPC Compute Cluster is comprised of:
Software
- Base operating platform: AlmaLinux 8
- All compute nodes are virtual machines running under Openstack.
- Scheduler: Slurm
Compute
- 32,600 CPU threads:
- Intel(R) Xeon(R) Gold 5320 CPU @ 2.20GHz
- Xeon Gold 6140 CPU @ 2.30GHz
- Xeon E5-2670 v2 @ 2.50GHz
- 500 compute nodes:
- Each Compute Node is typically 128GB, 192GB or 256GB of RAM.
- Four large compute nodes with the following RAM configurations: 2x 768GB each, 1x 1.4TB, and 1x 4TB
- 16 x GPU Nodes (for general use):
- 8 x NVIDIA V100 32GB Tesla GPU, 19 cores, 180GB RAM, 1.2TB local scratch space
- 1 x NVIDIA P100 16GB Tesla GPU (2 cards), 30 cores, 120GB RAM, 250G local storage
- 4 x NVIDIA_L40S, 27 cores, 240GB RAM, 550G local storage space
- 2 x NVIDIA_H100_NVL (2 cards), 54 cores, 234GB RAM, 2.3T local storage space
- 1 x NVIDIA_H100_HGX (4 SXM cards with NVLINK), 124 cores, 960GB RAM, 2.3T local storage space
Network
- Interconnects: multi-bonded 40GE and 100GE
- Core: HPE 12908 IRF
- Top of Rack: Mellanox SX6036, Dell S4000, Dell S4100 and Dell S5200 Series
Storage
- Tier 1 storage – Powerscale nodes with 13PiB Total Space
- 5.2 PB of Flash
- 7.8 PB of Hybrid
- Cluster access to Tier 1 storage: NFSv3
- End user access to Tier 1 storage: SMB
- Archive storage – iRODS. 12PB usable with offsite replica