Systems overview¶
CARC operates several high-performance computing clusters serving disciplines from traditional scientific computing to data analytics, artificial intelligence, and machine learning.
Compute clusters¶
Hardware inventory was collected directly from Slurm on September 29, 2026. Partition availability changes continuously; see resource limits for partition policies and limits.
| Cluster | All nodes | Scheduled compute nodes | CPU cores | GPUs | Compute-node RAM | Node interconnect | Switch fabric |
|---|---|---|---|---|---|---|---|
| Easley | 65 | 63 | 4,160 | 8 NVIDIA H100 + 36 NVIDIA L40S | 22.4 TB | ConnectX-7 at 200 Gbit/s | 800 Gbit/s |
| Hopper | 68 | 67 | 2,176 | 26 NVIDIA A100 + 2 NVIDIA V100 | 15.1 TB | ConnectX-4 at 100 Gbit/s | 400 Gbit/s |
The CPU totals include Easley's head and login nodes (128 additional cores) and Hopper's head node (32 additional cores). GPU and RAM totals describe the Slurm-scheduled compute nodes listed below.
All Easley nodes have 64-core Intel Xeon Gold 6438Y+ processors.
| Nodes | GPUs per node | RAM per node |
|---|---|---|
| 48 | None | 252 GB |
| 4 | 2 NVIDIA H100 | 1,008 GB |
| 9 | 4 NVIDIA L40S | 252 GB |
| 2 | None | 2,016 GB |
All Hopper nodes have 32-core Intel Xeon Gold 6226R processors.
| Nodes | GPUs per node | RAM per node |
|---|---|---|
| 43 | None | 92 GB |
| 6 | None | 504 GB |
| 2 | None | 1,512 GB |
| 1 | None | 376 GB |
| 10 | 2 NVIDIA A100 | 376 GB |
| 2 | 2 NVIDIA A100 | 188 GB |
| 2 | 1 NVIDIA A100 | 188 GB |
| 1 | 2 NVIDIA V100 | 188 GB |
For queue limits and walltimes, see resource limits. Historical specifications for retired systems (Wheeler, Taos, Gibbs, and Xena) are kept in the legacy cluster reference.
Storage¶
CARC operates several high-performance storage systems, portions of which are subdivided into allocations for PIs and research projects:
IBM Storage Scale (GPFS) and BeeGFS are parallel filesystems designed for high-performance computing workloads.
- 385 TB of all-flash IBM Storage Scale (GPFS) scratch
- 1.2 PB of BeeGFS working scratch
- 133 TB of NetApp enterprise storage, with automated hourly, daily, weekly, and monthly snapshots retained up to four months for user-directed recovery
See storage and backups for how home, project, and scratch spaces work, and resource limits for quotas.
In partnership with UNM Libraries, CARC also supports virtual machine infrastructure for custom research applications, secure data hosting, and flexible computing environments.
Web portals¶
| Portal | URL | Purpose |
|---|---|---|
| JupyterHub (Hopper) | hopper.alliance.unm.edu | Interactive notebooks on Hopper |
| JupyterHub (Easley) | easley.alliance.unm.edu/jupyter | Interactive notebooks on Easley |
| Open OnDemand | ood.alliance.unm.edu | Browser-based files, shells, and interactive apps |
| ColdFront | coldfront.alliance.unm.edu | Project, allocation, and publication management |
| XDMoD | xdmod.alliance.unm.edu | System usage metrics by PI |
| Help desk | support.alliance.unm.edu | Create or manage help tickets |
Status and downtime¶
Check the live monitors directly:
- Cluster login & website status — per-system up/down and response times
- UNM IT alerts — campus-wide IT outage notices
- Network performance — perfSONAR measurements for the CARC network
- Easley external DNS check — worldwide resolution of easley.alliance.unm.edu
- System usage (XDMoD) — utilization metrics by system and principal investigator
Networking¶
CARC systems connect to campus through multiple 10 Gbps links, including a dedicated 10 Gbps connection to UNM's Science DMZ research network. External connectivity includes 100 Gbps connections to ESnet and the Western Regional Network through the Albuquerque Gigapop.
Export control¶
The U.S. Government controls the export of sensitive equipment, software, and technology. Installation of export-controlled software on CARC systems requires prior written approval from UNM Export Control — see the Good Neighbor Use Policy.
Machine-readable versions of this page: Markdown twin · raw source on GitHub · llms.txt · llms-full.txt (whole site). See For AI agents.