Computing Infrastructure
The Neuroscience Institute (NI) and the Center for the Neural Basis of Cognition (CNBC) maintain a dedicated research computing environment that supports faculty, staff, students, and collaborative research groups across Carnegie Mellon University. The infrastructure provides secure enterprise storage, high-performance computing (HPC), advanced networking, and a comprehensive scientific software ecosystem supporting neuroscience, neuroimaging, artificial intelligence, machine learning, and data-intensive research.
The environment is a shared institutional resource supporting individual laboratories, multi-investigator collaborations, training grants, and large-scale research initiatives across the university. It is designed to provide both high reliability for long-term data management and high-performance resources for computationally intensive research.
Infrastructure at a Glance
| Resource | Capacity |
|---|---|
| Enterprise Research Storage | 500+ TB |
| Dedicated HPC Cluster Storage | 440 TB |
| Total Managed Storage | ~940 TB |
| Compute Nodes | 14 |
| GPU-Enabled Nodes | 11 |
| CPU-Only Nodes | 3 |
| Logical CPU Cores | 1,056 |
| Aggregate System Memory | 7.38 TB |
| NVIDIA GPUs | 64 |
| Aggregate GPU Memory | ~2.2 TB |
| High-Speed Network | 100 Gbps InfiniBand |
Enterprise Research Storage
Research data are housed in a dedicated, climate-controlled machine room within the Mellon Institute. The facility is protected by redundant power systems and uninterruptible power supplies (UPS) to ensure high availability and data integrity.
The enterprise storage environment provides more than 500 TB of protected research storage supporting laboratory data, shared research projects, institutional collaborations, and long-term data preservation.
Two primary storage systems support these needs:
- Raptor provides approximately 260 TB of protected storage supporting core research activities and large collaborative projects, including major federally funded initiatives.
- Condor provides approximately 240 TB of protected storage designed for multi-investigator laboratories and shared datasets.
Both systems use enterprise-grade storage infrastructure with redundant arrays, automated snapshots, and continuous replication to dedicated backup systems. Snapshot schedules include hourly, daily, weekly, monthly, and annual recovery points, ensuring strong protection against accidental deletion, corruption, and hardware failure. Access is provided securely via SSH/SFTP and SMB services for authorized users.
High-Performance Computing
Computational research is supported by the MIND Cluster, a dedicated high-performance computing environment located in the School of Computer Science data center. The system is professionally managed and continuously monitored by SCS system administrators in collaboration with NI computing staff.
The cluster consists of:
- 14 compute nodes total
- 11 GPU-enabled nodes
- 3 CPU-only compute nodes
- 1,056 logical CPU cores
- 7.38 TB of aggregate system memory
- 64 NVIDIA GPUs
- 100 Gbps InfiniBand high-speed interconnect
The cluster supports computational neuroscience, neuroimaging, connectomics, artificial intelligence, machine learning, statistical analysis, simulation, and large-scale image processing.
Cluster Storage
In addition to central enterprise storage, the MIND Cluster includes dedicated high-performance storage systems totaling approximately 440 TB. These storage systems support active computation and data-intensive workflows, including home directories, laboratory data, and shared research datasets.
This cluster storage environment provides fast, low-latency access to large datasets required for high-throughput computing. It is fully integrated into the computational environment and is also protected through snapshot-based recovery and replication to backup storage systems. This ensures both performance for active computation and reliability for research data protection.
GPU Resources
The MIND Cluster provides 64 NVIDIA GPUs spanning multiple hardware generations. This heterogeneous environment allows researchers to match computational workloads with appropriate hardware ranging from legacy pipelines to state-of-the-art AI systems.
| GPU Model | Qty | Architecture | Release | Memory |
|---|---|---|---|---|
| NVIDIA RTX PRO 6000 Blackwell | 10 | Blackwell | 2025 | 96 GB GDDR7 ECC |
| NVIDIA L40S | 18 | Ada Lovelace | 2023 | 48 GB GDDR6 ECC |
| NVIDIA RTX A5000 | 4 | Ampere | 2021 | 24 GB GDDR6 ECC |
| NVIDIA RTX 3090 | 2 | Ampere | 2020 | 24 GB GDDR6X |
| NVIDIA TITAN RTX | 10 | Turing | 2018 | 24 GB GDDR6 |
| NVIDIA GeForce RTX 2080 Ti | 16 | Turing | 2018 | 11 GB GDDR6 |
| NVIDIA TITAN Xp | 4 | Pascal | 2017 | 12 GB GDDR5X |
The inclusion of next-generation Blackwell and Ada Lovelace GPUs alongside prior-generation accelerators provides both cutting-edge AI capability and broad support for established neuroimaging and scientific computing workflows.
Software Environment
The computing environment provides a centrally managed scientific software stack supporting neuroscience, machine learning, and data science research. Key software includes AFNI, FreeSurfer, MNE-Python, SPM, MATLAB, R, DSI Studio, TensorFlow, PyTorch, Python, CUDA, and additional specialized tools.
The system supports reproducible research workflows through Conda environments and Apptainer (Singularity-compatible) containers. Workload scheduling and resource management are handled using SLURM, enabling efficient and fair sharing of cluster resources across users and projects.
Operations and User Support
The Computing Center operates as a 100% recharge facility. Operational costs are supported through user fees, while major hardware acquisitions and infrastructure expansions are funded primarily through faculty grants and startup resources. This model enables continuous modernization of the computing environment while ensuring long-term sustainability of shared research infrastructure.
The infrastructure is overseen by the NI Manager of Computing, who is responsible for system administration, user support, documentation, software deployment, and technical consulting. Users receive assistance with data management, computational workflows, software environments, performance optimization, and effective use of shared resources.
Summary
The NI and CNBC computing infrastructure provides a fully integrated research computing environment consisting of more than 940 TB of managed storage, a 14-node high-performance computing cluster, 1,056 CPU cores, 7.38 TB of system memory, 64 NVIDIA GPUs representing approximately 2.2 TB of GPU memory, and a 100 Gbps InfiniBand network fabric.
Together with enterprise storage systems, high-performance cluster storage, a modern scientific software ecosystem, and dedicated user support, this infrastructure enables large-scale computational neuroscience, artificial intelligence, neuroimaging, and data-intensive research across Carnegie Mellon University.