Welcome to pasteur
The CASL High-Performance Computing cluster — built for large-scale GPU and CPU workloads in research computing. Managed by Slurm, shared across the Complex Adaptive Systems Lab.
Cluster Specifications
pasteur is a single bare-metal compute node backed by a KVM-based login VM (pasteur-login) on the university network.
CPU Cores
384
Logical processors
GPUs
8×
NVIDIA H200
RAM
~2.2 TB
System memory
Storage
14 TB
/data — NFS shared
OS
Ubuntu
24.04 LTS (Noble)
Scheduler
Slurm
with slurmdbd + MariaDB
Cluster Architecture
Users log in through pasteur-login, a lightweight VM on the university network. Jobs are submitted via Slurm and run on the compute node pasteur. All storage under /data is shared via NFS.
Network Topology
[ You ]
──── SSH ────
pasteur-login
(KVM VM · university network)
│
└── Slurm / NFS ────
pasteur
(bare-metal · 8×H200 · 384 CPUs)
│
└── NFS exports ────
/data/{home,apps,scratch,shared}
Documentation
Getting Started
SSH access, first login, and environment setup
Storage Layout
Home, scratch, apps, shared dirs and quotas
Slurm Scheduling
Submitting jobs, partitions, QOS levels
GPU Jobs
Requesting GPUs, multi-GPU QOS, isolation
Software Stack
Conda/Mamba, Apptainer containers
Key Policies
Scratch is not backed up
Files in
/data/scratch/casl/<username> are temporary. Do not store important results there — move completed outputs to your home directory or shared project storage.
Default GPU limit: 1 per job
By default, each user can request at most 1 GPU per job. To request multiple GPUs, you need elevated QOS. Contact an admin if your workload requires it.
Do not run compute jobs on pasteur-login
The login node is a shared VM — running intensive workloads there affects everyone. Always submit jobs through Slurm on the compute node.