NUS Volta Cluster user guide covering access, hardware and storage resources, PBS job scheduling, Python package installation, and recommendations.
The Volta Cluster Pilot User Guide, dated February 2019, explains how users access the NUS HPC Volta cluster and describes its hardware and storage resources. It covers login methods, queues and the PBS job scheduler, short interactive and batch jobs, multi-GPU scaling, and Python package installation.
The guide also lists recommendations such as maximum and default walltimes, model checkpointing, copying large datasets to /scratch, and saving weights to /hpctmp. It ends with additional support and consultation contacts.
| Volta cluster nodes | 5 nodes |
|---|---|
| GPUs per node | 4 x Nvidia Tesla V100-32GB |
| CPU cores per node | 20 CPU cores/node |
| RAM per node | 375GB |
| Maximum walltime | 48 hours |
| Default walltime | 24 hours |
| Home directory quota | 20 GB |
| Working directory quota | 500 GB |
| Working directory retention | Files older than 60 days are purged automatically |
Connect to Web VPN first, then use SSH to log in to an NUS HPC login node.
The maximum walltime is 48 hours.
Files older than 60 days are purged automatically.
The Volta cluster has 5 nodes, each with 4 Nvidia Tesla V100-32GB GPUs.