This document introduces how to use NUS HPC for deep learning training.
This December 2018 guide explains how to access NUS HPC login nodes and use the available CPU and GPU resources. It describes storage directories, the PBS job scheduler, and job script templates for GPU and Atlas8 CPU queues.
It also includes steps for submitting jobs, checking log files, installing Python packages, and contacting additional support.
| GPU cluster configuration | 5 nodes x 4 Nvidia Tesla V100-32GB |
|---|---|
| Azure Data Science VM GPU | Nvidia Tesla P100-16GB |
| Home directory quota | 20 GB |
| Working directory purge age | 60 days |
| Atlas8 cluster CPU cores | 24 CPU cores |
Log in via ssh to one of the listed NUS HPC login nodes, such as atlas8-c01.nus.edu.sg. If you are connecting from outside the NUS network, connect to Web VPN first.
The guide lists 5 nodes with 4 Nvidia Tesla V100-32GB GPUs each. It also lists a Microsoft Azure Data Science VM instance with an Nvidia Tesla P100-16GB.
The home directory has a global disk quota of 20 GB and snapshot backup.
Files in /hpctmp older than 60 days are purged automatically.