Skip to content

NVIDIA DGX Cloud information for Spring 2025

NVIDIA DGX Cloud icon During the spring 2025 semester, UFIT Research Computing is working on an exciting upgrade to HiPerGator, bringing both B200 and L4 GPUs to the UF community. During this time, the resources on HiPerGator are more constrained than they had been. To facilitate research and teaching uses, most courses using GPUs are being set up on a resource from the NVIDIA DGX Cloud.

NVIDIA DGX Cloud only!

This page provides resources for using the NVIDIA DGX Cloud, and is not intended for HiPerGator users!

Quick facts

Access to NVIDIA DGX Open onDemand

The URL for Open onDemand for the NVIDIA DGX Cloud is: https://dgxcloud-ood.rc.ufl.edu/ (requires UF network! )

  • This is only available from the UF Campus network. You must either be on the UF network or connected to the VPN.
  • Note that the UFHealth network is not included. Please use the VPN from UFHealth network locations.
  • Please see the UFIT VPN instructions here.

Only users who have had accounts created on the DGX Cloud will be able to login at this site.

Storage

Home storage

Each user has a small home folder at the path of /home/USERNAME.

Space is limited and cannot be increased. The ncdu command will scan the current folder and show space usage. Hidden folders (starting with a ".") are often the culprit in using up space.

Primary storage

The primary storage for users has the path /lustre/fs0/GROUP_NAME, where GROUP_NAME is typically the prefix (lower case) and number of your class. Classes with multiple numbers will usually be combined. E.g. the group for BSC 4892 and BSC 6895 is bsc4892. All students in either section will be in the same group.

To check on your group, you can use the id command:

albert@slogin002:~$ id
uid=9999(albert) gid=9999(albert) groups=9999(albert),2004(bsc4892)

The last group in the list here is the course group, bsc4892.

By default, Jupyter will start in your home directory (/home/USERNAME). There isn't an easy way to navigate to your group's /lustre/fs0/ folder. To fix this, create a symbolic link in your home directory pointing to the /lustre/fs0 folder. Copy and paste the following line

 GROUP=`id -G -n | awk '{print $NF}'`; ln -s /lustre/fs0/$GROUP ~/$GROUP
Expand to see the explanation of this command

The first part of this command, 

GROUP=`id -G -n | awk '{print $NF}'`;

uses the id -G -n command to get a list of all your groups. Then it uses awk to get the last group name in the list. This is the name of your class group.

The second part of the command,

ln -s /lustre/fs0/$GROUP ~/$GROUP

uses the the $GROUP variable defined above, to make a symbolic link from a from your home directory to the group's folder in /lustre/fs0/. The link is named with the group name.

Documentation

Most of the documentation on the UFIT Research Computing docs.rc.ufl.edu site is still applicable. The main differences are:

  • The URL is different (the NVIDIA DGX Cloud Open onDemand site is at https://dgxcloud-ood.rc.ufl.edu/).
  • You can only connect from the UF network (either on campus or the VPN).
    • Note that the UFHealth network is not included. Please use the VPN from UFHealth network locations.
  • The applications available are limited, mostly to Jupyter and a Desktop environment.
  • The primary storage is not /blue/, but /lustre/fs0/.
  • Quotas on /home/ and /lustre/fs0/ are smaller.

Additional NVIDIA DGX Cloud Specific Documentation Pages

Interacting with Slurm from the terminal

If you want to interact with Slurm from the terminal, you will need to load the slurm module (module load slurm). To have that automatically loaded at login, please run this command to add that to your ~/.bashrc file:

echo "module load gcc slurm" | tee -a ~/.bash_profile

You will need to logout and back in for this change to take effect.

Software

To request additional software/kernels be installed, course instructors please open a support request. Please ensure to mention this request is for the DGX cloud.

As a reminder, only the instructors should use the support system. Students should work with the faculty if they need help.

VSCode Remote Tunnel

VSCode Remote Tunnels have been tested and do work with the NVIDIA DGX CLoud. The documentation here should work, using a terminal in either a Jupyter session or Desktop.