We’re excited to announce a new GPU queue designed specifically for interactive development and quick prototyping work: ic-express.
What is ic-express?
The ic-express partition provides fast, responsive access to H100 GPU resources for short-duration, interactive tasks. This queue is ideal for:
- Running Jupyter notebooks
- Model prototyping and tuning
- Exploratory development workflows
- Quick testing and debugging
Key Features:
- GPU Allocation: One Nvidia H100 MIG slice per user (1/7th of an Nvidia H100 with 10GB GPU memory)
- Hardware: Single node with 4x H100 GPUs, supporting up to 28 concurrent users
- Walltime: 30-minute default, 8-hour maximum
- Expected Wait Times: Typically <5 minutes (designed for fast access)
- Job Limit: One job per user at a time to ensure fair access
When to Use ic-express:
✓ Interactive development, testing, Jupyter notebooks
✗ NOT for: Long training runs, production jobs (use standard IllinoisComputes-GPU queue instead)
How to Use:
Option 1: Interactive Session For interactive work, use:
srun –nodes=1 –partition=ic-express –account=yourusername-ic
–time=04:00:00 –gres=gpu:1 –pty bash
(Replace ‘yourusername’ with your NetID or account name, e.g., jdoe-ic)
Option 2: Via Slurm Script To request a GPU in this partition, use the following in your Slurm script:
#SBATCH –partition=ic-express
#SBATCH –account=yourusername-ic # Replace ‘yourusername’ with your NetID
#SBATCH –gres=gpu:1
#SBATCH –time=04:00:00 # Adjust as needed, max 8 hours
Option 3: Via Campus Cluster Open OnDemand You can also access ic-express through the Campus Cluster Open OnDemand interface at https://cc-ood.campuscluster.illinois.edu/. When launching interactive apps (such as Jupyter), use:
- Partition: ic-express
- Account: yourusername-ic
Who Can Access This Queue?
If you currently have access to the IllinoisComputes or IllinoisComputes-GPU queues, you automatically have access to ic-express. No additional request is needed.
If you do not yet have access to Illinois Computes resources, please apply here: https://computes.illinois.edu/submitrequest/
Quick FAQ:
Q: Can I use this for overnight training runs? A: No, use the standard GPU queue for jobs >8 hours
This new partition complements our existing resources by providing a dedicated space for interactive work without competing with longer production jobs. We hope this improves your development experience and accelerates your research workflows.
If you have questions about using ic-express, please don’t hesitate to reach out to our support team at help@campuscluster.illinois.edu
Happy computing!
