Delta GPU is a GPU-focused high-performance computing resource operated by NCSA and allocated through ACCESS. The system is designed for GPU-accelerated and hybrid CPU-GPU workloads, including AI/ML training and inference, data analytics, scientific simulations, and applications transitioning from CPU-only execution to accelerated computing.
Delta provides multiple GPU node types, including NVIDIA A40, NVIDIA A100, NVIDIA H200, and AMD MI100/MI210 resources. The system also includes high-performance node-local SSD storage, shared project and work file systems, and access through traditional command-line Slurm workflows as well as Open OnDemand.
Jobs
Jobs on Delta GPU resources are submitted through the Slurm scheduler. Users can run batch jobs with sbatch, launch interactive work with srun or salloc, or access compute resources through Open OnDemand applications such as JupyterLab, VS Code Code Server, and noVNC Desktop.
For instructions and Delta-specific examples, see the NCSA Delta Batch Jobs documentation for sbatch and Interactive Jobs documentation for srun and salloc.
Common Slurm commands
| Command | Description |
|---|---|
squeue -u $USER | Displays the status of your queued and running jobs. |
scontrol show job JobID | Displays detailed information about a specific job. Replace JobID with the job's ID. |
sinfo -a | Displays summary information about all Delta partitions and their node states. |
scancel JobID | Cancels a queued job or terminates a running job. Replace JobID with the job's ID. |
Delta GPU jobs run in GPU-specific Slurm partitions for NVIDIA A40, NVIDIA A100, NVIDIA H200, and AMD MI100 resources. Standard GPU partitions are intended for production workloads, interactive partitions are intended for development, debugging, testing, and short interactive sessions, and preemptible partitions are available for flexible jobs that can tolerate interruption.
Most Delta GPU production partitions allow jobs up to 48 hours, while interactive partitions generally have a 1-hour maximum duration. Node sharing is the default behavior unless users request exclusive access by specifying the required consumable resources or using Slurm options such as --exclusive --mem=0.
Preemptible queues should be used only for jobs that include checkpointing or can safely restart, because preempted jobs may lose unsaved progress. Delta charges jobs based on the resources reserved for the job, and CPU and GPU allocations use separate local charge accounts. GPU jobs should therefore be submitted using the appropriate Delta GPU account.
For more information about running jobs on Delta, see the NCSA Delta Running Jobs documentation. Example scripts for single-GPU, PyTorch, Job arrays, and other job types are available in the NCSA Delta Sample Scripts section.
Queue specifications
Metrics updated 2026-06-16
| Name | Purpose | Nodes | CPU cores / node | GPUs / node | Node RAM | Jobs
30 days
|
Wait Time
30-day trend
|
Wall Time
30-day trend
|
|---|---|---|---|---|---|---|---|---|
| gpuA100x4 | Default A100 queue with 4 GPUs per node for standard GPU jobs | 100 | AMD Milan (64 cores) | 4 NVIDIA A100 (40 GB vRAM) | 256 GB | 104,480 |
|
|
| gpuA100x4-interactive | Interactive A100 sessions for developing, debugging, and testing; max 4 nodes per job | 100 | AMD Milan (64 cores) | 4 NVIDIA A100 (40 GB vRAM) | 255 GB | 51,498 |
|
|
| gpuA100x4-preempt | Preemptible A100 queue for checkpointing-capable, flexible jobs | 100 | AMD Milan (64 cores) | 4 NVIDIA A100 (40 GB vRAM) | 256 GB | 193 |
|
|
| gpuA100x8 | Large-scale multi-GPU workloads using 8 A100 GPUs | 6 | AMD Milan (128 cores) | 8 NVIDIA A100 (40 GB vRAM) | 2048 GB | 4,984 |
|
|
| gpuA100x8-interactive | Interactive large-scale A100 jobs for development/testing; max 2 nodes per job | 6 | AMD Milan (128 cores) | 8 NVIDIA A100 (40 GB vRAM) | 2048 GB | 580 |
|
|
| gpuA40x4 | Moderate GPU workloads | 100 | AMD Milan (64 cores) | 4 NVIDIA A40 (48 GB vRAM) | 256 GB | 58,503 |
|
|
| gpuA40x4-interactive | Interactive A40 usage for testing and development; max 4 nodes per job | 100 | AMD Milan (64 cores) | 4 NVIDIA A40 (48 GB vRAM) | 256 GB | 28,628 |
|
|
| gpuA40x4-preempt | Preemptible A40 jobs for flexible, lower-priority workloads | 100 | AMD Milan (64 cores) | 4 NVIDIA A40 (48 GB vRAM) | 256 GB | 4,571 |
|
|
| gpuH200x8 | High-end GPU queue for memory-intensive workloads; max 1 node per job | 8 | Intel Xeon Platinum 8558 (96 cores) | 8 NVIDIA H200 (141 GB vRAM) | 2048 GB | 6,439 |
|
|
| gpuH200x8-interactive | Interactive H200 sessions for development and tuning; max 1 node per job | 8 | Intel Xeon Platinum 8558 (96 cores) | 8 NVIDIA H200 (141 GB vRAM) | 2048 GB | 3,097 |
|
|
| gpuMI100x8 | AMD GPU workloads optimized for MI100 architecture | 1 | AMD Milan (128 cores) | 9 AMD MI100, AMD MI210 (32 GB vRAM) | 2048 GB | 239 |
|
|
| gpuMI100x8-interactive | Interactive AMD GPU sessions for development/testing | 1 | AMD Milan (128 cores) | 9 AMD MI100, AMD MI210 (32 GB vRAM) | 2048 GB | 32 | 0 gpuMI100x8-interactive wait time: 0 hours |
|
Datasets
| Name | Description |
|---|---|
| ColabFold | ColabFold predicts protein folding structures and complexes using MMseqs2, which gives faster performance than AlphaFold. The ColabFold databases are available on Delta at |
| ImageNet | ImageNet is an image database organized according to the WordNet hierarchy, where each node of the hierarchy is depicted by hundreds to thousands of images. ImageNet database(s), version ILSVRC 2012–2017, are available on Delta at |
| alphafold | Protein structure prediction datasets used for bioinformatics research with AlphaFold. |
| models(ollama) | Pretrained large language models used with Ollama for local inference and AI experimentation. |
Storage
File System
| Directory | Path | Quota | Purge | Backup | Notes |
|---|---|---|---|---|---|
| HOME | /u | 100 GB, 750,000 files | Not purged | Yes (30 days) | Area for software, scripts, job files, and so on. Not intended as a source/destination for I/O during jobs. |
| PROJECTS | /projects | 500 GB | Not purged | No/TBA | Area for shared data for a project, common data sets, software, results, and so on. |
| WORK-HDD | /work/hdd | 1 TB | Not purged | No | Area for computation, largest allocations, where I/O from jobs should occur. (this is now your scratch volume) |
| WORK-NVME | /work/nvme | Not purged | No | Area for computation, NVME is best for lots of small I/O from jobs. (NVME space is available upon request.) | |
| TMP | /tmp | 1536 GB | Purged after each job | No | Locally attached disk for fast small file I/O |
File Transfer
Delta supports several methods for moving data to and from the system. Users may use scp or rsync for small to modest transfers, while Globus is recommended for large data transfers. GUI applications that use SSH, SCP, or SFTP must support Duo multi-factor authentication.
Files can also be shared externally from the /projects file system using Globus. To share data, users should create a sharing directory under their project space, move or copy the data into that directory, and connect to the “ACCESS Delta” Globus endpoint to create the shared collection.
The Delta Globus endpoint provides access to Delta home directories and shared project/work file systems, including /work/hdd and /work/nvme.
For more information, see the NCSA Delta File Sharing and Transferring Data documentation.
| Supported Methods | Data Transfer Node | URL |
|---|---|---|
| GLOBUS | RECOMMENDED | https://app.globus.org/ | |
| SCP | dt-login[01-04].delta.ncsa.illinois.edu | |
| RSYNC | dt-login[01-04].delta.ncsa.illinois.edu |
Login to Delta GPU
2FA/MFA required for login
After you have been added to an active Delta allocation, NCSA will send you an email with your NCSA username and instructions for completing your NCSA account setup. Please check your spam or junk folder if you do not receive the email. Follow the instructions in that email to set your NCSA password and register for NCSA Duo multi-factor authentication.
You can also find your assigned NCSA username in your ACCESS Profile. After logging in, scroll to the Resource Provider Site Usernames table near the bottom of the page. To set or reset your NCSA password, use the NCSA Identity password reset page. Follow the NCSA Duo setup instructions to enroll a device and save your backup codes. NCSA Duo is separate from the University of Illinois Duo service.
Use of SSH key pairs is disabled for general Delta access. Most individual users, including principal investigators, cannot use an SSH key pair in place of password and Duo authentication. The only exception is for principal investigators with a Gateway allocation. If Gateway account key-pair access is required, submit a support request to NCSA.
Direct access to the Delta login nodes is available via SSH. To log in, use one of the SSH Login commands below. Users must enter their NCSA username and password and complete NCSA Duo multi-factor authentication. Delta provides four login nodes, labeled 01 through 04.
In addition to SSH access, Delta provides access through Open OnDemand, which offers a browser-based interface and interactive applications. For maintaining persistent login sessions, the tmux multiplexer is available on all Delta login nodes. This utility allows users to run multiple programs within a single terminal, detach sessions while keeping processes running, and later reattach from a different terminal.
For more information about accessing Delta, see the NCSA Delta Login Methods documentation.
SSH Login
$ ssh <your_username>@login.delta.ncsa.illinois.edu