Jetstream2 GPU provides on-demand GPU-enabled virtual machines for AI, machine learning, visualization, GPU-accelerated analysis, and interactive research software environments.
GPU access is provided through Jetstream2 GPU flavors. Current official GPU flavors include partial and full GPU options using A100, L40S, and H100 hardware, with available vCPU, system RAM, GPU type, and GPU memory depending on the selected flavor.
L40S, H100, and multi-GPU A100 flavors are not available by default. Access depends on the allocation's GPU quota and may require a justified request to Jetstream2 support.
Jobs
Jetstream2 GPU is a cloud virtual-machine resource, not a centrally scheduled Slurm cluster. Standard Jetstream2 GPU use therefore has no shared login node, Slurm queue, partition, sbatch, srun, or squeue workflow. Slurm commands apply only if a project separately deploys and administers its own virtual cluster.
Users select an available g3.*, g4.*, or g5.* GPU flavor in Exosphere, create a VM, and connect through Web Shell, Web Desktop, or SSH. GPU availability depends on the project's allocation and quota, and some full-card or multi-GPU flavors require a justified support request.
Applications run directly inside the GPU VM. Users can verify the assigned NVIDIA GPU and driver with nvidia-smi and then start their application directly, for example python train.py. Software may be installed inside the VM using the operating system package manager, Python environments, containers, or the Jetstream2 software collection.
An active GPU instance consumes service units even when no GPU process is running. Users should shelve an instance when it will not be used; a shelved instance consumes zero compute SUs.
See the Jetstream2 GPU flavors, GPU memory, and hourly SU costs, Jetstream2 GPU access and troubleshooting instructions, instructions for installing and running software on a Jetstream2 VM, and instructions for shelving and managing an instance to conserve SUs.
Queue specifications
| Name | Purpose | Nodes | CPU cores / node | GPUs / node | Node RAM | Jobs
30 days
|
Wait Time
30-day trend
|
Wall Time
30-day trend
|
|---|---|---|---|---|---|---|---|---|
| Indiana Jetstream2 GPU | GPU-enabled virtual machine flavors; not a Slurm partition | 122 | Mixed: AMD Milan 7713, Intel Xeon Platinum 8468, Intel Xeon Gold 6438M | 4 NVIDIA A100 SXM4, NVIDIA H100 SXM, NVIDIA L40S | — | — | — |
Storage
Jetstream2 provides cloud storage associated with virtual machine instances rather than a traditional shared HPC filesystem. Instances may use ephemeral or volume-backed root disks. Data on an ephemeral root disk is permanently removed when the instance is deleted; retention of a volume-backed root disk depends on the instance's delete-volume setting.
Persistent storage options include block volumes, OpenStack Manila file shares, and S3-compatible object storage. Volumes can be attached to one instance at a time, Manila shares can be mounted by multiple instances, and object storage can be accessed through S3-compatible tools or HTTP. All Jetstream2 allocations receive 1 TB (1000 GB) of storage quota by default; additional storage can be requested through the Indiana Jetstream2 Storage resource.
File Transfer
Jetstream2 does not use dedicated data transfer nodes. Files are transferred directly to and from VM instances or through web-based and managed transfer tools.
Common methods include the Exosphere web shell/web desktop file-transfer interface, SCP/SFTP to a VM with a public IP address, and Globus for larger or managed transfers. See the Jetstream2 File Transfer documentation for current guidance.
| Supported Methods | Data Transfer Node | URL |
|---|---|---|
| GLOBUS | RECOMMENDED | https://www.globus.org/data-transfer | |
| EXOSPHERE WEB SHELL/WEB DESKTOP | https://docs.jetstream-cloud.org/ui/exo/exo-filetransfer/ | |
| SCP | ||
| SFTP |
Login to Jetstream2 GPU
2FA/MFA required for login
Jetstream2 GPU is accessed primarily through the Exosphere web interface. Users do not create a separate Jetstream2-specific user account. They need an ACCESS account and must be the PI of, or be added to, an active allocation that includes Indiana Jetstream2 GPU. Access to Jetstream2 CPU does not automatically provide access to the GPU resource.
In Exosphere, select Add allocation, choose Add ACCESS Account, select ACCESS CI (XSEDE) as the CILogon identity provider, sign in with ACCESS credentials, and select the allocation containing the GPU resource. ACCESS uses Duo for multi-factor authentication. Users who need to enroll in Duo or manage their authentication devices should follow the ACCESS Duo MFA instructions.
Jetstream2 does not provide a shared SSH login node. Users must first create a GPU-enabled virtual machine in Exosphere. To create an SSH key pair on a local Linux, macOS, or Windows OpenSSH system, run ssh-keygen -o -a 100 -t ed25519 -C "descriptive-label" -f ~/.ssh/id_ed25519. During instance creation, open the advanced options and select or upload the generated public key, normally ~/.ssh/id_ed25519.pub. Only the public key should be uploaded; the private key must not be shared.
Assign the instance a public IP address when direct SSH access is required. Exosphere-created instances use the exouser account. Connect with ssh -i ~/.ssh/id_ed25519 exouser@<PUBLIC_IP>. Users may alternatively connect through Exosphere Web Shell or Web Desktop.