Description
Jetstream2 GPU is a hybrid-cloud platform that provides flexible, on-demand, programmable cyberinfrastructure tools ranging from interactive virtual machine services to a variety of infrastructure and orchestration services for research and education. This particular portion of the resource is allocated separately from the primary resource and contains 360 NVIDIA A100 GPUs -- 4 GPUs per node, 128 AMD Milan cores, and 512gb RAM connected by 100gbps ethernet to the spine.
RP Description
Jetstream2 GPU is a cloud VM service consisting of on-demand GPU-enabled virtual machines in partial- and full-card flavors on NVIDIA A100, L40S, and H100 hardware. It is particularly well suited for interactive GPU work and GPU-accelerated services rather than long batch training runs, and is often used for AI and machine learning, visualization, and GPU-accelerated analysis. It includes whatever software you install, alongside the Jetstream2 software collection.
Login Text
Full instructions are in [Jetstream2 User Interfaces]. Jetstream2 GPU has no shared login node - you reach a machine by creating a virtual machine yourself. Exosphere is the recommended interface and the one these instructions assume; Jetstream2 also supports other OpenStack interfaces if you prefer one. Before your first sign-in:
- You need an ACCESS account, and you must be the PI of - or be added to - an active allocation that includes Indiana Jetstream2 GPU.
- There is no separate Jetstream2 account. Sign in with your ACCESS credentials, choosing ACCESS CI (XSEDE) as the CILogon identity provider.
- ACCESS uses Duo for multi-factor authentication. To enroll or manage your devices, see [Enroll in or manage ACCESS Duo MFA].
Instances created through Exosphere use the exouser account. Once an instance has a public IP address you can connect through Web Shell, Web Desktop, or SSH.
Storage Text
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.
Jobs Information
Jetstream2 GPU is a cloud virtual-machine resource, not a centrally scheduled Slurm cluster. There is no shared login node and no Slurm queue, partition, sbatch, srun, or squeue workflow. Slurm commands apply only if your project separately deploys and administers its own virtual cluster.
Select an available g3.*, g4.*, or g5.* GPU flavor, create a VM, and connect through Web Shell, Web Desktop, or SSH. GPU availability depends on your project's allocation and quota, and some full-card or multi-GPU flavors require a justified support request. Applications run directly inside the VM, and you install software with the operating system's package manager, Python environments, containers, or the Jetstream2 software collection.
An active GPU instance consumes service units even when no GPU process is running. Shelve an instance when you are not using it; a shelved instance consumes zero compute SUs.
See [Jetstream2 Instance Flavors] and [Jetstream2 Instance Management and SU Conservation]