Sage

RP account needed

Sage is a national AI-at-the-edge research platform composed of sensor-equipped compute nodes deployed in natural, urban, agricultural, and environmental settings. It is particularly well suited for processing live sensor data near its source, and is often used for image, video, audio, weather, air-quality, soil, and water analysis. It includes a great deal of edge AI, sensor-data, and distributed-application software.

Submitting Jobs Documentation

Sage schedules work with the Sage Edge Scheduler rather than a batch scheduler such as Slurm, so there are no queues, no partitions and no wall-clock limits. A job is an instance of a science goal, written in YAML or JSON: it names the edge applications to run, the nodes to run them on, the science rules that decide when they run, and the conditions that complete the job. Nodes are targeted individually by their VSN, for example W023, and a job can only target nodes your account is authorized to schedule on.

Two kinds of node run jobs, and neither is a queue. Wild Sage nodes are weatherproof outdoor nodes built around an NVIDIA Xavier NX ARM64 node controller with 8 GB of shared CPU and GPU memory and 1 TB of NVMe storage; they are the target for field deployments that process camera, audio, weather, air-quality, soil or water data where it is collected. Blade nodes are rack servers for machine rooms and other controlled sites, with a multi-core ARM64 processor, 32 GB of memory, a dedicated NVIDIA T4 GPU and 1 TB of SSD storage; they suit work that needs more GPU or memory than a Wild Sage node provides. See [Sage Architecture] for the full hardware description.

Jobs are created, submitted and removed with the sesctl client, which needs job-submission permission from [Request Sage Access] and an access token before it will talk to the scheduler. [Sage Submit your job] carries the configuration steps and the full command set.

Sage does not enforce job success criteria, so a submitted job is served until you remove it - a job file may carry a wall-clock criterion and it will not stop the job. Watch your jobs with sesctl stat and remove them when they are no longer needed.

Queue specifications

Queue CPU cores / node GPUs / node Node RAM
Wild Sage Node
Best suited for outdoor, field-deployed edge AI workflows that need to process sensor data close to where it is collected, including environmental monitoring, camera/audio analysis, weather or air-quality sensing, and remote instrument deployments.
NVIDIA Xavier NX ARM64 node controller 1 NVIDIA Xavier NX shared CPU/GPU platform 8 GB
Blade Node
Best suited for edge AI workflows that need stronger GPU acceleration or more memory than Wild Sage nodes, especially deployments in machine rooms or controlled infrastructure sites that can support server/blade-style hardware.
Multi-core ARM64 1 NVIDIA T4 32 GB

Software Documentation

No software usage data is currently reported for Sage in XDMoD.

SEE ALL SOFTWARE AVAILABLE ON SAGE


File Transfer Documentation

Sage does not use a traditional HPC file-transfer workflow or data-transfer node. Use the Python Sage Data Client to query sensor data into a pandas DataFrame, or use the HTTP API for non-Python tools and service integrations. Large files such as images, audio, and video are referenced in API results and can be downloaded from their URLs with wget or curl. Protected data require an approved Sage account, a signed Data Use Agreement, and authentication with your Sage Portal username and access token. For instructions, see [Sage Access and Use Data].

Supported Methods Data Transfer Node / Globus Collection Notes
PYTHON SAGE DATA CLIENT | RECOMMENDED Sage Data Client
HTTP API Sage HTTP API