📄️ Account Hierarchy Setup
Slurm's accounting system tracks GPU usage by account — a hierarchical label that maps to your organization structure. Every user belongs to one or more accounts, and every job is charged to an account.
📄️ Fair-share, QoS, Priority & Preemption (Optional)
This page is optional. Follow this page to setup Fair-share, QoS, Priority and Preemption on your cluster.
📄️ Sync LDAP Users to Slurm Accounting (Optional)
This page is optional. If you manage users manually or via CSV import (see Account Hierarchy), skip this section.
📄️ Job Tagging Strategy
While Slurm accounts track which team uses GPUs, job tagging tracks which project or model version consumes capacity. This enables cost allocation per model version beyond team-level tracking.
📄️ Data Extraction
Slurm provides two tools for extracting GPU usage data: sreport for aggregated summaries and sacct for job-level detail. Together they produce the reports needed for chargeback.
📄️ S3 Upload Automation
Uploading GPU reports to S3 enables downstream analytics via Athena and QuickSight, persistent storage for audit trails, and access for non-cluster stakeholders.
📄️ Cost Allocation from CUR
This page converts GPU-hours (from Data Extraction) into dollar costs using your actual instance pricing from the AWS Cost and Usage Report (CUR).
📄️ QuickSight Dashboard (Optional)
Amazon QuickSight provides visual dashboards for GPU chargeback data. This page covers creating Athena tables from your S3 reports and building QuickSight visualizations.