Pre-Validation Setup | NVIDIA Minerva User Guide

The other ORR subsections all assume the steps below are complete. They take roughly 30 minutes and only need to be done once per deployment.

Before running through the validation areas, the operator team needs the preconfigured dashboards imported into the right tenants, agreement on which time window the cluster is idle, and a place to archive baseline screenshots.

Persona Use Cases

Persona	Primary Use Case
Cluster Administrator	Imports the dashboard sets into both tenants once, on behalf of the team.
Tenant Admins	Confirms tenant-scoped users have visibility on the imported dashboards.
System Administrator / SRE	Captures the idle time window and archives baseline screenshots.

Validation Procedure

Step 1: Confirm Tenant Context

Sign in to the NVIDIA Minerva UI.
Confirm that you have access to both the default (admin) tenant and the user-tenant tenant. Tenant assignment is managed in User Management.
The bulk import in Step 2 adjusts the username and tenant of imported views to whichever tenant you are signed into. Sign in to the tenant you intend to import into before running the import.

Step 2: Import the Preconfigured Dashboards

The preconfigured dashboards live in the reference-architecture-configs repo. Clone it to the workstation running your browser, then import per the workflow described in Preconfigured Dashboards.

For first-time setup, use the combined JSON files for a single-file bulk import per tenant:

Tenant	Folder	Combined file
`default` (admin)	`minerva-configs/dashboards/minerva-stack-monitoring/`	`reference-architecture_admin_dashboards_combined_views_filtered.json`
`user-tenant`	`minerva-configs/dashboards/user-tenant/combined/`	`reference-architecture_user-tenant_dashboards_combined_views_filtered.json`

Individual dashboard sets in user-tenant/ (cluster-health/, nvlink/, professor-x/, telemetry-health/, ai-insights/, jobs-and-performance/, gpu/) can also be imported one at a time.

From the UI:

Open the global configuration cog and select Actions > Import.
Select the JSON file and click Import.
Confirm there are no warnings or errors in the import dialog.

Step 3: Establish the Idle Time Window

Coordinate with the workload owners to confirm the cluster will be idle (no production jobs scheduled) during the ORR validation window.
Use the time filter at the top of each dashboard to scope it to that window. A 1-hour window is typical for snapshot validation; a 24-hour window is appropriate for capturing diurnal noise.
Document the window start and end (for example, 2026-04-30 08:00 PT to 2026-04-30 12:00 PT): this is part of the baseline archive.

Step 4: Set Up the Baseline Archive

Decide where idle-baseline screenshots will be stored (for example, a shared drive, ticket attachment, or wiki page).
Each subsection's pass criteria asks for "baseline captured": the same target archive applies to all of them.
Capture screenshots in the light theme and at a consistent resolution to make later comparisons easier.

Pass Criteria

[ ] The operator running the validation has tenant access to both default and user-tenant.
[ ] Preconfigured dashboards are imported and visible in both tenants.
[ ] An agreed idle time window is documented and applied as the dashboard time filter.
[ ] A baseline-archive location exists and the team knows where to put screenshots.

Common Pitfalls

Importing into the wrong tenant. The bulk import binds views to the currently logged-in tenant. If user-tenant dashboards land in the admin tenant they will not have the right scoping; remove and re-import after switching tenants.
Trying to validate during workload activity. Active jobs make idle baselines impossible to capture. If the cluster cannot be quiesced, scope the validation window to a known low-activity period and document it explicitly.
Skipping the time-filter step. Default time filters vary by dashboard (some default to "latest day," others to 1 hour). Inconsistent windows produce unreliable cross-dashboard comparisons.

Last updated: October 14, 2026