What Is NVIDIA Fleet Intelligence?
Fleet Intelligence gives GPU owners and operators a single view of health and integrity across an entire fleet, including deployments spread over multiple sites. It is a managed service: a low-footprint agent runs on each host and streams telemetry to NVIDIA’s cloud, which aggregates it into dashboards, reports, and alerts. It is generally available at no cost to data center GPU owners.
The agent is read-only. It collects machine telemetry and does not modify host configuration. It is installed through a Linux package manager or Helm, is released as open source on GitHub for auditability, and builds on existing NVIDIA tooling: DCGM, GPUd, and the Attestation SDK.
What It Reports
| Area | Signals |
|---|---|
| Power | Utilization and throttling |
| Temperature | Hotspots and airflow problems |
| Performance | GPU utilization, memory bandwidth, interconnect health |
| Health | ECC errors, XID errors, retired pages, RAS signals |
| Configuration | Driver, firmware, and BIOS consistency across the fleet |
Alongside monitoring, it provides inventory visualization across data centers, customizable alerting (email, Slack), health checks, and integrity verification through cryptographic GPU attestation via the NVIDIA Remote Attestation Service.
Configuration reporting is worth calling out separately. Driver and firmware drift across a fleet produces failures that are hard to attribute, because the same workload succeeds on one node and fails on another for reasons that have nothing to do with the workload. Catching that drift as a fleet-level report is cheaper than debugging it one job at a time.
Support and Scope
Fleet Intelligence supports Vera Rubin, Blackwell, and Hopper architectures. Attestation is limited to Vera Rubin and Blackwell.
It is the fleet visibility component of NVIDIA DSX OS, and it pairs with NVSentinel, which handles automated remediation. Fleet Intelligence tells you what the fleet looks like. NVSentinel changes cluster state when something on it breaks.
