DPS SDK and Simulator#
The DPS Software Development Kit (SDK) is a Kubernetes-based environment for running DPS without physical data center infrastructure. Use it to evaluate DPS, develop against the API, and build integrations.
Evaluate DPS Without Hardware#
In its Hardware Emulation and Workload-Aware profiles, the simulator emulates 144 DGX GB300 compute nodes across eight racks, together with the complete power distribution hierarchy, Redfish BMC endpoints, Grafana dashboards, and automated simulation playbooks. It serves the full DPS API, so integrations you build against the simulator target the same interface as real hardware.
The Pilot profile provides a dedicated four-rack, 72-node GB300 environment. Rehearse the baseline and managed workflows with simulated power traces and compare their GPU power in Grafana. Follow Rehearse the MaxLPS Power Pilot With the SDK Simulator for the full procedure. Inference performance and physical power compliance require a hardware pilot.
It is a good fit for:
Learning DPS concepts and workflows
Proof-of-concept demonstrations before committing to a deployment
Developing and testing custom integrations
Automated testing of power management scenarios in CI
Start here: the SDK Simulator User Guide covers the four deployment profiles, installation, and the simulation playbooks.
The SDK Simulator Topology guide explains the general eight-rack topology used by Hardware Emulation and Workload-Aware profiles.
Build integrations#
The following guides cover extending DPS rather than evaluating it. They assume you already have the simulator running.
Administrator Guides — deployment architecture, operations, and the partner integration guides for grid integrators, optimization developers, and workload schedulers
Developer Guides — building custom integrations and extensions
The gRPC API reference documents the
nvidia.dcpower.v1 service that all of these build against.
Deploying against real hardware#
When you are ready to move beyond the simulator, refer to Deploying DPS.