DPS 0.9.0 - Resource Group activation fails on DGX_B300 nodes ("policy capability max cannot be more than active psu policy limit max")

Accelerated Computing DGX Station / GB300 / GB200
1

Hello NVIDIA DPS Support Team,

I am evaluating DPS 0.9.0 using the official dps and dps-bmc-simulator 0.9.0 Helm charts on k3d. I found a reproducible issue specific to DGX_B300.

Environment:

  • DPS/dps-bmc-simulator: 0.9.0

  • k3d single-node cluster; dpsctl 0.9.0

  • 35 DGX_B300 simulator nodes plus 2 default DGX_H100 nodes

  • Custom topology with one 1 MW PowerDomain

  • Redfish credentials verified; topology activation and BMC health checks succeed for all nodes

Issue:
Resource Group activation consistently fails for every DGX_B300 with:

“policy capability max cannot be more than active psu policy limit max”

This occurs with MaxP, MaxQ, MaxLPS, and a custom percent-only policy, regardless of PSU redundancy mode, shared-GPU/DPM settings, topology re-import, or service restarts.

The identical workflow with DGX_H100 succeeds, returning power_cap_watts: 7140 and “Success”.

This strongly suggests a DGX_B300-specific issue in DPS 0.9.0, rather than topology, credentials, policy, or cluster configuration.

Questions:

  1. Is this a known DGX_B300 issue?

  2. Is any B300-specific PSU/device configuration missing from the public documentation?

  3. Is a fix or workaround planned for a future 0.9.x release?