skills.rest
NVIDIA
NVIDIAOFFICIAL
Explore all skills by this author →
WebsiteGitHub

doca-bf4-deployment

Guides BlueField-4 day-1 bring-up via BMC using UEFI HTTP Boot, PXE, Redfish Virtual Media, and PLDM firmware updates.

3.2k|370|Updated Feb 25, 2026
Search Tags:#doca#firmware update#BMC#redfish#pldm#dpu bring-up#bluefield-4
One-click install
npx skills add https://github.com/NVIDIA/skills --skill doca-bf4-deployment
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: doca-bf4-deployment
Source: https://github.com/NVIDIA/skills/tree/main/skills/doca-bf4-deployment
Command: npx skills add https://github.com/NVIDIA/skills --skill doca-bf4-deployment

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Standing up a brand-new BlueField-4 DPU is a high-stakes, error-prone process: operators must choose an OS install method, push firmware bundles through Redfish/PLDM, and avoid irreversible mistakes like bricked firmware or boot loops. This Skill gives an AI agent the documented, safety-gated workflow for BF4 day-1 platform bring-up driven entirely from the BMC.

Core Features & Use Cases

  • Three documented OS install methods: Install the BlueField/DOCA bundle ISO onto Grace (the Arm complex) via UEFI HTTP Boot (recommended), PXE Boot, or fully out-of-band Redfish Virtual Media.
  • PLDM firmware-update flow: Push .fwpkg bundles through the Redfish UpdateService, monitor the returned Task, verify pending images with pldmtool, and activate via power cycle across BMC, NIC firmware, SBIOS, and ERoT components.
  • Grace Ubuntu + cloud-init path: Install a Grace Ubuntu image with an optional CIDATA-labelled cloud-init seed through Redfish Virtual Media, including the fixed image.iso/config.iso URI rules.
  • Layered error taxonomy and safety gating: Diagnose failures across six layers (boot-source, virtual-media-attach, firmware-Task, activation, cloud-init, boot-loop), with every mutating step routed through the doca-hardware-safety meta-policy and explicit operator confirmation.
  • Use Case: An operator with a fresh BlueField-4 asks the agent to install the DOCA bundle ISO without setting up PXE; the agent walks them through UEFI HTTP Boot, confirms OOB console access, and runs the post-install verification sweep.

Quick Start

Ask your agent to bring up your new BlueField-4 from the BMC, for example: "Help me install the BlueField/DOCA bundle ISO onto my new BlueField-4 over UEFI HTTP Boot and verify the firmware levels."

Frequently Asked Questions about doca-bf4-deployment

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I install the DOCA bundle ISO on a new BlueField-4?▼

BlueField-4 supports three documented install methods: UEFI HTTP Boot (recommended, fewest moving parts), PXE Boot (when you already have DHCP/TFTP infrastructure), and Redfish Virtual Media (fully out-of-band). All are driven from the BMC with the OOB console monitored via obmc-console-client.

How do I update BlueField-4 firmware with PLDM over Redfish?▼

Push the .fwpkg bundle via a Redfish multipart POST to the UpdateService endpoint, monitor the returned Task until Completed, verify pending images with pldmtool fw_update GetFwParams, then activate with a power cycle. Every step is mutating and requires operator confirmation through the doca-hardware-safety policy.

Can I install Grace Ubuntu with cloud-init on BlueField-4?▼

Yes, via Redfish Virtual Media using a config ISO with the exact volume label CIDATA. The BMC-facing URIs must be exactly image.iso and config.iso, local eMMC hosting is limited to 5 GB combined, and media must be detached after install to avoid boot loops.

Why does my BlueField-4 keep re-entering the installer after reset?▼

This boot loop happens when install media was never detached and BootSourceOverride re-selects it. Detach all virtual media, confirm the Redfish VirtualMedia resource reports Inserted: false, and clear or re-point the boot override.

Does this skill cover BlueField-3 bring-up?▼

No, this skill is BlueField-4-specific: the Grace naming, dpu-bmc Redfish surface, and install methods differ from BF3. BlueField-3 bring-up routes to the separate doca-bf3-deployment skill.

What should I do when a Redfish firmware Task stalls at Running?▼

Do not re-push or power-cycle blindly. Capture the complete Task resource and Messages, classify it as stalled only after two unchanged observations at an operator-approved interval, then stop and escalate with the bring-up snapshot and OOB-console evidence.

Related Skills

Explore More Software Engineering Skills→
air-gappedair-gapped
5

nvidia-datacenter-bringup

Automate NVIDIA datacenter GPU bring-up on Ubuntu 24.04 with health validation.

Official
Advanced
00PrabalK0000PrabalK00
1

hardware-bringup-checklist

Validate power, buses, firmware, and interfaces during initial hardware bringup.

Community
Intermediate
mem9-aimem9-ai
85

drive9

Provide persistent network filesystem storage with natural-language search for AI agents.

Official
Intermediate
input-output-hkinput-output-hk
1.2k

bech32-encoding-decoding

Converts between base16, base58, and bech32 formats using a local CLI.

Official
Basic