Skip to content
nodeza.learn
DOCA course/2. Bring-up/Lesson 2.6

Dell-specific bring-up: iDRAC, BIOS, CPLD, aux power, DSE

S2·E6Three cards, three different Dell failures · Caged row, Friday, three hours before the acceptance demo

S2·E6Analyze~40 minsources checked todayverified against Dell KBs 000227031, 000225111, 000379421, 000300192; Dell R770 and R7725 Installation and Service Manuals (DPU specifications); NVIDIA BlueField-3 User Guide; 2026-09-06

Builds on: Switching modes safely

Before you read: what do you already know?

3 quick questions. Wrong answers are fine and expected; trying first makes the lesson stick.

After this lesson you can

  • Diagnose the Dell-documented BlueField-3 bring-up faults (No Memory Found, PCIe initialization failure, post-update fatal error) from their symptoms and versions.
  • Decide whether a given BlueField-3 SKU needs the 8-pin auxiliary cable in a PowerEdge and state what evidence supports it.
  • Place a BlueField-3 in a supported slot on R760, R770 and R7725 from the Dell manuals.
  • Explain Dell's support boundary between Dell-sourced and NVIDIA-channel cards, including VMware DSE.

Episode 6 — Three cards, three different Dell failures

The situation · Caged row, Friday, three hours before the acceptance demo

You are on the floor with the SE, the operator and the last three nodes of a rack that was supposed to be finished on Wednesday. Three BlueField-3 cards, three different symptoms. One R760 shows “No Memory Found” at POST after last night’s BIOS update. On another, the Lifecycle Log repeats “Device not detected: Nvidia Network Adapter”. A third card powers up and halts. The replacement request covering all three is already open, and procurement has already asked what an RMA does to a nine o’clock demo.

None of them is a dead card. The POST message is Dell KB 000227031: after BIOS updates or cold reboots the state machines of the BlueField-3, the iDRAC and the CPLD are not synchronized, the server waits about a minute, reboots itself, and memory is seen as expected.[1] The absent card is KB 000379421, a PCIe training failure caused by device firmware and fixed from 32.46.3048.[3] The halted one is a DPU SKU in a riser with no 8-pin PCIe cable attached, and its complaint is on the rshim console, not in the host’s dmesg.[8][10]

Dell writes these KBs because a DPU is not an ordinary adapter. It carries its own Arm system, its own power requirement and its own firmware clock, and it lands in a platform that already has three more.[1][7]

Four clocks — CPLD, iDRAC, BIOS, card firmware — and a symptom means nothing until you have read all four.

Segment 1 starts with those clocks.

1Three firmware clocks and the card's own

In a PowerEdge a BlueField-3 powers up alongside three platform components with their own firmware: the BIOS, the iDRAC and the CPLD. Dell KB 000227031 documents what happens when their timing and the card’s disagree. The symptom: at POST the server displays “No Memory Found”, “remains at the error message status in the BIOS … waits for approximately 1 minute, then it reboots automatically. Once the server reboots, it is fully functional and memory is seen as expected.”[1]

The cause is not memory. A firmware combination introduced a graceful DPU shutdown, and after BIOS updates, PSB updates, BIOS configuration changes or cold reboots “the timing of the various state machines are not synchronized” between the BlueField-3, the iDRAC and the CPLD.[1] Affected: PowerEdge R660, R760, R760XA and XE9680; CPLD 1.1.5 and 1.1.7; iDRAC 7.10.50.00; BlueField-3 firmware 32.40.1000; DPUs B3210e and B3220 “when operating in DPU mode”. The B3140H “comes set in Super NIC mode. If it is changed to DPU Mode, it can also encounter this event.”[1] Resolution: “Graceful shutdown is disabled in an upcoming CPLD release”; until then, wait for the automatic reboot.[1]

Two lessons for a bring-up plan. First, mode matters to the platform, not just to the card: a NIC-mode SuperNIC is outside this KB, the same card in DPU mode is inside it. Second, the fix lives in Dell’s CPLD, so your version inventory must include CPLD, iDRAC and BIOS next to the card firmware, and any escalation must quote all four.

The iDRAC9 7.x User’s Guide has a top-level “Data Processing Unit (DPU)” topic; the page fetched for this lesson rendered only its table of contents, so what iDRAC can do with a BlueField (license requirements, Redfish one-time boot, Arm UEFI and BMC firmware updates) is not taught here as fact.[11] No fetched Dell document ties the OS-to-iDRAC pass-through feature to BlueField management; do not tell a customer that dependency exists.

SymptomBlueField-3 on a Dell Pow…SymptomDell platform: POST, iDRA…Symptom"No Memory Found" POST e…SymptomvSphere Distributed Serv…SymptomCard disappeared after a…SymptomFirmware: Dell DUP vs NV…
Symptom

Dell platform: POST, iDRAC, vSphere, firmware

Start at the Dell branch: the card is in a PowerEdge and did not come up. The first splits are slot, aux power, POST message and LC log.

2PCIe initialization failure and the power-cycle rule

The second Dell KB is about a card that is simply absent. KB 000379421, “BlueField-3 DPU PCIe Initialization Failure”, describes a Lifecycle Log that repeats “Device not detected: Nvidia Network Adapter” (Slot 33, PR8): a PCIe training failure caused by BlueField-3 device firmware, “a known NVIDIA issue” fixed from firmware 32.46.3048, released August 14, 2025.[3] The resolution path is ordered: full power cycle; if the card returns, update to 32.46.3048 or later via the “NVIDIA DOCA Software Framework” download and keep the BFB aligned with the firmware; if it is still undetected, “open a service request for a replacement card”.[3]

The third is about what happens after the fix. KB 000300192 (XE9680L): after updating DPU firmware without a power cycle, the LC log reports “A fatal error was detected on a component at bus [number], device 0, function 0”; per NVIDIA, “a system power cycle is required after updating the DPU firmware”, and a warm reboot is not sufficient.[4] That is the same cold power-cycle rule you met in lessons 2.2 and 2.5, now with Dell’s LC log as the evidence.

Where the firmware comes from is a support-boundary question. Dell publishes a NIC-firmware DUP for BlueField-2 (4RR0M, version 24.36.75.06) that can “ONLY be applied via the iDRAC/Lifecycle Controller”; “Executing this firmware DUP from the host OS is not supported.”[13] For BlueField-3, KB 000379421 sends you to NVIDIA’s DOCA downloads (BF-FW-Bundle or BFB), not to a DUP.[3] No page fetched for this lesson shows a BlueField-3 firmware DUP or a BFB packaged as a DUP; if a customer’s process assumes one, ask them to show it before you plan around it.

3Auxiliary power: which SKUs need the 8-pin cable

Dell’s manuals give the wattage: on the R770, B3220 is 150 W and B3140H is 75 W; on the R7725, B3220 and B3240 are 150 W and B3140H is 75 W; all are FHHL, PCIe Gen5 x16.[5][6] NVIDIA’s BlueField-3 introduction says the DPU SKUs (B3240, B3220, B3210, B3210E) require the “supplementary 8-pin ATX power supply”, while SuperNICs run on slot power.[8]

The specifications page splits the requirement by product class, not by model: the SuperNIC OPNs (B3140H, B3140L, B3220L, B3210L) take “a minimum of 75W system power supply through the PCIe x16 interface”, while the B3220, B3240, B3210 and B3210E DPUs “require a supplementary 8-pin ATX power supply connectivity available through the external power supply connector”.[7] The introduction states the same rule outright and adds the cable type: “For B3240, B3220 and B3210E DPUs, you need an 8-pin PCIe external power cable to activate the card. The cable is not included in the package.”[8] The HGX B300 reference architecture agrees for the B3240, and a field report shows a B3210 in DPU mode halting with “ATX power not detected! Halting system!!” until an 8-pin PCIe (GPU-style, not EPS12V) cable was attached.[9][10]

Working rule, as documented: every BlueField-3 DPU SKU (B3210, B3210E, B3220, B3240) in a PowerEdge riser gets the 8-pin PCIe auxiliary cable; the 75 W SuperNICs, the B3140H included, run on slot power.[7][8] Dell’s cable kit part numbers per riser are not on the pages fetched for this lesson; get them from the Dell configurator or the account team, not from memory.[5][6] One more trap: the NVIDIA “PCIe Auxiliary Card Kit” is a PCIe-lane extension for Socket-Direct or downstream-port modes, not a power accessory; ordering it will not fix an ATX-missing halt.[14]

4Slots and risers: R760, R770, R7725

Dell fixes which slots take a BlueField-3, and the rules differ by generation. On the R770 the Installation and Service Manual’s “DPU Specifications” lists “NVIDIA BlueField-3 2 x 200 GbE B3220” and “NVIDIA BlueFIeld-3 1 x 400 GbE B3140H”; supported riser configurations are RC 1 and RC 2 (slots 31 and 36) and RC 6-2 and RC 11-2 (slots 7 and 2); in RC 11-2, “slot 5 supports installation of the NVIDIA BF3 DPU 1 x 400 GbE B3140H when slot 7 and slot 2 are populated with NVIDIA RTX Pro 6000 GPUs”; and “Riser configurations RC 6-2 and RC 11-2 support installation of GPU RTX Pro 6000 and DPUs on the same system”.[5]

On the R7725 the manual lists B3220, B3140H (asterisked: may not be available at launch) and B3240; B3220 and B3240 go in RC 2, RC 3, RC 4, RC 5 and RC 8, slots 2 and 7; the B3140H slot table reads “TBD”.[6] For 16G, KB 000227031 is the evidence that R660, R760, R760XA and XE9680 ship BlueField-3 options, but the fetched R760 material does not include a slot table, so take the slot from the R760 manual for the riser you have.[1] A search snippet suggests an R650 slot 1 and R750 slot 2 rule for older generations; it was not confirmed on a fetched page and is not a fact you should quote.

For the HGX-class XE9780, the only fetched evidence is the Dell Partner DPU listing (352-BBFH, dual-port 400GbE BlueField-3) naming XE9780 and XE8712 as compatible systems; the XE9780 manual’s DPU page rendered without a table.[12] Treat XE9780 support as partial until the manual is read.

5Channel versus Dell-sourced cards, and DSE

Dell draws a support line by where the card came from. KB 000225111 states: “VMware’s Distributed Services Engine is not supported on NVIDIA Channel cards by Dell. Dell has not validated this feature and is unable to support it, if it is enabled.” The option appears under “Device Level Configuration of Device Setup” on BlueField DPUs; the KB applies to XE8640, XE9640 and XE9680 with ESXi 7.x and 8.x.[2] Broadcom’s KB adds the factory state: the BlueField-3 “by default is in the DPU mode and VMware Distributed Services Engine (DPU) is disabled”.[15]

The Dell-sourced counterpart is the “Dell Partner DPU, NVIDIA BlueField-3 Dual Port 400GbE” (Dell part 352-BBFH, manufacturer part XWFTP), with the statement “This product has been tested and validated on Dell systems. It is supported by Dell Technical Support when used with a Dell system”.[12] A card bought through NVIDIA’s channel and installed in the same server is outside that statement for DSE.

Practical position for a Dell FAE: on PowerEdge, DSE is a Dell-SKU conversation; an NVIDIA-channel BlueField-3 in a VMware farm runs in NIC mode (SuperNIC-like) unless Dell qualifies that DPU for DSE. Everything else about the card, DOCA-Host, BFB, modes, works the same regardless of source; what changes is who answers the support call.

Triage a Dell bring-up ticket

Ticket: “R760, BIOS updated last night. This morning POST shows No Memory Found. Two BlueField-3 B3220 in DPU mode. Customer wants to pull the DIMMs.”

  1. Match the symptom to the KB. “No Memory Found” after a BIOS update on an R760 with B3220 in DPU mode is the KB 000227031 pattern. Ask: did the server reboot itself after about a minute and come up with all memory? If yes, this is the KB.[1]
  2. Confirm the version combination: CPLD (1.1.5 or 1.1.7?), iDRAC (7.10.50.00?), BlueField-3 firmware (flint q, 32.40.1000?). All three matching is confirmation; any one differing still leaves the KB likely but asks for the versions in the escalation.[1]
  3. Stop the DIMM pull. The KB’s cause is state-machine timing between the card, iDRAC and CPLD; memory is seen as expected after the automatic reboot.[1]
  4. Resolution: the fix ships in a CPLD release that disables graceful shutdown; until then the workaround is to let the automatic reboot happen. Plan the CPLD update as a Dell platform change, with its own power cycle.[1]
  5. Check adjacent risks while you are there: is firmware below 32.46.3048 (KB 000379421 exposure)?[3] Was the last firmware update followed by a power cycle (KB 000300192)?[4] Are the cards on the aux cable (150 W SKU)?[6]
  6. Write back with the KB number, the three versions, the expected behaviour, and the CPLD plan.

End of Episode 6 — Case closed

How it ended

The R760 is left alone through its automatic reboot and comes up with all its memory.[1] The absent card is power-cycled, returns, takes a firmware update from the NVIDIA DOCA download and stays.[3] The third gets an 8-pin PCIe cable, not the PCIe Auxiliary Card Kit somebody suggested ordering, which extends lanes and supplies no power at all.[14] Nothing ships back. At nine the demo runs on twenty-four nodes, the network lead writes his counter on the last free page of the notebook, and the SE’s spreadsheet is green for the first time in five days.

What you say to the customer: “Give me CPLD, iDRAC, BIOS and card firmware versions before anyone quotes hardware. On a PowerEdge, three of those four are Dell’s.”[3][1]

Lab

Inventory only; nothing here changes the server or the card.

  1. Platform identity: sudo dmidecode -s system-product-name; sudo dmidecode -s bios-version. Record model and BIOS. If the model is R660/R760/R760XA/XE9680, KB 000227031 is in scope.[1]
  2. iDRAC and CPLD versions over standard Redfish from a management host: curl -k -u <user>:<password> https://<idrac_ip>/redfish/v1/UpdateService/FirmwareInventory and follow the members for iDRAC and CPLD. Record both; compare with 7.10.50.00 and 1.1.5/1.1.7.[1] If Redfish is not enabled, read them from the iDRAC web UI’s firmware inventory instead.
  3. Card firmware: sudo mst start && sudo flint -d /dev/mst/mt41692_pciconf0 q | grep -E 'FW Version|FW Release Date'. Compare with 32.40.1000 (KB 000227031 exposure) and 32.46.3048 (KB 000379421 fix level).[1][3]
  4. Slot: sudo lspci -d 15b3: -vv | grep -E '^[0-9a-f]|Physical Slot'. Record the physical slot and check it against the manual for this model and riser.[5][6]
  5. Aux power: grep -E 'OPN_STR|DEV_INFO' /dev/rshim0/misc to identify the SKU, then decide from the wattage table whether a cable is required; visually confirm with the lab’s chassis photo or the account team’s riser sheet. If the card halts, sudo screen /dev/rshim0/console 115200 and look for ATX power not detected.[7][10]
  6. LC log: in iDRAC, export the Lifecycle Log and search for Device not detected: Nvidia Network Adapter and fatal error was detected on a component. Record hits with timestamps and match them to the last firmware update or reboot.[3][4]
  7. Mode: sudo mlxconfig -d /dev/mst/mt41692_pciconf0 q INTERNAL_CPU_OFFLOAD_ENGINE. A B3140H showing ENABLED(0) on an affected 16G combination is inside KB 000227031; note it.[1]
  8. Write the one-row inventory: model, BIOS, iDRAC, CPLD, card SKU, slot, aux cable, FW, mode, KB exposure. This row is what lesson 8.3’s validation plan expects per server.

Retrieval check

10 questions from memory. Answer before looking anything up; misses become flashcards.

Explain it to a Dell SE

Explain to a Dell SE, in five sentences, why a BlueField-3 that works perfectly on a lab bench can fail to come up in a PowerEdge, and what you check in what order.

14 flashcards for this lesson — 0 in deck. Spaced review lives at /review.

Sources

Facts in this lesson were checked against Dell KBs 000227031, 000225111, 000379421, 000300192; Dell R770 and R7725 Installation and Service Manuals (DPU specifications); NVIDIA BlueField-3 User Guide; 2026-09-06. Dates are when each page was fetched.

  1. Dell KB 000227031: 'No Memory Found' event on BlueField-3 enabled PowerEdge during POST · fetched 2026-09-06
  2. Dell KB 000225111: NVIDIA Channel DPU VMware DSE Support · fetched 2026-09-06
  3. Dell KB 000379421: BlueField-3 DPU PCIe Initialization Failure · fetched 2026-09-06
  4. Dell KB 000300192: XE9680L, a power cycle is required after updating DPU firmware · fetched 2026-09-06
  5. Dell PowerEdge R770 Installation and Service Manual: DPU Specifications · fetched 2026-09-06
  6. Dell PowerEdge R7725 Installation and Service Manual: DPU Specifications · fetched 2026-09-06
  7. BlueField-3 User Guide: Specifications (per-SKU power) · fetched 2026-09-06
  8. BlueField-3 User Guide: Introduction (8-pin ATX on DPU SKUs) · fetched 2026-09-06
  9. NVIDIA HGX AI Factory Enterprise Reference Architecture, Components (B3240 aux power) · fetched 2026-09-06
  10. NVIDIA developer forum: BlueField-3 in DPU mode stuck in FW pre-initializing, ATX power not detected · fetched 2026-09-06
  11. iDRAC9 7.x User's Guide: Data Processing Unit (DPU) topic · fetched 2026-09-06
  12. Dell Partner DPU, NVIDIA BlueField-3 Dual Port 400GbE (352-BBFH) · fetched 2026-09-06
  13. Dell driver 4RR0M: NIC firmware for NVIDIA BlueField-2 DPUs (DUP) · fetched 2026-09-06
  14. BlueField-3 User Guide: PCIe Auxiliary Card Kit (not a power accessory) · fetched 2026-09-06
  15. Broadcom KB 379391: Configuring BlueField-3 into DPU mode for vSphere DSE or NIC mode · fetched 2026-09-06

The same idea elsewhere

Other lessons that cover this ground, sometimes from another course's angle.