Skip to content
nodeza.learn
DOCA course/8. Dell OEM/Lesson 8.1

PowerEdge support matrix

S8·E1Two lines, one chip, two support answers · Hotel lobby near the Dell campus, the night before the signature

S8·E1Understand~30 minsources checked todayverified against Dell KB 000227031 / 000225111 / 000300192, Dell R770 and R7725 ISM DPU pages, Dell Partner DPU 352-BBFH page, DOCA 3.5.0 modes page, 2026-09-06

Builds on: The NVIDIA networking portfolio, Modes of operation: DPU, NIC, Zero-Trust, Dell-specific bring-up: iDRAC, BIOS, CPLD, aux power, DSE

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

  • Name the 16G and 17G PowerEdge models with documented BlueField-3 options and state which ones are confirmed only indirectly.
  • Read the R770 and R7725 riser and slot rules for the B3220, B3240 and B3140H from the Dell manuals.
  • Distinguish a Dell "Partner DPU" (consignment) quote line from a Dell SuperNIC SKU and say who supports each.
  • Predict which mode a given SKU ships in and whether it needs the 8-pin auxiliary power cable.
  • State Dell's vSphere DSE support boundary in one sentence.

Episode 1 — Two lines, one chip, two support answers

The situation · Hotel lobby near the Dell campus, the night before the signature

Nine at night, and the Dell SE slides his laptop across the hotel lobby table, holding a coffee that went cold two meetings ago. Tomorrow the imaging bureau’s platform architect signs for sixteen PowerEdge nodes that must carry radiology reads for forty rural hospitals before quarter end. Two quote lines hold the same silicon: a “Partner DPU, NVIDIA BlueField-3 Dual Port 400GbE” marked CONSIGNMENT at $0.00, and a priced “SuperNIC, NVIDIA BlueField-3 Single Port 400GbE/NDR QSFP112”.[4] Procurement asks why one of them is free. Dell catalogues the Partner DPU with the statement that it “has been tested and validated on Dell systems” and “is supported by Dell Technical Support when used with a Dell system”[4], while a separate Dell article says DSE “is not supported on NVIDIA Channel cards by Dell” and never defines which cards those are.[11] The SE’s spreadsheet of promises already has a DSE row. The network lead opens a notebook to a clean page: show me the counter.

One chip ships two ways because a BlueField-3 can be a computer or a network adapter, and a customer needs one at a time. In NIC mode the Arm cores are inactive and the device functions as an NVIDIA ConnectX adapter, which “reduces power consumption, improves network performance, and minimizes the host memory footprint”.[8] In DPU mode the Arm complex stays up to run services. Two jobs, two SKUs, two support answers.

On PowerEdge, the quote line decides the support answer, not the chip. You open the matrix where the answer starts: which line, which SKU, which server.

1Two sourcing channels, two support answers

On a Dell quote the same BlueField-3 silicon can appear under two headings, and the heading decides who answers the support call. Dell catalogues a “Dell Partner DPU, NVIDIA BlueField-3 Dual Port 400GbE” (Dell P/N 352-BBFH, manufacturer part XWFTP) with the statement that the product “has been tested and validated on Dell systems” and “is supported by Dell Technical Support when used with a Dell system”.[4] Dell’s DSE knowledge base article uses a different phrase, “NVIDIA Channel cards”, and never defines it; for those cards it says “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.”[11] Because Dell does not say which sourcing route earns that label, get Dell to confirm in writing whether a given quote line counts before you promise or refuse DSE.

Search snippets of Dell’s US configurators show “Partner DPU, NVIDIA BlueField-3” lines at $0.00 marked CONSIGNMENT next to priced “SuperNIC, NVIDIA BlueField-3 Single Port 400GbE/NDR QSFP112” lines; those configurator pages were not fetched for this lesson, so treat the pricing pattern as unconfirmed and read the actual quote in front of you.[4] The working interpretation is this: a “Partner DPU” is an NVIDIA-channel DPU SKU (B3220/B3240 class) consigned into the Dell build and booting in DPU mode; the “SuperNIC” line is the Dell-catalogued B3140H, which ships in NIC mode.[4][8]

The matrix below is the module’s reference table. Every row carries an evidence grade: verified (fetched Dell manual or KB), indirect (a Dell compatibility list but no DPU table), or unverified (search snippets only).[1][5]

PowerEdgeBF-3 offeringDPU modeNIC modeAux powervSphere DSENotable KB
R660
16G
R760
16G
R760XA
16G (GPU)
XE9680
16G HGX H100/H200
XE9680L
16G liquid
R7725
17G (AMD)
R770
17G (Intel)
XE9780 / XE9785
17G HGX B300 — Dell AI Factory

✓ yes · ✗ no · ◐ conditional · ? unknown · — n/a. Click a cell for the evidence.

⚠ = not confirmed on a fetched primary source (hover for why). Facts as of DOCA 3.5.0 (Sep 2026). Selections are saved.

Dell tab: PowerEdge model × BlueField-3 SKU × riser/slot × power × evidence grade. Click a cell for the source.

216G: R660, R760, R760XA, XE9680 — and what the KBs imply

Dell KB 000227031 lists the 16G platforms it calls “NVIDIA Bluefield 3 enabled”: PowerEdge R660, R760, R760XA and XE9680.[1] That is a bug article, but its affected-systems list is Dell engineering’s own enumeration of where BlueField-3 shipped and was tested, which makes it a better matrix source than a marketing page.[1] Dell KB 000225111, the DSE article, applies to PowerEdge XE8640, XE9640 and XE9680 with ESXi 7.x and 8.x — those are the systems where NVIDIA-channel DPUs met VMware in the field.[11]

Two more 16G data points. Dell KB 000300192 documents a BlueField-3 DPU, Dell part “HFRWM”, in the PowerEdge XE9680L.[14] NVIDIA’s November 2023 Spectrum-X announcement named PowerEdge XE9680 servers as the Israel-1 reference architecture host, with BlueField-3 SuperNICs and Spectrum-4 SN5600 switches.[15]

For the mainstream line, the strongest evidence is a firmware package rather than a manual. Dell’s BlueField-2 NIC-firmware DUP 4RR0M (version 24.36.75.06, 5 Dec 2023) lists R650, R660, R6615, R6625, R750, R760, R7615, R7625, XR7620 and vSAN Ready Nodes as supported systems, which shows Dell shipped Dell-sourced BlueField DPUs across the 15G/16G line for vSphere use.[13]

One caution: a Dell community thread asserts that BlueField-3 is “only supported on the XE9680, and not the R760”, and a search snippet says the R760 technical guide mentions a Management Interface Card (MIC) for the Dell DPU card; the thread returned HTTP 403 and neither claim was fetched.[1] KB 000227031 explicitly lists the R760, so treat the thread as stale or SKU-specific, and say so when a customer quotes it.[1]

317G: R770 and R7725 slot rules; XE9780 is indirect

The 17G evidence is direct: both Installation and Service Manuals have a “DPU specifications” page.[2][3]

Server SKU Ports Power Form factor Riser configurations and slots
R770 B3220 2x200 GbE 150 W FHHL, PCIe Gen5 x16 RC 1 (slots 31, 36), RC 2 (slots 31, 36), RC 6-2 (slots 7, 2), RC 11-2 (slots 7, 2)
R770 B3140H 1x400 GbE 75 W FHHL, PCIe Gen5 x16 Same as B3220; in RC 11-2 also slot 5 when slots 7 and 2 hold RTX Pro 6000 GPUs
R7725 B3220 2x200 GbE 150 W FHHL, PCIe Gen5 x16 RC 2, RC 3, RC 4, RC 5, RC 8 — slots 2 and 7
R7725 B3240 listed as 1x400 GbE 150 W FHHL, PCIe Gen5 x16 RC 2, RC 3, RC 4, RC 5, RC 8 — slots 2 and 7
R7725 B3140H 1x400 GbE 75 W FHHL, PCIe Gen5 x16 Slots listed as TBD; asterisk “may not be available at launch”

The R770 manual adds that RC 6-2 and RC 11-2 “support installation of GPU RTX Pro 6000 and DPUs on the same system”, which is the configuration an AI-inference customer will ask about.[2] The R7725 manual, as extracted, labels the B3240 “1x400 GbE”; NVIDIA’s specifications page lists the B3240 as 2x400G dual-slot FHHL, so confirm the OPN on the quote rather than trusting either label.[3][6]

XE9780 is a different case. Dell’s Partner DPU 352-BBFH lists “PowerEdge XE8712” and “PowerEdge XE9780” as compatible systems, but the XE9780 ISM’s dpu-specifications URL rendered only navigation, with no DPU table.[4][5] The XE9780 and XE9785 are Dell’s HGX B300 chassis, and NVIDIA’s HGX reference architecture places one BlueField-3 DPU per node for north-south traffic in DPU mode, which is consistent with a dual-port 400GbE Partner DPU on the XE9780.[16][9] Call XE9780 indirect. XE9785, R670 and R570 have no fetched Dell page confirming BlueField-3 options; say “unverified” and ask Dell product management.[5]

4Which mode a SKU ships in

Mode is a per-SKU default, and it drives what the customer sees on first boot. Dell KB 000227031 states that the B3140H “comes set in Super NIC mode”, and that it can encounter the DPU-mode boot issue only “if it is changed to DPU Mode”; the B3210e and B3220 are described “when operating in DPU mode”, which is their shipping state.[1] NVIDIA’s HGX reference architecture likewise runs the north-south BlueField-3 in “ECPF or DPU mode” by default.[9]

The switch is one firmware parameter. INTERNAL_CPU_OFFLOAD_ENGINE=0 selects DPU mode and =1 selects NIC mode; “for the configuration to take effect, Arm and NIC components must undergo a reset. Power cycle is recommended.”[8] In NIC mode “the Arm cores of BlueField are inactive, and the device functions as an NVIDIA ConnectX network adapter”; the mode “reduces power consumption, improves network performance, and minimizes the host memory footprint”, and multi-host is unsupported.[8] Zero-trust (restricted) mode goes the other way: it “prevents the host system administrator from accessing BlueField from the host side”, restricting port ownership, hardware counters, the tracer, RShim and firmware flashing.[8]

So the two Dell quote lines map onto two modes. A “SuperNIC” B3140H is a NIC-mode card: the Arm cores are inactive and the device functions as a ConnectX network adapter, so there is no Arm OS to console into and no DOCA services on the card — and none are needed for an east-west AI fabric.[8] The modes page does not say the RShim device disappears in NIC mode, so do not use the absence of /dev/rshim as the NIC-mode test; use mlxconfig q INTERNAL_CPU_OFFLOAD_ENGINE (1 = NIC mode).[8] A “Partner DPU” B3220 or B3240 boots in DPU mode and can run HBN, OVS-DOCA or a storage service on its Arm cores.[9][8] If a customer bought SuperNICs and now asks for DOCA services, the honest answer is a mode change plus a power cycle, and on the B3140H that change also exposes them to the boot-sync issue in the next lesson.[1]

5Auxiliary power: 150 W SKUs and the ATX halt

The auxiliary-power rule is stated on NVIDIA’s side and only implied on Dell’s, and an FAE should be able to say which is which. NVIDIA’s BlueField-3 User Guide requires the supplementary 8-pin ATX power connector on all four DPU SKUs (B3240, B3220, B3210, B3210E), while the SuperNICs run on slot power; the User Guide’s introduction states a “Minimum 75W or greater system power supply for all cards”, and the specifications page repeats “a minimum of 75W system power supply through the PCIe x16 interface” for the SuperNIC rows.[7][6] NVIDIA’s HGX B300 reference architecture repeats it for the B3240: “requires 8-pin ATX 12V PCIe”.[9] Dell’s manuals do not state the cable requirement; they rate the B3220 and B3240 at 150 W and the B3140H at 75 W, which is the same fact expressed as power.[2][3]

The field settles it. A B3210 P-series in DPU mode sat in “FW pre-initializing” with no host netdevs; its console read “ATX missing - initiating halt sequence… ATX power not detected! Halting system!!”, and it booted once an 8-pin PCIe cable (GPU-style, not EPS12V) was attached.[10] The User Guide puts the slot baseline at 75 W[7]; a 150 W card has to draw the rest through the connector, and without it the Arm complex never comes up.

Working rule: every 150 W BlueField-3 DPU SKU needs the 8-pin PCIe auxiliary cable in a PowerEdge riser; the 75 W B3140H does not.[7][3] Two things remain unconfirmed and should be asked of Dell rather than asserted: the exact cable-kit part numbers per riser, and whether Dell’s own documentation states the requirement.[3] Do not confuse NVIDIA’s “PCIe Auxiliary Card Kit” with a power accessory; it is a lane-extension kit.[6]

6vSphere DSE: a Dell-sourced-only story

VMware’s Distributed Services Engine runs ESXio on the DPU’s Arm cores to offload networking; on PowerEdge it has a hard support boundary. Dell 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.”[11] The same article notes that NVIDIA “proactively added the DSE configuration option” to BlueField DPUs, so the setting is present on every card whether or not Dell supports it.[11]

Broadcom’s KB 379391 supplies the mechanics. A BlueField-3 “out of the factory, by default is in the DPU mode and VMware Distributed Services Engine (DPU) is disabled”; DPU mode for DSE means the “embedded Arm system controls the NIC resources and data path” with DSE enabled in device settings, while in NIC mode “BlueField behaves exactly like an adapter card from the perspective of the external host”.[12] Crucially, “when you switch BlueField-3 from NIC mode to VMware Distributed Services Engine enabled DPU mode, you must re-install ESXi”, in either direction, on vSphere 8.0 U3b and later.[12] Dell’s own BlueField-2 firmware DUP targets ESXi 8.0 and Windows and is applied only through iDRAC/Lifecycle Controller, which shows the Dell-sourced vSphere path exists and is managed Dell’s way.[13]

The FAE position follows from those two documents: on PowerEdge, DSE is a Dell-SKU story; an NVIDIA-channel or consignment BlueField-3 in a VMware farm runs in NIC mode, behaving like a SuperNIC, unless Dell qualifies that DPU SKU for DSE.[11][12][8]

8:30, and what actually gets signed

How it ended

You take them in order. The SuperNIC is the B3140H: it “comes set in Super NIC mode”, runs on slot power, and offers no Arm console — the Arm cores are inactive.[1][7] The Partner DPU boots in DPU mode and needs the 8-pin auxiliary cable, like every 150 W BlueField-3 DPU SKU.[7] On DSE you do not improvise: Dell has not validated it on NVIDIA Channel cards and is unable to support it if enabled.[11] What you say: “Have Dell confirm in writing whether this consignment line counts as a Channel card.” The architect signs; the operator’s label maker tags every auxiliary cable “AUX — DO NOT PULL”. Eleven days later that rack takes a BIOS update, and at 03:40 the phone rings: the DPU is gone.

Lab

  1. Identify the platform: sudo dmidecode -s system-product-name — expect a PowerEdge model string. Record it in your evidence notes.
  2. Enumerate: lspci -nn | grep -i -E "mellanox|15b3" — expect the BlueField-3 PCI functions. If nothing appears, stop and check riser, slot and aux cable before touching software.[2][10]
  3. sudo mst start && sudo mst status -v — record the /dev/mst/ device path.
  4. sudo mlxconfig -d /dev/mst/<dev> q INTERNAL_CPU_OFFLOAD_ENGINE — 0 is DPU mode, 1 is NIC mode. Compare with the SKU’s expected default (DPU SKU = 0, B3140H = 1) and note any mismatch; do not change it.[8][1]
  5. sudo mlxfwmanager --query — record the part number/OPN and firmware version; match the OPN to the NVIDIA specifications page (e.g., B3220 900-9D3B6-00CV-AA0, B3140H 900-9D3D4-00EN-HA0).[6]
  6. Open the chassis (or use iDRAC photos if you have them) and confirm the 8-pin PCIe cable is fitted on any 150 W SKU. Record yes/no. This lab changes nothing.[7]

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 four sentences, why the same BlueField-3 silicon can appear on a quote as a $0 'Partner DPU' line or as a priced 'SuperNIC' line, and what that difference means for who supports it and which mode it boots in.

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

Sources

Facts in this lesson were checked against Dell KB 000227031 / 000225111 / 000300192, Dell R770 and R7725 ISM DPU pages, Dell Partner DPU 352-BBFH page, DOCA 3.5.0 modes page, 2026-09-06. Dates are when each page was fetched.

  1. Dell KB 000227031 — 'No Memory Found' on NVIDIA BlueField-3 enabled PowerEdge during POST · fetched 2026-09-06
  2. Dell PowerEdge R770 Installation and Service Manual — DPU specifications · fetched 2026-09-06
  3. Dell PowerEdge R7725 Installation and Service Manual — DPU specifications · fetched 2026-09-06
  4. Dell Partner DPU, NVIDIA BlueField-3 Dual Port 400GbE (352-BBFH) · fetched 2026-09-06
  5. Dell PowerEdge XE9780 Installation and Service Manual (dpu-specifications URL; rendered navigation only) · fetched 2026-09-06
  6. NVIDIA BlueField-3 Networking Platform User Guide — Specifications · fetched 2026-09-06
  7. NVIDIA BlueField-3 Networking Platform User Guide — Introduction · fetched 2026-09-06
  8. BlueField Modes of Operation · fetched 2026-09-06 · DOCA 3.5.0
  9. NVIDIA HGX AI Factory Enterprise Reference Architecture — Components · fetched 2026-09-06
  10. NVIDIA forum — BlueField-3 (DPU mode) stuck in FW pre-initializing, no host netdevs (root cause: ATX power not detected) · fetched 2026-09-06
  11. Dell KB 000225111 — NVIDIA Channel DPU VMware DSE Support · fetched 2026-09-06
  12. Broadcom KB 379391 — Configuring BlueField-3 into DPU mode for vSphere DSE or NIC mode · fetched 2026-09-06
  13. Dell driver 4RR0M — NIC firmware for NVIDIA BlueField-2 DPUs (DUP) · fetched 2026-09-06
  14. Dell KB 000300192 — XE9680L: power cycle required after updating DPU firmware · fetched 2026-09-06
  15. NVIDIA Spectrum-X available from Dell, HPE, Lenovo (Nov 20 2023) · fetched 2026-09-06
  16. Dell Technologies Unveils Next Generation Enterprise AI Solutions with NVIDIA (May 19 2025) · fetched 2026-09-06

The same idea elsewhere

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