NVIDIA Networking Firmware (MFT)

The NVIDIA Firmware Tools family (mst, flint, mlxfwmanager, mlxconfig, mlxfwreset, mlxlink, mlxcables), PSIDs, the NIC firmware lifecycle (query, burn, activate, verify), BlueField bundles, InfiniBand and Ethernet switch upgrades, and keeping a whole cluster on one recipe. Complements the HGX firmware guide, which covers the server side. Objectives 1.4, 2.1, and 4.6.

1. The big picture: firmware across the whole cluster

An AI cluster runs firmware in every box: server BIOS and BMC, the HGX GPU baseboard, every NIC and DPU, every switch, every active cable or transceiver, every drive. Different tools update different parts, and every part has its own activation rule.

2. Interactive firmware matrix: which tool for which component

3. The MFT tool family

MFT (NVIDIA Firmware Tools, formerly Mellanox Firmware Tools) is a package of command-line tools for NVIDIA networking devices: ConnectX adapters, BlueField DPUs/SuperNICs, and switches.

Less common tools in the same package that still show up in procedures:

Debug utilities. You'll mostly run them when NVIDIA support asks for their output:

4. mst: finding devices

sudo mst start             # load access modules, create /dev/mst/* devices
sudo mst status -v         # devices with PCI address, RDMA name (mlx5_X), netdev
# DEVICE_TYPE        MST                           PCI       RDMA     NET
# ConnectX7(rev:0)   /dev/mst/mt4129_pciconf0      17:00.0   mlx5_0   net-ibp23s0
# BlueField3(rev:1)  /dev/mst/mt41692_pciconf0     a1:00.0   mlx5_8   net-ens7f0np0

sudo mst ib add            # add InfiniBand switches reachable in-band (for flint -d lid-...)

Output is illustrative; columns vary by MFT version.

• Most MFT tools also accept a PCI address (-d 17:00.0) instead of the /dev/mst path.
• A dual-port adapter appears as one device for firmware purposes: one firmware image serves both ports. Updating it affects both.
• mst status showing nothing usually means mst start wasn't run, or MFT isn't installed for this kernel.

mst subcommands and options

5. PSIDs and OEM cards

Every adapter carries a PSID (Parameter-Set ID), for example MT_0000000XXX for an NVIDIA-branded card. It identifies the exact board variant and its configuration defaults. Firmware images are built per PSID.

• flint and mlxfwmanager check that the image's PSID matches the card's PSID and refuse a mismatch by default.
• Cards sold by server vendors (OEM cards) have OEM PSIDs. Their firmware comes from the server vendor's support site, not NVIDIA's generic downloads.
• flint ... --allow_psid_change forces a different PSID's image. It can remove OEM-specific settings, break support, or leave the card unusable. Only do it under explicit vendor or NVIDIA support direction.

6. flint: query, burn, verify

sudo flint -d /dev/mst/mt4129_pciconf0 q
# Image type:            FS4
# FW Version:            28.AA.BBBB
# FW Version(Running):   28.XX.YYYY        <- only shown when a newer image is burned but not yet active
# PSID:                  MT_0000000XXX
# Security Attributes:   secure-fw
# Base GUID / Base MAC:  ...

sudo flint -i fw-ConnectX7-<version>.bin q          # query an image FILE: version and PSID before burning
sudo flint -d /dev/mst/mt4129_pciconf0 -i fw-ConnectX7-<version>.bin burn
sudo flint -d /dev/mst/mt4129_pciconf0 verify        # check the image on flash is intact
sudo flint -d /dev/mst/mt4129_pciconf0 ri backup.bin # read the current image to a file (where permitted)

sudo flint -d lid-<switch-LID> q                     # in-band query of an InfiniBand switch (after mst ib add)

Version numbers and output fields are placeholders; the layout varies by device and MFT version.

• Burning writes flash; it does not change the running firmware. The card keeps running the old version until it's reset (mlxfwreset) or the server is rebooted/power-cycled.
• FW Version(Running) differing from FW Version means an activation is pending.
• Secure firmware (secure-fw) devices only accept NVIDIA-signed images; a modified or wrong image is rejected.
• Abbreviations: q = query, b = burn, v = verify. -y answers yes to prompts (scripted use).
• Exit codes: 0 success, 1 error, 7 the burn was declined at the prompt. Scripts that loop over nodes should check them.

flint subcommands

flint flags you need to know

Signing commands (sign, rsa_sign, --private_key, --hsm, set_public_keys, set_forbidden_versions) are for building custom signed images and aren't part of routine operations.

7. mlxfwmanager: package-based updates

sudo mlxfwmanager --query                  # every device: type, PSID, current version, available version, status
# Device Type:      ConnectX7
# Part Number:      <part-number>
# PSID:             MT_0000000XXX
# Versions:         Current        Available
#    FW             28.XX.YYYY     28.AA.BBBB
# Status:           Update required

sudo mlxfwmanager -u -i fw-ConnectX7-<version>.bin     # update from one image
sudo mlxfwmanager -u -D /path/to/fw-images/ -y          # pick the right image per device from a directory
sudo mlxfwmanager --online -u                           # download matching firmware from NVIDIA (needs internet)

• Walks all devices on the host and matches each to an image by PSID.
• The vendor-provided self-extracting firmware packages for a platform often wrap mlxfwmanager.
• Air-gapped clusters: stage the images on the head node or an internal share and use -D; --online needs internet access.
• After an update it tells you to reboot (or use mlxfwreset) to load the new firmware.

mlxfwmanager options

• mlxfwmanager only scans PCI devices on the host. In-band switches (lid-N) go through flint or UFM.
• The query also lists PXE and UEFI expansion-ROM versions under FW; an update can change those too.
• With several adapters, recent versions query and update them in parallel through the fwctl kernel driver (upstream Linux 6.15+, or from DOCA) when it's loaded.

8. flint vs mlxfwmanager

9. mlxconfig: NIC settings in non-volatile memory

mlxconfig changes settings stored on the adapter itself. They survive OS reinstalls and apply at the next firmware load.

sudo mlxconfig -d /dev/mst/mt4129_pciconf0 -e q          # all settings: Default / Current / Next Boot
sudo mlxconfig -d /dev/mst/mt4129_pciconf0 q LINK_TYPE_P1 LINK_TYPE_P2
sudo mlxconfig -d /dev/mst/mt4129_pciconf0 set LINK_TYPE_P1=2        # 1 = InfiniBand, 2 = Ethernet (VPI cards)
sudo mlxconfig -d /dev/mst/mt4129_pciconf0 set SRIOV_EN=1 NUM_OF_VFS=8
sudo mlxconfig -d /dev/mst/mt4129_pciconf0 reset                     # restore defaults (careful)

• Changes land in the Next Boot column. They apply after mlxfwreset, a reboot, or (for some settings) a full power cycle.
• Every node of a type should have the same settings; include mlxconfig queries in your inventory.

mlxconfig commands and options

Parameters you'll meet often: LINK_TYPE_P1/P2, SRIOV_EN, NUM_OF_VFS (the firmware caps it; 127 on current adapters), ATS_ENABLED, ADVANCED_PCI_SETTINGS, PCI_ATOMIC_MODE, INTERNAL_CPU_OFFLOAD_ENGINE (BlueField DPU vs NIC mode), EXP_ROM_*_ENABLE (PXE/UEFI boot ROMs). show_confs is the source of truth for a given card.

10. mlxfwreset: activating without a reboot

mlxfwreset resets the adapter so it loads the newly burned firmware and pending mlxconfig settings, without rebooting the whole server.

sudo mlxfwreset -d /dev/mst/mt4129_pciconf0 query      # supported reset levels/types and what each affects
sudo mlxfwreset -d /dev/mst/mt4129_pciconf0 reset      # default (recommended) level
sudo mlxfwreset -d /dev/mst/mt4129_pciconf0 -l <level> -y reset   # a specific level, no prompt

• Reset levels range from a live update that keeps the driver and links up, through a driver restart with PCI reset, to a warm reboot. Which ones work depends on the device, firmware, and server platform. query tells you; don't memorize numbers.
• The reset affects the whole device: both ports, every PCI function, every VF. RDMA traffic, storage mounts, and IPoIB on those ports are interrupted.
• On BlueField, some resets also affect the Arm side; follow the DPU documentation.
• If mlxfwreset isn't supported on the platform, fall back to a reboot. Some changes need a cold reboot or power cycle; a warm reboot may keep the adapter powered and running old firmware on some servers.

mlxfwreset options

mlxfwreset -d <dev> q                                   # supported levels / types / sync, and the default
mlxfwreset -d <dev> status [--json]                     # what the pending change needs: level 3, warm reboot, or power cycle (beta)
mlxfwreset -d <dev> -y -l 3 --sync 2 --method 1 r       # hot-reset the PCIe switch above the card (ConnectX-7 / BlueField)
mlxfwreset -d <dev> -y reset                            # default level
mlxfwreset -d <dev> -y -l 3 -t 0 --sync 1 reset         # explicit level, type, sync
mlxfwreset -d <BlueField> -y -l 3 --sync 1 r            # BlueField: driver-owned sync (the BF2/BF3 default)

• A level that query reports as "Not Supported" fails with an error; pick one that's supported, or reboot.
• If flint finishes a burn without printing -I- To load new FW run mlxfwreset or reboot machine, a reboot is required.
• If mlxfwreset fails between "Stopping driver" and "Starting driver", the driver stays down: start it by hand.
• Some mlxconfig changes print -W- PCI rescan is required after device reset.
• Switches: no level/type/sync options, and it doesn't work in-band or remotely (for unmanaged switches use flint -d lid-N swreset). It also doesn't run on NVOS, SONiC, or Cumulus.
• BlueField: level 3 with sync 1. On BlueField-2 and BlueField-3, sync 1 is the default and sync 0 isn't supported. Level 4 alone needs mlxfwreset on both the host and the Arm side; the Arm must be shut down first.

11. The NIC firmware lifecycle

1. Query: mlxfwmanager --query (or flint q): current version, PSID, part number.
2. Get the image for that PSID at the version your recipe specifies (DGX/OEM release notes; NIC firmware must match the DOCA/OFED driver's supported range).
3. Drain the node.
4. Burn with mlxfwmanager -u or flint burn.
5. Activate with mlxfwreset or a reboot/power cycle.
6. Verify: running version, mlxconfig settings, link state and speed, then a quick fabric test (ib_write_bw, nccl-tests) before returning the node to service.

12. Hands-on lab: six simulated tickets

A mock cluster you work like the real thing. Pick a ticket, start with cat notes.txt, and type help on any host for its commands. State carries through: burned firmware stays pending until activated, mlxconfig changes sit in Next Boot, resetting a busy node fails its job, rebooted switches drop out of the fabric for a moment, and a reinstalled switch comes back with a blank config.

• FW-1024: one node, ConnectX-7 and BlueField-3 firmware plus a port type fix (guided; start here).
• INC-2087: from the head node, use pdsh and dshbak to find three different stragglers in an 8-node fleet and fix them.
• NET-3307: InfiniBand switches: in-band flint for unmanaged leaves, NVOS for the managed spine, one at a time, then ibdiagnet.
• NET-4410: Ethernet switch OS: a Cumulus MLAG pair through ONIE (backup and restore) and a SONiC switch.
• DPU-5150: a BlueField-3 stuck in NIC mode, then a BFB bundle upgrade over RShim.
• CBL-6060: LinkX transceiver firmware with flint --linkx: a leaf switch's modules in two halves (range burn), a port that doesn't come back by itself, and the host-side module with --linkx_auto_update.

13. BlueField: BFB bundles, NIC firmware, modes

A BlueField DPU has two worlds: the NIC part (ConnectX firmware, like any adapter) and the Arm computer running its own OS (Ubuntu with DOCA) plus boot firmware (ATF/UEFI). BlueField-3 also has its own BMC.

ls /dev/rshim*                                  # RShim devices on the host (rshim service running)
sudo bfb-install --bfb <bf-bundle>.bfb --rshim rshim0
# then follow the release notes: usually a host power cycle so the new NIC firmware loads
sudo flint -d /dev/mst/mt41692_pciconf0 q       # confirm the NIC firmware version afterwards

• A BFB bundle keeps the Arm OS, DOCA, and NIC firmware versions consistent with each other: the preferred way to move a DPU to a new release.
• .bfb is a BlueField image, not a server firmware package (.fwpkg is for DGX/HGX).
• Don't restart the host or the DPU while bfb-install runs. An interrupted install can leave the Arm side unreachable; rerun bfb-install to recover.

bfb-install options

• Check the result on the Arm side with bfver or cat /etc/mlnx-release, and the NIC with flint q. flint -d <dev> query_bfb_components lists BFB component versions from the host.
• More on DPU modes and the Arm side in the bring-up guide's BlueField objective.

14. InfiniBand switch firmware

ibswitches                         # list switches: GUID, LID, description
sudo mst ib add                    # make in-band switch devices visible to MFT
sudo flint -d lid-<LID> -qq q      # switch firmware version and PSID (-qq: quick query)

Two versions to confirm: switch OS and ASIC firmware

A managed switch has a switch OS (NVOS, or MLNX-OS / Onyx on older systems) running on its management CPU, and firmware on the switch chip (the ASIC) that actually forwards packets. Objective 4.6 asks you to confirm both against the cluster's recipe. An unmanaged switch has no OS, only ASIC firmware.

# NVOS
admin@ib-leaf01:~$ nv show platform firmware ASIC
                 operational          applied
---------------  -------------------  -------
part-number      MQM9700-NS2X-NS_Ax
actual-firmware  31.2014.1000
auto-update      enabled              enabled
fw-source        default              default

# MLNX-OS / Onyx
switch (config) # show asic-version
===================================================
 Module      Device      Version
===================================================
 MGMT        QTM         31.2014.1000
switch (config) # show asic-version | include QTM     # filter like grep

• Upgrade one switch at a time (or one per redundancy group) and let the fabric reconverge. A rail-optimized fabric has redundancy at the spine layer, but a leaf reboot takes down that rail for its nodes.
• Keep every switch of the same role on the same version; version mismatches cause odd adaptive-routing or SHARP behavior.
• The Subnet Manager (UFM / OpenSM) version is part of the recipe too.

15. Ethernet switch OS upgrades

Spectrum Ethernet switches run a network OS that includes the ASIC firmware. You upgrade the whole OS image, not the firmware separately.

• Back up the configuration first (Cumulus: NVUE config; SONiC: config_db.json).
• Upgrade redundant pairs one at a time; check BGP/MLAG, RoCE QoS, and link state after each (see the RoCE guide's switch section).
• For Spectrum-X, switch OS, SuperNIC firmware, and DOCA versions are released together; upgrade them as a set.

16. LinkX cable and transceiver firmware (flint --linkx)

Optical transceivers and active optical cables (AOCs) run their own firmware, separate from both NIC and switch firmware. NVIDIA calls the in-field update IFFU: flint sends the image to the NIC or switch, and that device passes it to the modules plugged into it.

• Through a switch (-d lid-N, in-band, works on unmanaged switches) or a host NIC (-d /dev/mst/..., over the PCI register interface). Managed switches can also run it from the switch itself.
• Each device only updates the modules plugged into it. A link has two ends: the far end (the other switch, or the host NIC) needs its own run.
• --downstream_device_ids and the range options use port numbers (from 1). Passive copper (DAC) has no firmware and shows as not present.
• Two stages: transfer (download the image into the module's inactive bank) and activate (switch banks; the link flaps). Neither happens unless you ask for it.
• You can't burn the version a module already has. That fails with Cable burn failed, error is LinkX downstream transfer failed for device index N, which saves time in large clusters.
• Plan about 1.5 minutes per module for the transfer, plus about 30 seconds to activate.

# 1. Query: firmware banks, running image, versions (ports 1, 2, 5)
sudo flint -d lid-1 --linkx --downstream_device_ids 1,2,5 q          # comma-separated, NO spaces
sudo flint -i <cable-image>.bin q                                    # version and part number in the image file

# 2a. Range: half the ports first (here ports 1-4), transfer then activate
sudo flint -d lid-1 --linkx --downstream_device_id_start_index 1 --num_of_downstream_devices 4 --download_transfer -i <img> b
sudo flint -d lid-1 --linkx --downstream_device_id_start_index 1 --num_of_downstream_devices 4 --activate b
for p in 1 2 3 4; do sudo mlxlink -d lid-1 -p $p -m | grep -E "State|Part|FW"; done   # links back up?

# 2b. Auto-update: every supported module on the device in one go
sudo flint -d <dev> --linkx --linkx_auto_update --download_transfer -i <img> b
sudo flint -d <dev> --linkx --linkx_auto_update --activate b
sudo flint -d <dev> --linkx --linkx_auto_update --download_transfer --activate -i <img> b   # both in one command
sudo flint -d <dev> --linkx --linkx_auto_update --activate --activate_delay_sec 10 b      # delayed activation

mlxlink and mlxcables for modules

sudo mlxlink -d mlx5_0 -m                          # host NIC port: state, module part number, FW version
sudo mlxlink -d lid-1 -p 7 -m                      # switch port 7, in-band
sudo mlxlink -d lid-1 -p 7 -a TG                   # toggle the port (DN / UP / TG)
for i in $(seq 1 40); do echo $i; sudo mlxlink -d lid-1 -p $i -m | grep -E "State|Part|FW"; done
sudo mlxlink -d mlx5_0 --cable --ddm               # DDM: temperature, voltage, Rx/Tx power, bias
sudo mlxlink -d mlx5_0 --cable --dump              # raw EEPROM pages
sudo mst cable add; sudo mlxcables -q              # cable devices; query them all
sudo mlxcables -d <cable dev> --DDM               # DDM through mlxcables
sudo mlxcables -d <cable dev> -r -p 0 -o 165 -l 3 # read bytes (page 0, offset 165, length 3)

mlxcables only reads and writes module memory, and gets no new features; mlxlink --cable is the way forward. Its options: -q (query all cables), --DDM, --dump, -r/-w with -p <page> -o <offset> -l <len> (or -a <address>), -b <bytes per line>, --print_raw, and --read_reg/--read_all_regs/--show_all_regs to decode named fields. Page and offset follow SFF-8636 for QSFP and SFF-8472 for SFP (the A0/A2 addresses).

17. Fleet inventory and drift

A cluster stays healthy when every node of a type runs the same versions. Inventory regularly and after every maintenance window.

# Parallel query across nodes (pdsh, or BCM's parallel execution from the head node)
pdsh -w node[001-064] 'mlxfwmanager --query | grep -E "Device Type|PSID|FW "' | dshbak -c
pdsh -w node[001-064] 'flint -d /dev/mst/mt4129_pciconf0 q | grep -E "FW Version|PSID"' | dshbak -c
# Switches
ibswitches; for l in <LIDs>; do flint -d lid-$l q | grep -E "FW Version|PSID"; done

• dshbak -c groups nodes with identical output: one group means no drift; several groups show the outliers.
• Record a per-node inventory (NIC firmware, mlxconfig settings, switch versions, module firmware) next to the server-side inventory from the BMC.
• In BCM, install MFT in the software image so every node has the same tool version, and keep firmware images on the head node for air-gapped updates.

18. Interactive drift checker

Paste lines of node device version psid (any whitespace). The checker finds the most common version per device type and PSID, and flags outliers. Example data uses placeholder names.

19. Failures and recovery

20. Sample outputs

What the commands in this guide print. Expand a command to see a typical result and what to check in it.

Starting MST (Mellanox Software Tools) driver set
Loading MST PCI module - Success
Loading MST PCI configuration module - Success
Create devices
Unloading MST PCI module (unused) - Success

MST modules:
------------
    MST PCI module is not loaded
    MST PCI configuration module loaded
PCI devices:
------------
DEVICE_TYPE             MST                           PCI       RDMA            NET                                     NUMA
ConnectX7(rev:0)        /dev/mst/mt4129_pciconf0      18:00.0   mlx5_0          net-ibp24s0                             0
ConnectX7(rev:0)        /dev/mst/mt4129_pciconf1      40:00.0   mlx5_3          net-ibp64s0                             0
ConnectX7(rev:0)        /dev/mst/mt4129_pciconf2      4f:00.0   mlx5_4          net-ibp79s0                             0
ConnectX7(rev:0)        /dev/mst/mt4129_pciconf3      5e:00.0   mlx5_5          net-ibp94s0                             0
...
BlueField3(rev:1)       /dev/mst/mt41692_pciconf0     29:00.0   mlx5_1          net-enp41s0f0np0                        0
BlueField3(rev:1)       /dev/mst/mt41692_pciconf0.1   29:00.1   mlx5_2          net-enp41s0f1np1                        0
BlueField3(rev:1)       /dev/mst/mt41692_pciconf1     aa:00.0   mlx5_7          net-enp170s0f0np0                       1
...

-I- Discovering the fabric - Running: /usr/sbin/ibnetdiscover -s
-I- Added 3 in-band devices

Image type:            FS4
FW Version:            28.43.1014
FW Version(Running):   28.39.2048
FW Release Date:       4.9.2024
Product Version:       28.43.1014
Rom Info:              type=UEFI version=14.36.16 cpu=AMD64,AARCH64
                       type=PXE version=3.7.500 cpu=AMD64
Description:           UID                GuidsNumber
Base GUID:             000000000a010001        8
Base MAC:              00000a010001            8
Image VSD:             N/A
Device VSD:            N/A
PSID:                  MT_0000000A01
Security Attributes:   secure-fw

Image type:            FS4
FW Version:            28.43.1014
FW Release Date:       4.9.2024
Product Version:       28.43.1014
Rom Info:              type=UEFI version=14.36.16 cpu=AMD64,AARCH64
                       type=PXE version=3.7.500 cpu=AMD64
Description:           UID                GuidsNumber
Base GUID:             N/A                     8
Base MAC:              N/A                     8
Image VSD:             N/A
Device VSD:            N/A
PSID:                  MT_0000000A01
Security Attributes:   secure-fw

    Current FW version on flash:  28.39.2048
    New FW version:               28.43.1014

    FSMST_INITIALIZE - OK
    Writing Boot image component - OK
-I- To load new FW run mlxfwreset or reboot machine.

FS4 failsafe image

     /0x00000000-0x0000001f (0x000020)/ (HW_POINTERS) - OK
     /0x00010000-0x0001ffff (0x010000)/ (BOOT2) - OK
     /0x00500000-0x00dfffff (0x900000)/ (ITOC) - OK
     ...
-I- FW image verification succeeded. Image is bootable.

Image type:            FS5
FW Version:            31.2012.1000
FW Release Date:       14.2.2024
Description:           UID                GuidsNumber
Base GUID:             000000000000f001        1
Base MAC:              N/A                     1
PSID:                  MT_0000000S01
Security Attributes:   secure-fw

Querying Mellanox devices firmware ...

Device #1:
----------

  Device Type:      ConnectX7
  Part Number:      MCX75310AAS-NEAT_Ax
  Description:      NVIDIA ConnectX-7 HHHL Adapter Card; 400GbE / NDR IB (default mode); Single-port OSFP; PCIe 5.0 x16
  PSID:             MT_0000000A01
  PCI Device Name:  /dev/mst/mt4129_pciconf0
  Base GUID:        000000000a010001
  Versions:         Current        Available
     FW             28.39.2048     28.43.1014
     PXE            3.7.0500       3.7.0500
     UEFI           14.36.0016     14.36.0016

  Status:           Update required

Device #2:
----------

  Device Type:      ConnectX7
  Part Number:      MCX75310AAS-NEAT_Ax
  Description:      NVIDIA ConnectX-7 HHHL Adapter Card; 400GbE / NDR IB (default mode); Single-port OSFP; PCIe 5.0 x16
  PSID:             MT_0000000A01
  PCI Device Name:  /dev/mst/mt4129_pciconf1
  Base GUID:        000000000a010009
  Versions:         Current        Available
     FW             28.43.1014     28.43.1014
     PXE            3.7.0500       3.7.0500
     UEFI           14.36.0016     14.36.0016

  Status:           Up to date

...

Querying Mellanox devices firmware ...

Device #1:
----------

  Device Type:      ConnectX7
  Part Number:      MCX75310AAS-NEAT_Ax
  PSID:             MT_0000000A01
  PCI Device Name:  /dev/mst/mt4129_pciconf0
  Versions:         Current        Available
     FW             28.39.2048     28.43.1014

  Status:           Update required

...
---------
Found 3 device(s) requiring firmware update...

Device #1: Updating FW ...
FSMST_INITIALIZE - OK
Writing Boot image component - OK
Done
Device #2: Updating FW ...
...
Done

Restart needed for updates to take effect.

Device #1:
----------

Device type:        ConnectX7
Name:               MCX75310AAS-NEAT_Ax
Description:        NVIDIA ConnectX-7 HHHL Adapter Card; 400GbE / NDR IB (default mode); Single-port OSFP; PCIe 5.0 x16
Device:             /dev/mst/mt4129_pciconf0

Configurations:                                          Default             Current             Next Boot
        MEMIC_BAR_SIZE                              0                   0                   0
        MEMIC_SIZE_LIMIT                            _256KB(1)           _256KB(1)           _256KB(1)
        ...
        NUM_OF_VFS                                  0                   0                   0
        SRIOV_EN                                    False(0)            False(0)            False(0)
        LINK_TYPE_P1                                IB(1)               IB(1)               IB(1)
        ...
        ADVANCED_PCI_SETTINGS                       False(0)            False(0)            False(0)
        ATS_ENABLED                                 False(0)            False(0)            False(0)
        ...
The '*' shows parameters with next value different from default/current value.

Device #1:
----------

Device type:        ConnectX7
Name:               MCX75310AAS-NEAT_Ax
Description:        NVIDIA ConnectX-7 HHHL Adapter Card; 400GbE / NDR IB (default mode); Single-port OSFP; PCIe 5.0 x16
Device:             /dev/mst/mt4129_pciconf0

Configurations:                                          Next Boot
        LINK_TYPE_P1                                IB(1)
-E- The Device doesn't support LINK_TYPE_P2 parameter

Device #1:
----------

Device type:        ConnectX7
Name:               MCX75310AAS-NEAT_Ax
Description:        NVIDIA ConnectX-7 HHHL Adapter Card; 400GbE / NDR IB (default mode); Single-port OSFP; PCIe 5.0 x16
Device:             /dev/mst/mt4129_pciconf0

Configurations:                                          Next Boot       New
        SRIOV_EN                                    False(0)        True(1)
        NUM_OF_VFS                                  0               8

 Apply new Configuration? (y/n) [n] : y
Applying... Done!
-I- Please reboot machine to load new configurations.

Reset-levels:
0: Driver, PCI link, network link will remain up ("live-Patch")      -Not Supported
1: Only ARM side will not remain up ("Immediate reset")              -Not Supported
3: Driver restart and PCI reset                                       -Supported     (default)
4: Warm Reboot                                                        -Supported
Reset-type (relevant only for reset-levels 1,3,4):
0: Full chip reset                                                    -Supported     (default)
1: Phy-less reset (keep network port active during reset)            -Not Supported
2: NIC only reset (for SoC devices)                                   -Not Supported
3: ARM only reset                                                     -Not Supported
4: ARM OS shut down                                                   -Not Supported
Reset-sync (relevant only for reset-level 3):
0: Tool is the owner                                                  -Supported
1: Driver is the owner                                                -Supported     (default)
2: FW is the owner                                                    -Not Supported

The reset level for device, /dev/mst/mt4129_pciconf0 is:

3: Driver restart and PCI reset
Continue with reset?[y/N] y
-I- Sending Reset Command To Fw             -Done
-I- Stopping Driver                         -Done
-I- Resetting PCI                           -Done
-I- Starting Driver                         -Done
-I- Restarting MST                          -Done
-I- FW was loaded successfully.

Pushing bfb
1.42GiB 0:02:04 [11.7MiB/s] [================================>] 100%
Collecting BlueField booting status. Press Ctrl+C to stop...
 INFO[PSC]: PSC BL1 START
 INFO[BL2]: start
 INFO[BL2]: boot mode (rshim)
 INFO[BL31]: start
 INFO[UEFI]: UEFI loaded
 INFO[MISC]: Ubuntu installation started
 INFO[MISC]: Installation finished
 INFO[MISC]: Updating NIC firmware...
 INFO[MISC]: NIC firmware update done
 INFO[MISC]: Linux up
 INFO[MISC]: DPU is ready

Switch  : 0x000000000000f001 ports 129 "MF0;ib-leaf01:MQM9790/U1" enhanced port 0 lid 1 lmc 0
Switch  : 0x000000000000f002 ports 129 "MF0;ib-leaf02:MQM9790/U1" enhanced port 0 lid 2 lmc 0
Switch  : 0x000000000000f003 ports 129 "MF0;ib-spine01:MQM9700/U1" enhanced port 0 lid 3 lmc 0

Operational Info
----------------
State                              : Active
Physical state                     : LinkUp
Speed                              : IB-NDR
Width                              : 4x
FEC                                : Standard_RS-FEC - (544,514)
Loopback Mode                      : No Loopback
Auto Negotiation                   : ON

Supported Info
--------------
Enabled Link Speed                 : 0x00000080 (NDR)
Supported Cable Speed              : 0x000000c0 (HDR,NDR)

Troubleshooting Info
--------------------
Status Opcode                      : 0
Group Opcode                       : N/A
Recommendation                     : No issue was observed

Tool Information
----------------
Firmware Version                   : 28.43.1014
amBER Version                      : 3.4
MFT Version                        : mft 4.30.0-139

Module Info
-----------
Identifier                         : OSFP
Cable Technology                   : 1310 nm DFB
Cable Type                         : Optical Module (separated)
Vendor Name                        : NVIDIA
Vendor Part Number                 : MMA4Z00-NS400
Vendor Serial Number               : MT2400FT00001
FW Version                         : 47.120.0020
...

Module Info
-----------
Temperature [C]                    : 52
Voltage [mV]                       : 3291.3
Bias Current [mA]                  : 7.012,7.051,7.003,6.998
Rx Power Current [dBm]             : 1.21,0.98,1.05,1.12
Tx Power Current [dBm]             : 1.53,1.48,1.60,1.51
...

----------------
node[001-002,004-064]
----------------
FW Version:            28.43.1014
PSID:                  MT_0000000A01
----------------
node003
----------------
FW Version:            28.43.1014
FW Version(Running):   28.39.2048
PSID:                  MT_0000000A01

21. Scenarios

22. Searchable reference

23. Flashcards (click to flip)

24. Exam traps

25. Practice quiz (35 questions)