Nine times out of ten the card is fine. The flash fails because the card carries an original-vendor PSID — a firmware identity written for Dell, HPE or Lenovo — and the standard NVIDIA update package will not write over it. Read the PSID first, then decide whether you need a flash at all.
What a PSID Actually Identifies
PSID stands for Parameter Set ID: sixteen characters burned into the card’s flash that describe the configuration the image was built for. Not a serial number, not a part number. It records port count, which line rates the card may negotiate, PCIe lane count, whether the silicon is VPI (InfiniBand and Ethernet) or Ethernet only, and which features the image enables.
That is why similar names do not share images. A dual-port MCX516A-CCAT is PCIe 3.0 x16 silicon; the dual-port MCX516A-CDAT is ConnectX-5 Ex on PCIe 4.0 x16. Write a single-port PSID onto a dual-port card and you get a card reporting ports it cannot light.
Three commands read the identity. On Linux with the firmware tools installed, mst start then flint -d <device> q prints PSID, firmware version, GUIDs and device ID; mlxfwmanager –query returns the same fields. Save that output — it is the only thing worth comparing against a listing.
| Field | Where to read it | What it decides |
|---|---|---|
| PSID, 16 characters | flint -d q or mlxfwmanager –query | Which firmware images the card will accept at all |
| Firmware version, e.g. 16.35.2000 | Same output | Which branch you are on — not what the card can do |
| Part number, e.g. MCX516A-CCAT | Label plus query output | Port count, line rate, PCIe generation |
| Link type, IB or ETH on VPI parts | mlxconfig -d q | A runtime setting you change without flashing |
| original-vendor subsystem ID | lspci -nn | Which original-vendor built it, if any |
Why the Update Refuses to Run
The burn tool compares the image’s PSID against the card’s before writing a byte. When they differ it stops with a mismatch error — a safety interlock, not a defect. The image was built for a different card shape.
Four things explain it. An original-vendor card behind a generic package is the common one. A card already on the newest image for its branch reports nothing newer — the end of that branch, not a fault. When mlxfwmanager sees no devices, mst was never started or the inbox driver grabbed the card. And an image built for the wrong port count or PCIe width stops there for good reason: that is the mistake that damages hardware.
Firmware does not raise line rate. A 25GbE port is 25GbE because of the ASIC and port configuration. Updates fix link training, FEC, PXE and UEFI issues and security items. If a card links at the wrong speed, look at the transceiver, the cable or the switch port — we walked through that in running a 25GbE card on a 100GbE switch port.
Why original-vendor Cards Behave Differently
Dell, HPE and Lenovo buy these cards against their own PSID and firmware branch. The card then reports an original-vendor subsystem ID in lspci -nn, and the vendor’s package lists your PSID as not applicable. The tooling that does know it belongs to the platform: iDRAC and Lifecycle Controller on Dell, iLO with the support pack on HPE, XClarity on Lenovo.
In a same-brand chassis this is an advantage: inventory treats the card as a known part and the out-of-band controller updates it with everything else. Moved into a whitebox or another brand’s server it still works at the OS level — driver binds, ports link, RDMA runs. Only the firmware path belongs to someone else.
original-vendor branches also ship their own defaults for PXE, SR-IOV, wake-on-LAN and power. If a used card acts oddly in a new host, read mlxconfig -d <device> q and reset it before assuming the card is faulty.
When Forcing a Cross-Flash Is Worth It
You can override the interlock: flint -d <device> -i image.bin -allow_psid_change burn. You gain one firmware branch across a mixed fleet and the feature flags it carries. You give up the original-vendor support path, the label no longer matches what the card reports, and any feature the original-vendor branch exposed but the generic one does not quietly disappears.
The real risk is image shape. A single-port image on a dual-port card, or an x8 image on an x16 design, can leave a card that never trains again. Dump the flash first with flint -d <device> e backup.bin, record the original PSID, and keep a matching card within reach.
Do not flash for its own sake. If the card works, sits on the current image for its branch and lives in a same-brand server, leave it alone. A flash earns its risk when it fixes a bug you can name.
What to Ask Before You Buy Used
Ask four things and you avoid most of this. The text output of flint -d q or mlxfwmanager –query, so you can read PSID and version yourself. The lspci -nn line, which reveals the subsystem ID. A photo of the label showing part number and PSID sticker. And what is in the box — both bracket heights, and above 25GbE the transceiver type the ports expect.
Ask one more that saves arguments: is this a VPI part or an Ethernet-only part? On cards such as the MCX653105A-ECAT the answer changes what it plugs into, and it is written in the PSID long before anyone reads the label. Our test line queries every card the same way before listing, and that output is what we send when a buyer asks about firmware.
For 25GbE the two generations sit side by side in our stock: the MCX4121A-ACAT on ConnectX-4 Lx silicon and the MCX512A-ACAT on ConnectX-5. Both are PCIe 3.0 x8 with two SFP28 cages, so the choice is offload and driver branch, not bandwidth. For the 100GbE pair the slot decides it — see MCX516A-CCAT versus MCX516A-CDAT — and for HDR100 half-rate parts see MCX653105A-HDAT versus MCX653105A-ECAT.
The Mellanox cards listed on zhongxindata.hk are quoted per part number, and availability is confirmed per model and per quantity. Pairing them with a switch? The SN2700-CS2F 100GbE switch takes 100G QSFP28 directly and splits to 25G.
Frequently Asked Questions
Why won’t my used Mellanox card take a new firmware?
Most often because the adapter only accepts a specific firmware progression and will not jump directly to the newest release. Moving through the intermediate image in order resolves it. Other causes are an image that does not cover your adapter variant, or the card being in the wrong burn mode. If the card is genuinely faulty, the flash failure is itself a useful diagnostic.
Does a failed firmware update mean the card is broken?
Not usually. A refused update is more often an image or sequencing problem than a hardware fault. That said, a card that fails every valid flash path is worth treating as suspect, and because we re-flash during testing we know whether the unit accepts firmware before we list it. We report the firmware level the unit ships with.
How do I find the right firmware for my Mellanox adapter?
Match the image to the exact adapter type and variant, not just the family name — the suffix identifies the variant and the image covers a specific set. We confirm the adapter type, port configuration and firmware level during testing and state all three, so you can select the correct image without guessing.
Is it safe to flash a used Mellanox card?
It is a normal procedure, provided the image matches the adapter and you follow the sequence the card requires. The risk is writing the wrong image rather than the act of flashing. We re-flash during testing on our own test line, which also confirms the card’s flash path works before the unit is listed.
What firmware level do used adapters ship with?
We re-flash during testing and state the level each unit ships with, so you know the starting point rather than inheriting whatever the previous owner left behind. If you are standardising a fleet on a specific version, tell us what you need and we will confirm what we can ship.
How do you verify a used Mellanox adapter?
We clean, re-flash and bench-test on our own test line, then confirm the part number, port count, negotiated link rate and firmware level, and we read the error counters so a card that links but produces errors does not get listed. Ask for photos of the exact unit and its firmware level before you commit.
Send the part number and quantity and we will check availability. WhatsApp +86 13068616914 · zhongxindata.hk
The intermediate image step, which is the usual answer
- Some adapters accept only a specific firmware progression and reject a direct jump to the newest release
- Moving through the intermediate image in order is the standard remedy
- This is a known behaviour of the firmware packages, not a sign of a failed card
Other reasons a flash can be refused
- An adapter type or variant that the image does not cover will refuse the update
- The burn or the image mode matters — a card in the wrong mode will not accept the write
- A card with a genuine hardware fault will fail the flash, which is why the flash result itself is a useful test
What to check before and after flashing
- Confirm the adapter type and the current firmware version before you start
- Have the correct image for your exact variant, and the correct burn utility for it
- After a successful flash, confirm the version and check the error counters
Related reading
- MCX516A-CCAT adapter — a ConnectX-5 unit whose firmware path matters
- ConnectX adapter comparison — the model context
- 7-point verification checklist — the firmware validation step in context
- Network adapters — category hub
Frequently Asked Questions
Why won’t my used Mellanox card take a new firmware?
Most often because the adapter only accepts a specific firmware progression and will not jump directly to the newest release. Moving through the intermediate image in order resolves it. Other causes are an image that does not cover your adapter variant, or the card being in the wrong burn mode. If the card is genuinely faulty, the flash failure is itself a useful diagnostic.
Does a failed firmware update mean the card is broken?
Not usually. A refused update is more often an image or sequencing problem than a hardware fault. That said, a card that fails every valid flash path is worth treating as suspect, and because we re-flash during testing we know whether the unit accepts firmware before we list it. We report the firmware level the unit ships with.
How do I find the right firmware for my Mellanox adapter?
Match the image to the exact adapter type and variant, not just the family name — the suffix identifies the variant and the image covers a specific set. We confirm the adapter type, port configuration and firmware level during testing and state all three, so you can select the correct image without guessing.
Is it safe to flash a used Mellanox card?
It is a normal procedure, provided the image matches the adapter and you follow the sequence the card requires. The risk is writing the wrong image rather than the act of flashing. We re-flash during testing on our own test line, which also confirms the card’s flash path works before the unit is listed.
What firmware level do used adapters ship with?
We re-flash during testing and state the level each unit ships with, so you know the starting point rather than inheriting whatever the previous owner left behind. If you are standardising a fleet on a specific version, tell us what you need and we will confirm what we can ship.
How do you verify a used Mellanox adapter?
We clean, re-flash and bench-test on our own test line, then confirm the part number, port count, negotiated link rate and firmware level, and we read the error counters so a card that links but produces errors does not get listed. Ask for photos of the exact unit and its firmware level before you commit.