The ConnectX-4 is a QSFP+ 40GbE adapter from Mellanox (now NVIDIA), on PCIe 3.0 x8 or x16 depending on SKU, using the mlx5 driver that ships in the mainline Linux kernel. It is end-of-life at its maker, which means no new firmware - not that it has stopped working. 1 listing in stock, from $330.15.
Prices updated · live figures update every 4-5 hours
Thin coverage today - fewer than three listings, so this page is not indexed until stock returns.
| Card | Ports | Interface | Condition | Price | |
|---|---|---|---|---|---|
| ConnectX-4 Mellanox | 2 | QSFP+ | New | $330.15 | Buy |
Passive direct-attach copper, by length. A DAC is a fixed-length cable with the transceivers moulded on - you buy the whole run at once rather than a cable plus two modules.
Every "compatible with" phrase below is the seller's own wording, recorded as a claim and badged. We have not tested any module against any switch.
ConnectX-4 (not the Lx variant) is the dual-port QSFP28-cage card in the Mellanox 25/40/100GbE family, and the one listing we track states 40/56GbE - 40GbE being the standard, interoperable Ethernet rate, and 56GbE a Mellanox-proprietary rate that only negotiates between two compatible Mellanox endpoints (or through an SX10xx-class switch), not a rate you can assume works against arbitrary 40GbE gear.
The QSFP28 cage on this card is backward compatible with QSFP+ cabling and optics, the same relationship SFP28 has with SFP+ on the 25G side of this family - so a 40GbE build using this card takes ordinary QSFP+ direct-attach cable rather than requiring 100G-rated QSFP28 cabling. The cabling guide's QSFP+ section is the one to use for this card.
The lifecycle picture here is more mixed than its siblings. NVIDIA's own manual confirms the ordering part numbers are end-of-life without stating the plain-language banner sentence the ConnectX-4 Lx and ConnectX-5 manuals both carry, and NVIDIA's current DOCA-Host software package list leaves plain ConnectX-4 off entirely - Lx and ConnectX-5 are both still included. None of that changes whether the in-kernel mlx5 driver recognises the card today; it does. It is a reasonable factor in how far into the future you plan to rely on vendor-side software support beyond the kernel driver itself.
Given the thin listing coverage we currently track, this page exists to document what NVIDIA states about the card rather than to claim a deep market - check the live table above before assuming availability.
NVIDIA's Linux Kernel Upstream Release Notes list mlx5 for ConnectX-4 by name alongside ConnectX-4 Lx and ConnectX-5, in the mainline kernel. That matters more than any marketing claim, because a card whose driver is in the mainline kernel is a card your Linux-based NAS or hypervisor recognises without you doing anything.
We state in-kernel driver names as documented facts because they are - they come from the kernel's own driver documentation, not from a forum thread reporting that something worked. Where community experience is the only evidence available for something, we say so rather than dressing it up.
NVIDIA's ConnectX-4 EN manual carries an "EOL'd Ordering Part Numbers" table listing the MCX413A/414A/415A/416A family - confirming the line is closed for new orders - but, unlike the ConnectX-4 Lx and ConnectX-5 manuals, does not carry the same plain-language "has moved to end-of-life" banner sentence. Stated as that asymmetry rather than smoothed into one claim, because it is what NVIDIA's own documentation actually shows. A further difference worth knowing: NVIDIA's current DOCA-Host supported-device list (dated September 2026) includes ConnectX-5 and ConnectX-4 Lx, but not the standard ConnectX-4 - so a ConnectX-4 buyer today is relying on the in-kernel mlx5 driver rather than NVIDIA's newer software packages. Those are two different questions and conflating them is the most common error in discussions about this card.
Windows support is a separate question from Linux support and varies by card age. Check the maker's current driver page for your specific part number before buying if Windows is the target, because an in-kernel Linux driver tells you nothing about it.
Our tracked listing for this card is Mellanox-branded. Coverage is thin at the time of writing - one verified listing - so treat the live table above as the actual state of stock rather than this page implying a deep market.
The general shape of the crossflash question is the same across every card in this category: an OEM board carries the server maker's firmware, that firmware sometimes behaves differently from the silicon vendor's, and there is a community practice of replacing it. We link the maintained community documentation rather than reproducing the procedure here. A stale copy of a flashing guide is actively dangerous, and the people who maintain the canonical versions do so because the details change.
If you are buying an OEM-branded card specifically because it is cheaper, factor in the possibility that you will need to do this. If that prospect does not appeal, the vendor-branded card at a higher price is buying you the absence of that job.
This is a QSFP+ card, so the port is a cage rather than a socket and you choose what goes in it. For two machines near each other, a passive direct-attach copper cable is the cheapest, coolest and simplest answer, and it is what most builds using this card should use.
QSFP+ ports also support breakout cabling - one cable splitting a single high-speed port into several lower-speed ones (a 100G QSFP28 port to four 25G legs, or a 40G QSFP+ port to four 10G legs), which is how a single card and a single switch port often end up feeding several hosts more cheaply than one-to-one cabling would. See the cabling guide for the live breakout listings and the exact split each standard defines.
The compatibility question belongs here rather than being buried. Third-party module sellers list the equipment they claim compatibility with, and those are claims we record and badge rather than verify. Vendor coding is a documented phenomenon: cards and switches can and do refuse modules that do not identify as approved parts. A direct-attach cable avoids the entire question, which is another reason it is the default recommendation for a two-machine build.
The cabling guide covers every QSFP+ DAC, AOC and optic in one place · The 25/40/100GbE hub compares every card in this cluster and the switched-versus-point-to-point arithmetic · 10GbE vs 25GbE is the step-up decision, priced both ways · used servers almost always have a spare PCIe slot for one · NAS devices is usually what sits at the other end · rack disk shelves are the other common thing on the far end of a fast link.
No - different cards in the same family. ConnectX-4 Lx is the SFP28 card covered on its own page in the 10GbE cluster (it runs 10G/25G); plain ConnectX-4 is the QSFP28-cage card on this page, running 40G standard Ethernet (with a 56G Mellanox-proprietary option between compatible endpoints). Check the exact part number and cage type before buying, since the names are easy to conflate.
No. Its QSFP28 cage is backward compatible with QSFP+ cabling, the same relationship SFP28 has with SFP+ on the 25G side - so a 40GbE build with this card uses ordinary QSFP+ direct-attach cable, not the pricier 100G-rated QSFP28 DACs.
Its ordering part numbers are listed as end-of-life in NVIDIA's own manual, but that manual does not carry the same explicit "has moved to end-of-life" sentence the Lx and ConnectX-5 manuals do - and NVIDIA's current DOCA-Host software package list does not include plain ConnectX-4 the way it includes those two. Practically similar outcome, documented slightly differently by NVIDIA itself, and stated here exactly that way rather than flattened into one claim.
A Mellanox-proprietary link rate above standard 40GbE Ethernet, available only between two compatible Mellanox/NVIDIA endpoints (or via certain Mellanox switches) - not a rate you should plan a build around unless every device on the link is confirmed to support it. 40GbE is the standard, interoperable rate this card is built for.
mlx5, the same driver family as ConnectX-4 Lx and ConnectX-5, confirmed by NVIDIA's own current Linux Kernel Upstream Release Notes.
One at the time of writing. This is a thin category compared to ConnectX-5 or Intel E810 in our catalogue - check the live table above rather than assuming broad availability.
Specification claims on this page were checked against the pages below on 4 August 2026. Where a widely-repeated figure did not survive that check, we report what the source says now and say so in the text rather than repeating the familiar number.