Mellanox ConnectX-5 - Live Prices and Driver Support

Quick answer

The ConnectX-5 is a SFP28 25/100GbE adapter from Mellanox (now NVIDIA), on PCIe 3.0 x8 or x16 depending on SKU; the ConnectX-5 Ex variant adds PCIe 4.0 on its higher-speed SKUs, 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. 6 listings in stock, from $244.00.

Prices updated · live figures update every 4-5 hours

InterfaceSFP28
Portssingle and dual port variants (25G SFP28 SKUs; separate 100G QSFP28 SKUs also exist and are in the table below)
BusPCIe 3.0 x8 or x16 depending on SKU; the ConnectX-5 Ex variant adds PCIe 4.0 on its higher-speed SKUs
Linux driver (in-kernel)mlx5
MakerMellanox (now NVIDIA)
LifecycleEnd-of-life at maker

ConnectX-5 listings in stock

CardPortsInterfaceConditionPrice
ConnectX-5
Mellanox
2SFP28New$244.00Buy
ConnectX-5
Mellanox
2SFP28New$450.00Buy
ConnectX-5
Mellanox
1QSFP28New$485.00Buy
ConnectX-5
Mellanox
2QSFP28Refurb$489.99Buy
ConnectX-5
Mellanox
1QSFP28New$679.00Buy
ConnectX-5
Mellanox
2QSFP28New$750.00Buy

Direct-attach cables that pair with it

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.

LengthBrandTypePrice
0.3m10GtekSFP28 passive$13.99Buy
0.5mH!FiberSFP28 passive$11.99Buy
0.5mH!FiberSFP28 passive$12.99Buy
0.5m10GtekSFP28 passive$13.99Buy
0.5m10GtekSFP28 passive$13.99Buy
1m10GtekSFP28 passive$14.39Buy
1m10GtekSFP28 passive$15.99Buy
1m10GtekSFP28 passive$15.99Buy

Modules and optics, with seller claims

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.

ModuleTypeBrandSeller compatibility claimsPrice
SFP28 25GBASE-SR module25GBASE-SR10Gteknone stated$19.99Buy
SFP28 25GBASE-SR module25GBASE-SR10Gteknone stated$19.99Buy
SFP28 25GBASE-SR module25GBASE-SR10Gteknone stated$19.99Buy
SFP28 AOCAOC10Gteknone stated$19.99Buy
SFP28 25GBASE-LR module25GBASE-LR10Gteknone stated$22.99Buy
SFP28 AOCAOCipolex25GBase Active Optical Cable, 25Gbps SFP28, SFP28-25G-AOC5Mseller claim$31.99Buy
SFP28 AOCAOC10Gteknone stated$35.99Buy
SFP28 AOCAOC10Gteknone stated$37.99Buy

What the ConnectX-5 is

ConnectX-5 is the card most homelab buyers land on when they decide to step past 10GbE, and the reason is the SKU spread: the same family covers 25G dual-port SFP28 cards and 100G dual-port QSFP28 cards, both on the mlx5 driver, both from listings that are almost entirely Mellanox-branded rather than third-party boards. That first-party concentration is unusual in this market and worth knowing before you shop - see the OEM note below for why.

The honest headline: NVIDIA's own documentation now calls ConnectX-5 end-of-life, the same status ConnectX-4 Lx carries in the 10GbE cluster below. That is a commercial statement about new orders, not a statement about whether the card works or is supported in the kernel you are about to install - NVIDIA's own, separately dated release notes confirm mlx5 keeps supporting it. Read both facts together rather than either alone.

The 25G and 100G SKUs are physically different cards - SFP28 cage versus QSFP28 cage - and the cabling and DAC tables on this page default to the 25G SFP28 side, since that is the more direct step up from a 10GbE SFP+ build and the connector most of our tracked listings use. A 100G QSFP28 ConnectX-5 needs QSFP28 cabling instead - see the cabling guide's QSFP28 section for those live listings, and the /networking/100gbe hub for a build priced around the 100G SKU specifically.

PCIe requirement varies by SKU rather than being one number for the whole family: base ConnectX-5 EN cards run PCIe 3.0 at x8 or x16 depending on port count and speed, and the ConnectX-5 Ex variant adds PCIe 4.0 on its 100G SKUs. Check the exact part number's PCIe line in NVIDIA's own manual before assuming a slot is wide enough - a 100G card in a physically x16 but electrically x4 slot is the category's classic expensive mistake.

Driver and operating system support

NVIDIA's own Linux Kernel Upstream Release Notes (dated 2026) list mlx5 as the driver for ConnectX-5 by name, in the mainline kernel and actively maintained - the same driver family as ConnectX-4 and ConnectX-4 Lx. 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-5 user manual states plainly: "The ConnectX-5 product line has moved to end-of-life" (confirmed 2026-09-13). NVIDIA's own kernel release notes, dated the same year, still list mlx5 as actively supporting it, and NVIDIA's current DOCA-Host device list (dated September 2026) still includes ConnectX-5 - so the commercial line is closed for new orders while both the in-kernel driver and NVIDIA's current software packages keep supporting cards already in the field. 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.

OEM-branded versions and firmware

Every ConnectX-5 listing we track right now is Mellanox-branded (tier 1) rather than a third-party board asserting the chipset - unlike the Intel E810 tier below, this specific card has not attracted the same third-party-board ecosystem in what we currently hold in stock. That can change as more of these get pulled from decommissioned servers.

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.

What to plug into it

This is a SFP28 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.

For longer runs you need optics - a module at each end plus a fibre between them - or an active optical cable, which packages the same thing as a single sealed object. Both cost more than direct-attach copper and both go much further.

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.

Where to go next

The cabling guide covers every SFP28 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.

Frequently asked questions

Is ConnectX-5 a good card to buy used in 2026, given it's end-of-life?

For most home and homelab builds, yes - the EOL status is commercial (no new firmware or hardware revisions from NVIDIA), while the mlx5 driver keeps this card supported in the mainline Linux kernel per NVIDIA's own current release notes. Treat it the same way the site treats the ConnectX-3 and ConnectX-4 Lx: sound for a machine you expect to rebuild within a few years, a longer-horizon gamble for one you want to leave untouched for a decade.

What's the difference between the 25G and 100G ConnectX-5 cards?

Physically different connectors - SFP28 for the 25G SKUs, QSFP28 for the 100G ones - on the same mlx5 driver and the same EOL status. They need different cables. This page's cable tables default to SFP28; a 100G QSFP28 build needs the cabling guide's QSFP28 section instead.

Which driver does ConnectX-5 use, and is it still maintained?

mlx5, confirmed by name in NVIDIA's own Linux Kernel Upstream Release Notes, dated 2026 - current documentation, not an old snapshot. The same page confirms mlx5 covers ConnectX-4 and ConnectX-4 Lx too, so troubleshooting knowledge mostly transfers across the family.

Does ConnectX-5 need a specific PCIe slot?

It depends on the exact SKU - base EN cards run PCIe 3.0 x8 or x16, and the ConnectX-5 Ex variant adds PCIe 4.0 on its 100G SKUs, per NVIDIA's own manual. Check your specific part number rather than assuming, and check the slot's electrical width (not just its physical size) before buying a 100G card for it.

Are third-party ConnectX-5 boards common?

Not in what we currently track - every listing we hold right now is Mellanox-branded. That is unusual for this market and may simply reflect current stock; check the live table above rather than assuming it stays that way.

Is ConnectX-5 still supported for NVIDIA's current driver packages, or only the in-kernel one?

Both, as of NVIDIA's current documentation: it's in NVIDIA's DOCA-Host supported-device list (dated September 2026) as well as the in-kernel mlx5 driver. That puts it ahead of the standard (non-Lx) ConnectX-4, which the same DOCA-Host list does not include.

Sources

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.

  1. NVIDIA (Mellanox) - ConnectX-3 EN product line and lifecycle status (nvidia.com, 2026)
    Used for: that ConnectX-3 has reached end-of-life status with NVIDIA, and the MCX311A / MCX312A / MCX314A part-number family. EOL means no new firmware or driver work from the maker; it does not mean the card stops working, and the page says so in those terms.
  2. The Linux kernel project - mlx4, mlx5, ixgbe and i40e in-tree Ethernet drivers (kernel.org, 2026)
    Used for: that mlx4 (ConnectX-3), mlx5 (ConnectX-4 and later), ixgbe (Intel 82599/X520, X540, X550) and i40e (X710) ship in the mainline Linux kernel. This is why an end-of-life card can still be a sane purchase, and it is a documented fact rather than a forum claim.
  3. DatacenterDisk - Live catalogue - NIC, DAC and transceiver listings (datacenterdisk.com, 2026-08-11)
    Used for: every price on this page, read from our catalogue at request time. Seeded 11 August 2026: 24 network cards across 6 models, 24 DAC cables, 8 transceivers. Every card is split tier 1 (the listing is that part) or tier 2 (a third-party board asserting that silicon).
  4. DatacenterDisk - Sourcing note - power figures and compatibility claims (datacenterdisk.com, 2026-08-11)
    Used for: We publish no per-port wattage for either cabling path. The relationship - 10GBASE-T draws more power and runs hotter than SFP+ direct attach - is documented at product level and is stated here. The magnitude is not, because we have not fetched and logged a source we can cite for it. Separately: every 'compatible with' phrase shown on this site is stored as a seller claim at tier 2 and badged wherever it renders. We have not tested any module against any switch.
  5. NVIDIA - ConnectX-5, ConnectX-4 Lx and ConnectX-4 EN user manuals - end-of-life status (networking-docs.nvidia.com, 2026-09-13)
    Used for: that NVIDIA's own manuals state "The ConnectX-5 product line has moved to end-of-life" and "The ConnectX-4 Lx product line has moved to end-of-life" in those exact words (connectx5enhw and connectx4lxhw respectively, both dated 2026). The standard ConnectX-4 EN manual (connectx4enhw) carries an "EOL'd Ordering Part Numbers" table for every MCX413A/414A/415A/416A SKU but not the same plain-English banner sentence - stated as that asymmetry rather than smoothed into one claim. Also the per-SKU PCIe generation and lane count (PCIe 3.0 x8 or x16 depending on SKU; PCIe 3.0/4.0 on ConnectX-5 Ex variants) from the same manuals.
  6. NVIDIA - Linux Kernel Upstream Release Notes - driver and firmware support table (networking-docs.nvidia.com, 2026-09-13)
    Used for: that ConnectX-4, ConnectX-4 Lx and ConnectX-5 all use the mlx5 driver family per NVIDIA's own "Supported Uplinks to Servers" table, dated June 2026 - current documentation, confirming the in-kernel driver is still maintained for hardware NVIDIA itself calls end-of-life for ordering. Also that NVIDIA's current DOCA-Host device list (docs dated September 2026) includes ConnectX-5 and ConnectX-4 Lx but not the standard ConnectX-4 - a real difference in ongoing software-package support between the two, stated as found rather than smoothed over.
  7. The Linux kernel project - i40e and ice in-tree Ethernet driver documentation (kernel.org, 2026-09-13)
    Used for: that the i40e driver explicitly lists Intel Ethernet Controller XXV710 as supported hardware alongside X710/XL710/X722 (same page also notes XXV710's speed cannot be forced via ethtool - auto-negotiation only). The companion ice driver page (device_drivers/ethernet/intel/ice.html) is titled for the "800 Series" - the family E810 belongs to - without naming E810 verbatim in the rendered portion, which is stated as the limit of what this specific page confirms rather than stretched into a direct quote. Intel's own ARK product pages did not render in this sourcing pass (see the 100GbE and NIC pages' own note); PCIe generation and EOL status for XXV710 and E810 parts are marked not sourced for that reason.
  8. Arista Networks - 7050X and 7050X3 Series switch specifications, and End-of-Sale advisories (arista.com, 2026-09-13)
    Used for: the 7050X family's own published figures: "10GbE Typical Power: 1.2W per port", and for the 7050QX-32S specifically "Typical Power Draw: 150W (4.5W per port)" - the only per-port power figures this cluster publishes, because they are the only ones a vendor datasheet actually rendered and stated. Also Arista's own End-of-Sale advisory archive (arista.com/en/support/advisories-notices/endofsale), which carries a dated 2016 end-of-sale notice for "Select Models of the Arista 7050SX, 7050TX and 7050QX Series" and, separately, 2026-dated end-of-sale notices for the newer 7050SX3-48YC12 and 7050SX3-48YC8/48C8 models - a used 7050-family switch may be running silicon its own maker has since end-of-saled, stated with the notice's own date rather than ours.
  9. DatacenterDisk - Live catalogue and sourcing note - 25/40/100GbE cluster (datacenterdisk.com, 2026-09-13)
    Used for: every price in this cluster's tables, read from our catalogue at request time. Seeded 13 September 2026: 12 NICs across 3 models (ConnectX-5, ConnectX-4, Intel E810), 40 DAC cables (25G/40G/100G including breakout), 18 transceivers, 11 used-enterprise and current-SMB switches. Two vendor documentation gaps stated plainly rather than filled: NVIDIA/Mellanox's per-model switch datasheets (SN2010/SN2100/SN2410/SN2700) did not render in this sourcing pass beyond a series-level marketing page, so those four switches carry no sourced power, acoustic or EOL figures; Intel's ark.intel.com did not render at all in this pass, so XXV710 and E810 PCIe generation and formal EOL status are marked not sourced rather than assumed from the 700/800-series family pattern.
Share:WhatsAppFacebookPost on X