The 10GbE Upgrade - Live Card, Cable and Module Prices

10GbE network cards in stock now

Tier badges matter here: a badged row is a third-party board asserting the named chipset, not the named part. Both are legitimate purchases; only one of them is the maker's own card.

Direct-attach copper cables (DAC)

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.

SFP+ transceivers and active optical cables

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.

The six cards worth knowing

Who actually needs 10GbE, and who does not

Start with the arithmetic, because it settles most of the argument before any hardware is bought. Network link speeds are quoted in bits per second and file transfers happen in bytes, so divide by eight. Gigabit Ethernet tops out around 125 megabytes per second. 2.5GbE tops out around 312. 10GbE tops out around 1,250. Those are ceilings defined by the standard, not results anyone measured, and real transfers land below them because of protocol overhead.

The useful question follows immediately: can the storage on either end of the link produce or absorb data faster than the cheaper ceiling? If it cannot, the faster port changes nothing, and you have bought a card, a cable and possibly a switch to watch the same transfer take the same time.

This is where a lot of home 10GbE money goes to waste. A NAS with one or two mechanical drives is very unlikely to feed a 10GbE link, and often will not saturate 2.5GbE either. We are not going to tell you what your drives sustain, because we have not measured them and neither has anyone else who is quoting you a number - but every drive maker publishes a sustained transfer rate for each model, and that figure, multiplied by however many drives can actually work in parallel in your array, is the number to compare against 125, 312 and 1,250. Do that arithmetic before buying anything on this page.

10GbE earns its cost in a smaller set of cases than the enthusiasm suggests. Multi-drive arrays where several spindles read in parallel. NVMe pools, where the storage is no longer the constraint by a wide margin. Video editing straight off shared storage, where the working file is enormous and the wait is the job. Backup windows that have to close overnight. Virtualisation hosts moving disk images between nodes. And the case that motivates most of this page: two machines that need a fast path between them and nothing else.

If none of those describe you, 2.5GbE is inexpensive, runs over the cabling you already own, and is very often the honest answer. We would rather say that on the hub page than sell you a card.

SFP+ or 10GBASE-T: the decision that shapes everything after it

10GbE reaches your machine one of two ways, and the choice constrains every cable and switch decision that follows.

10GBASE-T is 10GbE over familiar RJ45 twisted pair. It plugs into the same shape of socket as every other Ethernet device you own, runs over Cat6 or Cat6a, and needs no thought about modules. That convenience is the entire case for it, and it is a strong one if you are running through walls you are not going to reopen.

SFP+ is a cage rather than a socket. You put a module in it - or a cable with the module ends already attached - and the cage does not care which, so the same port can take copper for a short run or fibre for a long one. Enterprise gear standardised on it, which is why the second-hand market is full of SFP+ cards and why the community default for a home 10GbE build is SFP+ rather than RJ45.

The power and heat difference is real and it runs one way: 10GBASE-T draws more power per port and runs hotter than SFP+ direct-attach copper. That is a documented product-level characteristic, and it is the reason an RJ45 module in an SFP+ cage is the single most complained-about component in this category - the cage was designed around the thermal budget of a passive copper cable, and a 10GBASE-T module puts considerably more into it.

We are not going to give you a wattage. Per-port figures for both paths are widely quoted, and we have not fetched and logged a source we can point you at for them, so publishing one would be a number you could not check. When we have sourced the datasheets the figures will appear here with a citation attached. Until then: the direction is documented, the magnitude is not ours to assert.

The practical version. If both machines are in the same room or the same rack, SFP+ with a direct-attach cable is cheaper, cooler and simpler. If the run goes through a wall and terminates in a wall plate, 10GBASE-T is worth the heat.

The four ways to actually connect two SFP+ ports

Once you have SFP+ cages at both ends there are four cabling paths, and they differ in cost, distance and heat rather than in the speed they carry.

A passive direct-attach copper cable - a DAC - is a fixed-length twinaxial cable with the transceiver ends moulded on permanently. You buy the whole run as one object. It is the cheapest path by a distance, it draws essentially nothing because there is no active circuitry, and it is what most short-run builds should use. The trade is that the length is fixed at purchase and passive DAC does not run far; the practical ceiling is around seven metres, which is where our catalogue's longest passive cables sit.

An active optical cable - an AOC - is the same idea with optics inside and a fibre between them. It costs more than a DAC and goes much further, and like a DAC it is a sealed fixed-length object.

Fibre plus separate optics is two transceiver modules and a fibre patch lead between them. It costs the most per link and it is the only path that scales to real distance - multimode short-range optics for within a building, single-mode long-range for between them. It is also the most flexible, because the fibre and the optics are separate purchases and either can be changed.

An RJ45 module in an SFP+ cage converts the port to 10GBASE-T so it can meet copper structured cabling. It is the most expensive per port of the four, it is the hottest, and it exists to solve a specific problem: an SFP+ machine that has to reach an RJ45 wall socket. If you have that problem it is the answer. If you do not, one of the other three is cheaper and cooler.

Live prices for all four paths are in the tables above. The gap between the cheapest DAC and a pair of optics is large enough that it should decide the build for anyone whose machines sit near each other.

Why used enterprise cards dominate this category

The reason a 10GbE card can cost less than a mid-range cable is that datacentres bought these cards by the pallet and then replaced them. Enterprise refresh cycles put enormous volumes of perfectly functional SFP+ cards into the second-hand market, and the supply overwhelms the hobbyist demand that meets it.

One honest caveat about our own coverage. We track Amazon, and the deepest pool of used enterprise cards is not on Amazon - it is on the auction and surplus markets, where the ex-datacentre pulls actually land. Everything in our tables is priced as new stock. That does not change the underlying economics much, because the cheapest new listings for the older cards are already priced against that used supply, but if you have read that these cards go for very little and our lowest figure is higher than you expected, the gap is a marketplace difference rather than a disagreement about the market.

The end-of-life question comes up constantly and deserves a direct answer. Mellanox ConnectX-3 is end-of-life at NVIDIA. That means the maker has stopped producing new firmware and driver work for it. It does not mean the card stops working, and it does not mean your operating system has dropped it: the mlx4 driver that runs ConnectX-3 ships in the mainline Linux kernel, as ixgbe does for the Intel 82599, X540 and X550 families and i40e does for the X710. Those are documented facts about what is in the kernel tree, not community reassurance.

What end-of-life does mean is that the risk profile changes over time rather than today. A card whose driver is in-tree will keep working for as long as that driver stays in-tree, and drivers are removed from the kernel slowly and with notice. Buying an EOL card for a system you will rebuild in three years is a different decision from buying one for something you intend never to touch again.

Separately: OEM-branded versions of these cards exist in quantity - the same silicon on a board carrying a Dell, HPE or Lenovo part number, sometimes with firmware that behaves differently with third-party modules. There is a well-established community practice of reflashing those cards to the generic maker firmware. We are not going to write that procedure out as our own instructions, for the same reason we do not rewrite RAID controller flashing steps: getting it wrong bricks the card, and the people who maintain the canonical guides maintain them for a reason. If you need it, find the maintained community documentation for your specific card and follow that.

Reading our tables

Every price here is the cheapest live listing we hold at the moment you loaded the page, and this category is priced in whole dollars rather than per terabyte for the obvious reason - a network card has no capacity, so a cost-per-terabyte column would be meaningless and we do not render one.

The card table carries a tier badge, and it is the most important thing on this page to understand. Some listings are the named part: Intel selling an Intel X520. Others are third-party boards built around the same silicon, and their titles say so - phrases like "compare to Intel X710-DA2" or "compatible for Intel X550-T2". Those boards are often genuinely built on the chipset they name, they are frequently cheaper than the first-party card, and plenty of people run them happily. But the chipset attribution is the seller's assertion about their own product, so we badge it rather than restating it as ours.

The same rule governs the transceiver table. "Compatible with Cisco SFP-10G-SR" on a third-party module is a claim the seller is making, recorded as a claim and shown as one. We have not tested any module against any switch. It is documented that switches can and do reject modules they do not recognise - vendor coding is a real phenomenon, not a myth - but which specific switch accepts which specific module is not something we can tell you, and anyone who tells you confidently without having tried it is guessing.

Multi-pack listings are excluded. This category is full of them and they are exactly the trap they look like: a ten-pack of modules priced as a unit would sit at the top of a price ranking while costing ten times what the row implies. Our first discovery pass kept several before the filter was tightened.

Where to go next

The two-machine build is the cheapest 10GbE that exists - two cards, one cable, no switch · 10GbE deals ranks every part against its category · DAC cables and transceivers cover the cabling in detail · 10GbE NAS devices lists which units ship the port · SFP+ vs 10GBASE-T, ConnectX-3 vs X520 and DAC vs fibre are the three decisions people get stuck on · NAS devices is where the other end of most of these links lives · the RAID planner helps work out whether your array can actually feed a 10GbE port · used servers mostly ship with spare PCIe slots for exactly this upgrade.

Frequently asked questions

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.