Realtek RTL8156 USB 2.5GbE Adapters: The Cheap Upgrade That Lies
The chip that put 2.5-gigabit Ethernet on every mini PC's desk, and why the number on the box rarely shows up in a real transfer
Contents
The RTL8156 is the reason 2.5-gigabit Ethernet became a £15 impulse buy instead of a proper networking upgrade. It’s the chip inside most of the cheap USB-A and USB-C dongles from Ugreen, Anker, ASUS and a dozen white-label brands, and it’s genuinely useful — a mini PC or laptop with only gigabit onboard can plug in one of these and, on paper, triple its ceiling. The problem is that “on paper” carries most of the weight in that sentence. Published tests consistently find real throughput sitting well short of the advertised 2.5 Gbps, and the gap is large enough, and consistent enough across reviewers and platforms, to be worth understanding before you buy one expecting a straightforward multiplier on your existing gigabit link.
What the chip actually promises
RTL8156 is a USB-to-Ethernet bridge chip: it takes a USB 3.x connection on one side and presents a 2.5GbE MAC/PHY on the other, letting a machine without built-in 2.5GbE (which is most machines still in service) get there via a cheap external dongle instead of a PCIe card or a motherboard swap. There are two relevant silicon revisions in the wild: the original RTL8156 and the later RTL8156B, and the difference between them matters more than most buyers realise. The original chip had well-documented problems with frequent link drops and poor power management under sustained load. The B revision specifically fixed both, and separately cut power draw dramatically compared to the original — reviews note reductions on the order of 60%+ in some measurements, alongside less heat and fewer of the random disconnects that plagued the first version. If you’re buying today, checking that a listing specifies RTL8156B rather than the bare RTL8156 is worth the extra thirty seconds, because the two chips behave differently enough to matter in daily use.
Why 2.5 Gbps on the box rarely shows up in a transfer
Here’s where the “lie” in the title earns its keep. Multiple independent reviews across Windows, macOS and Linux consistently measure real-world throughput well under the rated 2.5 Gbps — commonly somewhere in the 750 Mbps to 1.2 Gbps range for actual file transfers, occasionally reaching closer to 2 Gbps on one direction under ideal conditions but rarely both directions at once. The causes are a mix of driver quality, USB controller overhead, and thermal throttling in the compact dongle enclosures these chips usually ship in, and the split matters, because a driver problem is fixable with a config change while a genuine bandwidth ceiling stays fixed no matter what you install. On Linux specifically, the choice of driver makes an outsized difference — the generic cdc_ncm driver that many distributions load by default manages well under 500 Mbps in some tests, while the dedicated r8152 driver (which handles the RTL8152/8153/8156 family properly) recovers most of the advertised bandwidth, commonly reaching the 900+ Mbps range in each direction. That’s still short of 2.5 Gbps, but it’s the difference between “barely better than gigabit” and “a genuine, worthwhile upgrade.”
Getting the right driver loaded
On most current Linux distributions the r8152 driver ships in-tree and loads automatically, but older kernels or minimal installs sometimes fall back to the slower generic USB networking driver instead. Checking which one is actually bound to the device is the first troubleshooting step for anyone who plugs in one of these dongles and doesn’t see the speed bump they expected:
| |
If driver: reports cdc_ncm or cdc_ether instead of r8152, that’s the gap — either the in-tree module needs updating (some distributions ship an older r8152 that predates full RTL8156 support, in which case Realtek’s own out-of-tree driver package is worth installing manually), or a udev/modprobe rule is explicitly blacklisting it in favour of the generic driver. On Windows and macOS the equivalent fix is simpler in practice: install the vendor driver from the dongle manufacturer’s site rather than relying on whatever the OS bundled, since the generic in-box drivers on both platforms lag behind Realtek’s own releases.
USB-A vs USB-C, and the power budget nobody checks
Most RTL8156-based dongles come in both USB-A and USB-C flavours, and the choice isn’t purely cosmetic. A USB-C connection to a port that actually supports USB 3.x data (rather than a USB-C port wired for charging or DisplayPort alt-mode only) tends to deliver marginally more consistent throughput in reviewer testing, partly because USB-C ports on modern hardware are more likely to be wired to a faster, less contended internal controller than legacy USB-A ports on the same machine. More importantly, these dongles are entirely bus-powered — there’s no separate power input — and draw enough current under sustained load that a heavily loaded USB hub, or a port already feeding a keyboard and a couple of other peripherals, can starve the adapter of the power it needs for stable operation. Plugging a 2.5GbE dongle directly into a host port rather than through a hub, especially an unpowered one, avoids a category of intermittent-disconnect complaints that have nothing to do with the chip’s own driver or firmware.
Choosing between the flood of near-identical listings
Because the RTL8156B chip itself is a commodity part that Realtek sells to any manufacturer who wants it, the market is flooded with dongles that are functionally identical inside a slightly different plastic shell — Ugreen, Anker, and a long tail of unbranded listings all use the same silicon, and the actual difference between them comes down to enclosure quality, cable/connector build, and how conservatively the manufacturer specs the thermal design. Reviews from established testing sites (ServeTheHome and similar outlets have run direct comparisons) generally find little meaningful throughput difference between reputable brands once the correct driver is loaded — the variance shows up more in build quality, connector durability after repeated plugging, and whether the manufacturer bothers to publish the chip revision at all. Treat “does the listing actually say RTL8156B” as the real filter, not brand name recognition, since several white-label products from the same factory get resold under a dozen different storefronts.
USB controller and cable matter more than people expect
Because these are USB devices riding on a USB 3.x link, the actual host controller and cable in the chain cap performance just as much as the chip itself. A dongle plugged into a USB 3.0 port shared on an internal hub with three other high-bandwidth devices — a common situation on compact mini PCs where several ports trace back to the same internal controller — will contend for bandwidth in ways that a dedicated port won’t. A cable longer than about a metre, or a passive USB-C-to-A adapter added into the chain, introduces enough signal degradation on cheaper cables to measurably affect sustained throughput. None of this is unique to RTL8156 devices, but it explains a lot of the “I got completely different speeds on two different machines with the identical dongle” reports that show up in forum threads — the dongle was never the only variable.
Where 2.5GbE dongles are a genuinely good fit
The honest use case for these adapters isn’t matching a proper 2.5GbE or 10GbE backbone — it’s bridging the gap for a machine that would otherwise be stuck at gigabit. A mini PC being used for an eGPU build via OCuLink or as a compact always-on server, where an internal NIC upgrade isn’t practical, is the target scenario: even at a realistic 900 Mbps-1 Gbps sustained rather than the rated 2.5 Gbps, that’s still a real improvement over an internal 100 Mbps port, and a meaningful one over gigabit if your switch and the rest of your network can actually push past it. Compare that against something like an Odroid H4 vs Raspberry Pi 5 build, where the H4’s onboard 2.5GbE port entirely sidesteps this whole conversation — if the built-in NIC already does 2.5GbE properly, a USB adapter is solving a problem you don’t have.
macOS specifics worth knowing
macOS testing of these dongles has produced some of the most lopsided results in the reviews I’ve seen: one comparison of several 2.5GbE dongles on Monterey found upload speeds reaching over 2 Gbps while downloads on the same hardware, same cable, same test ranged from under 1 Gbps to a bit over 1.2 Gbps — a genuinely asymmetric result that comes down entirely to how macOS’s networking stack and Realtek’s driver for that platform handle the two directions differently, with the network path itself ruled out. Apple doesn’t bundle a Realtek driver in the box; getting a RTL8156-based dongle working at all on macOS means installing the vendor’s kernel extension or (on Apple Silicon under recent macOS versions) a system extension, and skipping that step is the single most common reason a Mac user reports the dongle “not working” rather than “working slower than expected.” Check the specific dongle’s product page for a macOS driver download before assuming plug-and-play will get you anywhere close to the advertised speed.
Sizing expectations against the rest of your network
None of this matters if the rest of the path can’t use the extra bandwidth anyway. A 2.5GbE dongle plugged into a switch with only gigabit ports, talking to a NAS with a gigabit NIC, delivers exactly nothing beyond what gigabit already gave you — the adapter’s ceiling is irrelevant if every other hop in the chain is the actual bottleneck. Before buying one of these expecting a network-wide speed boost, confirm the switch port, the cabling (Cat5e is technically enough for 2.5GbE over short runs, but Cat6 removes any doubt), and the device on the other end all support the higher speed. The adapter is one link in a chain, and a chain runs at its slowest link regardless of how fast any single dongle claims to be.
Troubleshooting notes
The single most common complaint — “I’m only getting gigabit speeds through a 2.5GbE adapter” — is almost always the driver issue above; check ethtool -i before assuming a hardware fault. Random disconnects under sustained load, particularly on the original (non-B) RTL8156 chip, point to a known power-management issue with that specific silicon revision; disabling USB autosuspend for that device (echo -1 > /sys/module/usbcore/parameters/autosuspend_delay_ms as a blunt system-wide fix, or a targeted udev rule for just that device) resolves most cases. Overheating dongles that throttle after extended transfers — a genuine issue in the smaller, cheaper enclosures with no thermal mass to speak of — benefit from simply not being tucked directly against a warm mini PC chassis; an inch of clearance and airflow makes a measurable difference in sustained throughput tests. And speeds that vary wildly between two apparently identical dongles from the same listing usually trace back to counterfeit or bin-sorted chips in cheaper third-party listings rather than genuine variance in the RTL8156B itself — buying from a brand that publishes its actual chip revision, rather than an unbranded generic listing, avoids most of this. And on macOS specifically, a dongle that shows a solid Ethernet link light but never appears as a network interface almost always means the vendor’s driver extension was never approved in System Settings after installation — a step that’s easy to click past without noticing, since the approval prompt doesn’t always appear where people expect it.
The honest verdict
For roughly £15-20, an RTL8156B-based dongle is a legitimate way to get a gigabit-only machine most of the way to 2.5GbE performance, provided you check the driver is actually loaded correctly and the rest of your network can use the extra bandwidth. It won’t reliably hit the number printed on the box, and treating that number as a real-world guarantee is where most of the disappointment in reviews comes from. Treat it as “meaningfully faster than gigabit, somewhat short of true 2.5GbE,” and it’s one of the cheapest useful upgrades in a homelab toolbox.
The gap between the number on the box and the number in a real transfer shows up across most USB-attached hardware, where the interface spec describes a theoretical ceiling and the actual silicon, driver, and cable determine how much of that ceiling you ever see. RTL8156B dongles are simply the most visible current example, because 2.5GbE went from enthusiast curiosity to £15 commodity item faster than driver support across three operating systems could keep pace with it. Buy one expecting a genuine upgrade over gigabit, verify the driver is doing its job once it arrives, and the shortfall against the advertised figure stops being a disappointment and starts being just another spec-sheet number to read with appropriate scepticism.
Related Content
