10 Gigabit External Ethernet Adapters: Do They Really Deliver?

Spoiler: yes, some of them can deliver, but there are quite a few nuances.

I’m sure many of you have said goodbye to the good old 1 Gigabit Ethernet networks, and rightfully so: even top-notch Wi-Fi can beat 1 Gbps these days. And yet, nothing compares to a super-fast wired connection: stable, reliable, interference-free. It’s just that 1 Gbps is not enough anymore.

What options are there on the table in 2026? If you’re looking beyond 1 Gbps, the upgrade path is no longer just about switches and cabling, it’s also about the Ethernet adapters themselves. In corporate environments, the most common choice is still PCIe network cards, available in everything from 2.5 GbE up to 100 GbE, often with SFP+ or QSFP ports for fiber and DAC connectivity. These are ideal for servers and high-performance workstations where you want maximum throughput and low latency. On the SOHO side, we’re looking at USB or Thunderbolt Ethernet adapters, including 2.5, 5, and 10 GbE options. They provide (or, rather, promise to provide) a simple plug-and-play upgrade without opening the machine, making it a practical way to access full 10G speeds externally.

So, shall we give these external adapters a spin? Can they really deliver 10 Gbps?

Test Setup

Testing a pair of network adapters is straightforward. What you need is:

• Two fast modern computers: you don’t want the computers’ hardware to be a bottleneck. Therefore, the buses should be modern and fast, the CPU shouldn’t be five or ten years old, and they should be running a modern OS. USB4, Thunderbolt, and all that jazz.

• Two identical Ethernet adapters to be tested. That’s easy, all you need about a hundred dollars for each. Yes, they’re not exactly cheap.

• A decent Ethernet RJ45 cable, Cat6a or better.

• Good software that can reliably measure network throughput; Tessabyte would be a great choice.

Now, plug in the adapters and configure their TCP/IP settings so that they can “talk” to each other without going through another network. For example, manually set IPv4 addresses to 192.168.0.10 and 192.168.0.11 and the subnet mask to 255.255.255.0. You may also want to disconnect these two computers from any other networks. Install testing software on both of the computers, run the server component on one of them and the client component on the other, and yeah, that’s it, you’re good to go.

Baseline Test – 5 Gbps

A while back, I played with a pair of WAVLINK 5Gbps USB C to Ethernet adapters: not yet 10 Gbps, but inexpensive and quite reliable.

So, I used these adapters as a baseline and to see if moving from 5 Gbps to 10 Gbps hardware makes any difference. Here are my results for these adapters:

As you can see, the results are very decent; 4,500 Mbps is close to the maximum achievable throughput. UDP is slightly worse, but this is not uncommon for any network adapter test for the reasons I’ll explain below. Note the figures and let’s move on to the 10 Gbps gear.

OWC Thunderbolt 3 10G Adapter

Now, the real 10 Gbps contestant, OWC Thunderbolt 3 10G Ethernet Adapter. By the way, OWC stands for “Other World Computing”, in case you were wondering.

Maybe in the “other worlds” it performs well. In our lab, it mostly failed:

There isn’t much to say about this chart. TCP throughput is unstable and, most importantly, at least 50% below the expected level for a 10 Gbps. UDP throughput is so low that any decent 1 Gbps adapter would beat these numbers. I’ve tried to change the setup to ensure that this is not my or my software or hardware fault, with some limited success. Running the whole thing on two Macs slightly improved the situation.

On macOS, as you can see above, the adapters could achieve about 9.4 Gbps of TCP throughput in one of the directions, but not in the other one. Still, my verdict is “It didn’t deliver, at least on Windows”.

SABRENT USB4/Thunderbolt 3/4 Adapter

The final contestant is SABRENT 10G Ethernet Adapter, a nice bus-powered (like all of the other adapters mentioned above, by the way) box in aluminum housing.

Let’s see if this one can deliver consistent 10G throughput:

Finally — stable 9.2–9.4 Gbps of TCP throughput in both directions!

When it comes to UDP, the numbers are not that shiny. There’s a reason for that. Seeing TCP outperform UDP on a 10 Gbps link is not unusual. In many 10 Gbps adapter tests, TCP can be much faster than UDP not because TCP is “lighter,” but because the whole network stack is heavily optimized for TCP. Modern NICs and drivers use offloads such as segmentation offload, checksum offload, large receive offload, interrupt moderation, and receive-side scaling. These features let the OS hand large chunks of TCP data to the adapter efficiently, instead of processing every packet individually. UDP is different. If a test sends a flood of typical-sized datagrams, the system often hits a packet-per-second ceiling imposed by CPU, driver, and NIC processing limits long before it reaches the theoretical 10G bandwidth. Once the sender, receiver, driver, or adapter buffers can’t keep up, packets are simply dropped. At that point, UDP throughput collapses while TCP continues to push data efficiently, which is why TCP often looks “faster” in real-world adapter tests.

Too long? Here’s the short version: small UDP packets increase packet rate and reduce throughput.

We used Tessabyte that sends UDP packets with a payload size of 1,300 bytes by default (chosen conservatively to avoid fragmentation on most IPv4/IPv6 paths), and with that payload size, a sniffer shows 170,000 UDP packets per second (that’s for the throughput you see on the chart). Can we do anything about it? Sure, let’s try the same test with the UDP payload size set to 9,216 bytes. Using jumbo UDP works fine in a LAN with jumbo frames enabled end-to-end, but only in a clean, consistently configured environment. Across a WAN, jumbo UDP packets won’t work well (and frankly, such huge UDP packets are not typical for general Internet traffic anyway), but hey, we’re just testing. The theory says that if we decrease packet rate by increasing UDP packet size, we’ll see better UDP throughout.

And here is what happened (I’m testing UDP only, in order not to clutter the chart):

We’ve improved UDP throughput to 1.6 – 1.7 Gbps by increasing payload size (the observed packets-per-second rate dropped to about 50,000 pps). Still not much, especially given that the 5Gbps WAVLINK adapter we tested above easily beats these UDP throughput numbers. A better driver? A better chipset? Frankly, I don’t know.

Conclusions

So yes, if you wondered whether 10 Gbps external adapters can deliver – some of them can, at least when we talk about TCP.  UDP performance lags behind, but first of all, that’s not exactly their fault, UDP is kind of TCP’s poor cousin. Second, most users won’t need stellar UDP performance, but UDP-heavy professional workloads should test it explicitly.

Don’t touch the adapters, though :-) I mean that literally. All of the tested devices get really hot. One can’t fry eggs on them, but they are about 60°C / 140°F under full load.

Oh, and In Case You Had Doubts…

We’re not affiliated with any of the above-mentioned hardware vendors. We’re simply passionate about network testing at Netmantics!