DEV Community

Cover image for Studio Display or Dock Ethernet Slow? Your USB Adapter Cuts Downloads 75%
Alan West
Alan West

Posted on Originally published at blog.authon.dev

Studio Display or Dock Ethernet Slow? Your USB Adapter Cuts Downloads 75%

If your Ethernet adapter is plugged into a Studio Display, a monitor's USB hub or a Thunderbolt dock, your download speed may be a fraction of what you pay for, and nothing on screen will tell you. Oscar Molnar has a 2.5 gig symmetrical line. He'd had the adapter behind his Studio Display XDR for half a year without running a speed test through it. When he finally did, fast.com showed 622 Mbps down and 2.4 Gbps up.

In his words: "600 Mbps was more than fast enough for me to never realise I wasn't getting three quarters of the speed I was paying for."

I read his write-up and both comment threads (HN, Lobsters). Despite the title of his post, the monitor is mostly a bystander. The adapter uses a cheap USB Ethernet chip that can't tolerate a few microseconds of extra hub latency, so the same thing can happen on docks and other brands of monitor, and possibly with gigabit adapters. I'll go through the cause, a two-minute iperf3 test you can run yourself, and the fixes that actually helped.

The setup that loses 75% of the download speed

  • MacBook Pro M3 Pro
  • 2026 Studio Display XDR, one cable to the laptop
  • UGREEN USB-C 2.5G Ethernet adapter (Realtek RTL8156B) plugged into the display's USB port
  • UniFi Dream Router 7

His work laptop showed the same slow speed. The ISP checked their side and found nothing. When he plugged the same adapter straight into the Mac, iperf3 showed 2.3 Gbps and fast.com showed 2.2 down / 2.3 up. Moving it back behind the display dropped it to about 600 again.

Bandwidth isn't the limit. Molnar gets much more than 600 Mbps from an external SSD on the display's Thunderbolt port, and uploads through the adapter were fine. That rules out the "USB bandwidth is shared with the video signal" theory people reach for first. On HN, ricardobeat put it this way: "The bandwidth in question is nowhere near saturating even a single USB3 connection."

Why Ethernet through a monitor or dock gets slow

It comes down to hub latency combined with a small burst size. A USB 3 device sending data to the host sends a few packets, then waits for the host to say "go ahead" before sending more. Molnar read the adapter's USB descriptors:

$ python3 -c "import usb.core; e = usb.core.find(idVendor=0x0bda)[0][(0,0)][0]; print('packet size', e.wMaxPacketSize, 'bMaxBurst', e.extra_descriptors[2])"

# old adapter (RTL8156B)
packet size 1024 bMaxBurst 3

# new adapter (RTL8157)
packet size 1024 bMaxBurst 7
Enter fullscreen mode Exit fullscreen mode

bMaxBurst counts the packets after the first one. So the RTL8156B sends 4 KB per burst. His SSD enclosure sends 16 packets.

Now plug in the timings from the post. The go-ahead round trip through the display's hub takes around 7.6 µs, and sending 4 KB takes around 8.2 µs. Add those and you get 4 KB every 15.8 µs or so, which works out to roughly 2 Gbps from adapter to Mac. The router is pushing 2.5 Gbps into the adapter. The adapter's buffer fills, frames drop before macOS ever sees them (so nothing shows up in the logs), TCP sees loss and backs off hard. A 2 Gbps ceiling ends up as 600 Mbps actually delivered.

He proved it with smooth fixed-rate UDP from a Linux box on 2.5G, flow control off, reading the drop counter on the RTL8156B itself:

Sent Lost behind display Frames/s reaching the Mac Lost direct on Mac
1.46 Gbps 0% 123,800 0%
1.75 Gbps 0.05% 148,600 0%
1.94 Gbps 0.1% 165,000 0%
2.03 Gbps 2.7% 168,000 0%
2.12 Gbps 7.2% 168,000 0%
2.20 Gbps 10.5% 168,000 0.001%
2.28 Gbps 13.5% 168,000 0.02%

Behind the display, delivery flattens at about 168,000 frames/s, roughly 2 Gbps, whatever you send.

So who's at fault? The HN thread argues about it at length. dcrazy blames UGREEN for "cheaping out on a part", then adds that the Thunderbolt hop makes it worse: "The extra work of unwrapping the Thunderbolt tunnel around the USB protocol introduces additional latency." db48x blames Realtek: "It would only have cost pennies to double the buffer sizes." thefz blamed Apple, and Molnar answered: "Nope. This doesnt work on Lenovo either. Nothing to do with Apple"

I think the adapter deserves most of the blame. A 4 KB burst window on a 2.5G NIC only works when it sits right next to the host, and nobody puts "only works plugged directly into the laptop" on the box. Still, crote's point holds: when a $3,300 monitor performs worse than "a $35 docking station", that's "unacceptable".

Gigabit adapters may be affected too

Molnar's spare adapters are all gigabit, and by his testing the trouble doesn't start at 1 gig but around 2 Gbps. Commenters disagree:

  • brigade, with an RTL8153 gigabit adapter: "the combination of the Studio Display and ARM MacBooks is especially bad. Using an Intel MacBook with the display, or a different USB hub instead of the display, gets the full NIC bandwidth." In a second comment: "the speed of gigabit Ethernet adapters are also known to cut 60% behind a studio display".
  • unixhero says monitor Ethernet on "Philips, Lenovo get to about 300mbps and can not saturate 1gbps."
  • gh123man hit it on a Thunderbolt dock. A non-Thunderbolt cable between dock and laptop gave "full speed with no packet loss", which fits the Thunderbolt-latency theory. What they actually run with is flow control.

If you have gigabit fiber and a monitor hub, I'd test it rather than assume you're fine.

How to test Ethernet speed through a dock or monitor in two minutes

Do it twice: once through the dock or monitor, once with the adapter plugged straight into the laptop.

1. Run iperf3 against something on your LAN

A speed test website also measures your ISP, so use a local box if you can. Molnar ran the iperf3 server on his UniFi router for the direct test. -R reverses the direction, so the server sends to you, which is the download case:

# on the LAN machine
iperf3 -s

# on the laptop: download test, then upload
iperf3 -c 192.168.1.10 -R -t 20
iperf3 -c 192.168.1.10 -t 20
Enter fullscreen mode Exit fullscreen mode

If upload is fast and download isn't, that's the same pattern Molnar saw.

2. Check for loss with UDP at a fixed rate

Set the rate just under your link speed and see what the loss column says:

iperf3 -c 192.168.1.10 -u -b 2.2G -R -t 10
Enter fullscreen mode Exit fullscreen mode

3. Check the negotiated link speed

A 2.5G adapter that came up at 1000baseT will cap you at gigabit regardless of the dock.

# macOS: find the adapter (often en5-en9), look at "media:"
networksetup -listallhardwareports
ifconfig en7 | grep media

# Linux
ethtool eth0 | grep -i speed
ethtool -S eth0 | grep -i -E 'drop|miss|err'
Enter fullscreen mode Exit fullscreen mode

On a Mac, System Information shows the same under Network, and the USB section tells you which hub the adapter actually sits behind. On Linux, lsusb -t shows the tree and the negotiated USB speed for each device.

4. Repeat with a real speed test

Run fast.com or something similar both ways so you know what the difference costs you on your real line.

Fixes for slow Ethernet behind a Studio Display or dock

Fix Result in the source Catch
Adapter straight into the laptop 2.3 Gbps iperf3 You lose the one-cable desk
RTL8157 adapter (UGREEN 5G) behind the display 0 drops up to 2.4 Gbps sent Costs another adapter
Flow control on the router/switch ~2 Gbps down Pause frames can add latency for other machines
Realtek's driver instead of Apple's generic one 1.1 Gbps Still far short
Firmware update for bigger bursts None reported crote: "good luck with that"

Molnar went with the RTL8157. It sends 8 packets per burst over a USB link twice as fast, which he puts at a theoretical 4.5 Gbps from adapter to Mac, well above the 2.5 coming in. The hub latency doesn't change, but the buffer never fills.

cozzyd asked whether BBR congestion control, which doesn't back off as hard on packet loss, would help, then noted "it seems Apple still doesn't support BBR." toast0 replied that the client barely matters: on a download the sender's congestion control is the one that counts, so fast.com (or whatever server you pull from) would need to run BBR. toast0 thinks Netflix, which runs fast.com, already does. Either way, you can't fix it from the laptop.

My take: if you paid for a fast line, plug the laptop's adapter in directly for anything big, or buy an adapter with a larger burst size. Don't trust a dock's Ethernet port until you've run iperf3 through it. The one-cable desk is great, and it's also why this went unnoticed for half a year.

So, is your Ethernet going through a dock or a monitor? Run the -R test both ways and post your numbers with the dock, monitor or adapter model. I'd especially like to see gigabit setups on Apple Silicon, because brigade and Molnar disagree on whether those are safe.

Sources

Top comments (0)