Realtek RTL8157 5 GbE adapters with 10 Gb USB interface - Wavlink, WisdPi

[tusk9541]

Wavlink: https://www.amazon.com/gp/product/B0DBZ41BPL
WisdPi: Realtek RTL8157 USB 3.2 5Gbps Wired Base-T Ethernet Adapter (WP-UT5)

So these have been out for a couple months now, and there's one other thread, but it's a review request and it's not popping up on search, so I'll post here in the proper subforum for discussion. I'm gonna copy my post from that thread below:

__________

I've been using the Wavlink on a Thinkpad P16s AMD 7840u. Out of the USB4 port I have a Kensington SD2600T Thunderbolt 4 hub. Connected directly to one of the TB downstream ports, the Wavlink maxes out on iperf3, ~4.74Gb/s, both upload and download. Connected to the USB-A 10Gbps port in the front with an adapter, I get full upload, but the download drops to ~2-3 Gb/s.

But... I've also experienced disconnections. Before taking the hub out of the equation I decided to search and it appears that this thing is still unstable with the latest Realtek drivers 1157.15.20.327, so I'm gonna wait cause testing without the hub is going to be a bit of a hassle due to my setup and lack of space.

Also I've noticed in the Amazon listing they've added another "style", besides the "340G" that we have, there's a "341G" now, without any discernible difference, with identical photos and description. One wonders if the disconnections aren't due to the Realtek chip and something else in the 340G is faulty. I'm past my return date, but if anyone that has the 340G and can still return it wants to test...

 

[jerrythea]

To further the discussion in the proper forum, I am using the Realtek 8157 based on the WisdPi implementation, which seems a little bit more stable.

However, there is denying that It is still bleeding edge.

With Proxmox or other Linux Kernel on 6.11++, and the driver provided on the realtek site, I have had good luck on almost all requirements.

iperf3, scp, etc get expected performance.

However, I am now having an issue with NFS server on top of this NIC. It is odd, but it locks up the CPUs, sometimes in unrelated network drivers (igc-intel 226v) and sometimes in r8152 (custom driver).

If you need it to just work, you will need either a PCIe slot, or get the thunderbolt 10GbE NIC from OWC. It runs really hot!, and is not cheap (~$150) but it might be more stable.

I went ahead and put the NFS server on the OWC 10 GbE adapter, and now I have no issues on either client or server-side.

Here are some NFS stats:

# dfu .
Filesystem     Type  Size  Used Avail Use% Mounted on
/dev/nvme1n1p2 ext4  1.8T  1.2T  592G  67% /mnt/pve/pve4vm
# dfu  /mnt/pve/BigNFSDisk/tmp/
Filesystem                         Type  Size  Used Avail Use% Mounted on
fastnfshost.jabramson.com:/BIGDISK nfs4  3.6T  653G  2.8T  19% /mnt/pve/BigNFSDisk

ll OracleLinux-R8-U6-x86_64-dvd.iso
11G -rw-r----- 1 root root 11G 2024/11/20-03:20:03.04 GMT OracleLinux-R8-U6-x86_64-dvd.is

Reading from NFS to local fast NVMe storage:

# dd if=/mnt/pve/BigNFSDisk/tmp/OracleLinux-R8-U6-x86_64-dvd.iso of=OracleLinux-R8-U6-x86_64-dvd.iso bs=1k status=progress
11163569152 bytes (11 GB, 10 GiB) copied, 25 s, 447 MB/s
11082752+0 records in
11082752+0 records out
11348738048 bytes (11 GB, 11 GiB) copied, 25.7294 s, 441 MB/s

Writing from local fast NVMe storage to NFS:

# dd of=/mnt/pve/BigNFSDisk/tmp/OracleLinux-R8-U6-x86_64-dvd.iso if=OracleLinux-R8-U6-x86_64-dvd.iso bs=1k status=progress
11261029376 bytes (11 GB, 10 GiB) copied, 15 s, 751 MB/s
11082752+0 records in
11082752+0 records out
11348738048 bytes (11 GB, 11 GiB) copied, 23.0598 s, 492 MB/s

 

[jerrythea]

Here is an expected speed result from oneWisdPi to another, via the Microtik CRS304-4XG [4 port 10GbaseT, 1 port 1GBaseT]:

scp foobar.qcow2 fastpve:/mnt/pve/pvevm/tmp
foobar.qcow2                                                             100%   47GB 485.1MB/s   01:38

 

[jerrythea]

And same result when going from one WisdPi to an OWC 10GbaseT TB3 host (speed difference likely due to current traffic on this host):

scp foobar.qcow2 fastnfshost:/BIGDISK/tmp
foobar.qcow2                                                              100%   47GB 406.0MB/s   01:57

And some iperf3 results:
WisdPi to WisdPi:

iperf3 -c fastpve
Connecting to host fastpve, port 5201
[  5] local 192.168.90.2 port 56464 connected to 192.168.90.1 port 5201
[ ID] Interval           Transfer     Bitrate         Retr  Cwnd
[  5]   0.00-1.00   sec   564 MBytes  4.73 Gbits/sec    0   1.20 MBytes
[  5]   1.00-2.00   sec   561 MBytes  4.71 Gbits/sec    0   1.20 MBytes
[  5]   2.00-3.00   sec   561 MBytes  4.71 Gbits/sec    0   1.20 MBytes
[  5]   3.00-4.00   sec   560 MBytes  4.70 Gbits/sec    0   1.20 MBytes
[  5]   4.00-5.00   sec   561 MBytes  4.71 Gbits/sec    0   1.35 MBytes
[  5]   5.00-6.00   sec   561 MBytes  4.71 Gbits/sec    0   1.35 MBytes
[  5]   6.00-7.00   sec   561 MBytes  4.71 Gbits/sec    0   1.35 MBytes
[  5]   7.00-8.00   sec   561 MBytes  4.71 Gbits/sec    0   1.43 MBytes
[  5]   8.00-9.00   sec   561 MBytes  4.71 Gbits/sec    0   1.43 MBytes
[  5]   9.00-10.00  sec   561 MBytes  4.71 Gbits/sec    0   1.50 MBytes
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bitrate         Retr
[  5]   0.00-10.00  sec  5.48 GBytes  4.71 Gbits/sec    0             sender
[  5]   0.00-10.00  sec  5.48 GBytes  4.71 Gbits/sec                  receiver

WisdPi to OWC

iperf3 -c fastnfshost
Connecting to host fastnfshost, port 5201
[  5] local 192.168.90.2 port 59998 connected to 192.168.90.5 port 5201
[ ID] Interval           Transfer     Bitrate         Retr  Cwnd
[  5]   0.00-1.00   sec   562 MBytes  4.71 Gbits/sec    0   1.48 MBytes
[  5]   1.00-2.00   sec   561 MBytes  4.71 Gbits/sec    0   1.55 MBytes
[  5]   2.00-3.00   sec   561 MBytes  4.71 Gbits/sec    0   1.55 MBytes
[  5]   3.00-4.00   sec   561 MBytes  4.71 Gbits/sec    0   1.64 MBytes
[  5]   4.00-5.00   sec   561 MBytes  4.71 Gbits/sec    0   1.64 MBytes
[  5]   5.00-6.00   sec   560 MBytes  4.70 Gbits/sec    0   1.74 MBytes
[  5]   6.00-7.00   sec   561 MBytes  4.71 Gbits/sec    0   1.74 MBytes
[  5]   7.00-8.00   sec   561 MBytes  4.71 Gbits/sec    0   1.74 MBytes
[  5]   8.00-9.00   sec   561 MBytes  4.71 Gbits/sec    0   1.74 MBytes
[  5]   9.00-10.00  sec   561 MBytes  4.71 Gbits/sec    0   1.74 MBytes
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bitrate         Retr
[  5]   0.00-10.00  sec  5.48 GBytes  4.71 Gbits/sec    0             sender
[  5]   0.00-10.00  sec  5.48 GBytes  4.70 Gbits/sec                  receiver

 

[MountainBofh]

I have the Wavlink. Test computer is an HP Spectre X360 notebook running Windows 10 21H2 LTSC with newest drivers from Realtek, and I also tested using Parted Magic with kernel 6.8. Test switch is a Unifi XG24 in my work data center. I ran Iperf3 to a server on the same switch, and speedtest to a few different test servers.


Starting with baseline info. A Debian 12 server on the same Unifi switch was used to test speedtest numbers. I received roughly 9.5gb down and up,. Iperf3 to another server one switch over showed consistent 9.5gb speeds.

The laptop in Windows showed inconsistent numbers, both in Iperf3 and speed test. While the Windows based Iperf3 is known to be unreliable in itself, both set of tests gave roughly 2.5 to 3.5gb speeds. Tweaking the receive and transmit buffers helped some, but numbers were still all over the place.

In Parted Magic, iperf3 showed a very consistent 4.5gb connection. Speedtest was still all over the place, giving roughly the same performance as Windows.



Conclusions so far - As I expected, the Realtek 8157's drivers are still rather immature at this point in Windows. I'll try to test later with a newer kernel and see if that leads to more reliable numbers. I was also running on battery power, so maybe testing this on a desktop system might show some insights as well.

 

[jerrythea]

On Linux the driver's may be a little bit more reliable, and this was for the WisdPi version of the Realtek 8157, which I believe is a more reliable hardware implementation.
USB 3.2 To 5GbE Adapter (WP-UT5) Realtek RTL8157

I still do not know what was going on with NFS, but some sort of interaction caused issues. It might not actually be the hardware or driver, but some sort of Kernel bug.

But for standard file transfers and other types of network activity they have been reliable.

I recommend ordering from Amazon:
USB 3.2 to 5GbE Adapter (WP-UT5) Realtek RTL8157 Wired LAN Network Connection 5G, 2.5G, 1G, 100Mbps

I would not recommend using the copy/paste driver from the WisdPi WiKi

The latest Realtek drivers I used are below:

USB NIC Linux driver for kernel up to 6.10

2.19.2

2024/11/05

84 KB

 

[MountainBofh]

I tested my Wavlink on my Debian 12 testbox at home. Long story short - it refused to work, both with the built in driver in kernel 6.11, and after compiling the realtek driver.

The testbox is an AMD 5700x using a B550 motherboard, so I wonder if the current state of the realtek driver or firmware just doesn't like AMD based systems.

 

[agreenbhm]

I tested my Wavlink on my Debian 12 testbox at home. Long story short - it refused to work, both with the built in driver in kernel 6.11, and after compiling the realtek driver.

The testbox is an AMD 5700x using a B550 motherboard, so I wonder if the current state of the realtek driver or firmware just doesn't like AMD based systems.

@MountainBofh - which Realtek driver did you compile? I too am using kernel 6.11 (though on Ubuntu 24.10). The built-in driver (cdc_ncm, IIRC) gave a max of 705mbps at half-duplex with auto-negotiation off (none of those parameters could be changed). However, compiling this driver worked flawlessly. Just want to make sure that's the one you tried. My system is an HP Zbook with an Intel CPU though so can't speak to whether AMD may be the source of the issue.

 

[MountainBofh]

I used the latest driver from their website last night, 2.19.2.

If I can locate a newer Intel machine to test this on, I'll repeat my test. Most of the better machines I have access to, are AMD though

 

[MountainBofh]

Tested again using a HP Prodesk 600G6 that has a 10th gen Intel I5 in it running Win10 LTSC 21H2, and a USB-C 10gb port.

In Windows 10, I got the following
Ran speedtest, and the card was able to hold my full internet speed (2gb fiber) over several tests.
Iperf3 to my Debian 12 box via my mikroktik switch held a consistent 3.6gb.

Booting into Linux (Debian 12 with kernel 6.11 and the Realtek 2.19.2 drivers)
Speedtest was the same, delivering internet line speed.
iperf3 gave about 3.7gb per sec.


Conclusions so far about the R8157.
I believe the drivers are still fairly raw, and you're going to see compatibility and speed issues as a result. I've not seen the same issue with the R8126 (the 5gb PCI-e card), so I'm guessing that Realtek needs to iron out the usb interface a bit more on this card.

I'm guessing that in about a year, this will be a more solid choice with driver updates.

 

[pimposh]

Would you mind doing a 100% load test for 72h please? AQC111 based adapters were also hitting circa 450MB/s sustained transfers but were (at least two units of one model i tried) randomly disconnecting so were unreliable in any use case but temporarily next to NIC-less notebook...


These days I'm running a couple of (2.5Gbps) R8156BG adapters and I find them stable and good enough as an upgrade for slightly older servers without spending almost no budget, but if the R8157 is stable over sustained load, given it costs almost the same, I'm tempted to give it a try.

 

[MountainBofh]

What OS do you want me to use for the load test?

 

[pimposh]

What OS do you want me to use for the load test?

Linux in general, feel free to choose something that you favor as long as it's not some super obsolete distri. Thanks!

 

[agreenbhm]

Tested again using a HP Prodesk 600G6 that has a 10th gen Intel I5 in it running Win10 LTSC 21H2, and a USB-C 10gb port.

In Windows 10, I got the following
Ran speedtest, and the card was able to hold my full internet speed (2gb fiber) over several tests.
Iperf3 to my Debian 12 box via my mikroktik switch held a consistent 3.6gb.

Booting into Linux (Debian 12 with kernel 6.11 and the Realtek 2.19.2 drivers)
Speedtest was the same, delivering internet line speed.
iperf3 gave about 3.7gb per sec.


Conclusions so far about the R8157.
I believe the drivers are still fairly raw, and you're going to see compatibility and speed issues as a result. I've not seen the same issue with the R8126 (the 5gb PCI-e card), so I'm guessing that Realtek needs to iron out the usb interface a bit more on this card.

I'm guessing that in about a year, this will be a more solid choice with driver updates.

I just got another one today and tested between my 2023 MBP and HP Zbook directly and saw 4.7Gbps on iperf3 (on the default test). I ordered the supposedly new model (341G) but received the original (340G) that I already had one of. I wonder if it's just something they listed on Amazon to differentiate batches because I don't see anything on the Wavlink site mentioning a new model.

 

[MountainBofh]

I'll use Debian 12 (that's my preferred server OS) with the 6.11 kernel. To get a proper test, I'll need to do this in my data center. Give me a few days to set this up and run the test.

 

[tusk9541]

I have an update about the "341G". I ordered one as well, and received the actual model: "WL-NWU340G Rev. a"
The older one I have, presumably "340G" although it may not have been labeled that on Amazon because at the time it was the only "style", is model "WL-NWU340G-A Rev. a".

So far the newer one (341G) hasn't disconnected though, whereas the older one (340G) disconnected pretty much every time I reconnected the cable, and had to reconnect several times or disable/enable it on Device Manager for it to work. I didn't update the drivers, it just used the same drivers previously installed.

 

[jerrythea]

My nodes continue to intermittently lockup. Not sure if it is unrelated fiddling, or if this unfinished driver from RealTek is the cause.

It is likely the driver, so my experiment with these 5 GbE dongles may be coming to an end for Proxmox and Linux.

 

[MountainBofh]

My nodes continue to intermittently lockup. Not sure if it is unrelated fiddling, or if this unfinished driver from RealTek is the cause.

It is likely the driver, so my experiment with these 5 GbE dongles may be coming to an end for Proxmox and Linux.

Just curious, what motherboard are you running your realtek's on? I suspect that the Realtek USB 5gb cards are temperamental on some USB controllers.

 

[agreenbhm]

I have an update about the "341G". I ordered one as well, and received the actual model: "WL-NWU340G Rev. a"
The older one I have, presumably "340G" although it may not have been labeled that on Amazon because at the time it was the only "style", is model "WL-NWU340G-A Rev. a".

So far the newer one (341G) hasn't disconnected though, whereas the older one (340G) disconnected pretty much every time I reconnected the cable, and had to reconnect several times or disable/enable it on Device Manager for it to work. I didn't update the drivers, it just used the same drivers previously installed.

Both of mine show "Rev. A", though only the second one was bought under the "341" listing; the other was from AliExpress which didn't list anything specific.

 

[WhiteNoise]

I asked and got this reply. Not sure if there is any difference or if they are doing some listings magic on Amazon.