Quick comparison, 1Gbps, 2.5Gbps, 5Gbps, 10Gbps Ethernet

[TheLostSwede]

Honestly, I don't know what's going on with this ancient thread, but as I just got myself a Realtek 2.5 USB adapter for my notebook, here are some quick tests using iPerf3 with my NAS with the same 10 Gbps card in it.

C:\iperf>iperf3 -c 192.168.1.250 -t10
Connecting to host 192.168.1.250, port 5201
[  4] local 192.168.1.34 port 50931 connected to 192.168.1.250 port 5201
[ ID] Interval           Transfer     Bandwidth
[  4]   0.00-1.00   sec   278 MBytes  2.33 Gbits/sec
[  4]   1.00-2.00   sec   274 MBytes  2.30 Gbits/sec
[  4]   2.00-3.00   sec   283 MBytes  2.37 Gbits/sec
[  4]   3.00-4.00   sec   282 MBytes  2.36 Gbits/sec
[  4]   4.00-5.00   sec   279 MBytes  2.34 Gbits/sec
[  4]   5.00-6.00   sec   282 MBytes  2.37 Gbits/sec
[  4]   6.00-7.00   sec   283 MBytes  2.37 Gbits/sec
[  4]   7.00-8.00   sec   279 MBytes  2.34 Gbits/sec
[  4]   8.00-9.00   sec   282 MBytes  2.36 Gbits/sec
[  4]   9.00-10.00  sec   282 MBytes  2.36 Gbits/sec
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bandwidth
[  4]   0.00-10.00  sec  2.74 GBytes  2.35 Gbits/sec                  sender
[  4]   0.00-10.00  sec  2.74 GBytes  2.35 Gbits/sec                  receiver

iperf Done.

I saw about a five percent increase in CPU usage on an "old" quad core Core i9-9900K when running iPerf. (Using an old Corsair One at the moment.)

Seems like I got an adapter with one of the old Realtek 8156 chips, as the download performance is arse.

C:\iperf>iperf3 -c 192.168.1.250 -t10 -R
Connecting to host 192.168.1.250, port 5201
Reverse mode, remote host 192.168.1.250 is sending
[  4] local 192.168.1.34 port 49336 connected to 192.168.1.250 port 5201
[ ID] Interval           Transfer     Bandwidth
[  4]   0.00-1.00   sec  97.0 MBytes   814 Mbits/sec
[  4]   1.00-2.00   sec  83.7 MBytes   702 Mbits/sec
[  4]   2.00-3.01   sec  91.4 MBytes   760 Mbits/sec
[  4]   3.01-4.01   sec  62.2 MBytes   522 Mbits/sec
[  4]   4.01-5.01   sec   144 MBytes  1.20 Gbits/sec
[  4]   5.01-6.01   sec  78.7 MBytes   663 Mbits/sec
[  4]   6.01-7.00   sec  57.2 MBytes   481 Mbits/sec
[  4]   7.00-8.00   sec  61.3 MBytes   515 Mbits/sec
[  4]   8.00-9.02   sec  83.7 MBytes   693 Mbits/sec
[  4]   9.02-10.00  sec  96.5 MBytes   821 Mbits/sec
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bandwidth       Retr
[  4]   0.00-10.00  sec   857 MBytes   719 Mbits/sec  8182             sender
[  4]   0.00-10.00  sec   856 MBytes   718 Mbits/sec                  receiver

iperf Done.

 

[Assimilator]

Seems like I got an adapter with one of the old Realtek 8156 chips, as the download performance is arse.

C:\iperf>iperf3 -c 192.168.1.250 -t10 -R
Connecting to host 192.168.1.250, port 5201
Reverse mode, remote host 192.168.1.250 is sending
[  4] local 192.168.1.34 port 49336 connected to 192.168.1.250 port 5201
[ ID] Interval           Transfer     Bandwidth
[  4]   0.00-1.00   sec  97.0 MBytes   814 Mbits/sec
[  4]   1.00-2.00   sec  83.7 MBytes   702 Mbits/sec
[  4]   2.00-3.01   sec  91.4 MBytes   760 Mbits/sec
[  4]   3.01-4.01   sec  62.2 MBytes   522 Mbits/sec
[  4]   4.01-5.01   sec   144 MBytes  1.20 Gbits/sec
[  4]   5.01-6.01   sec  78.7 MBytes   663 Mbits/sec
[  4]   6.01-7.00   sec  57.2 MBytes   481 Mbits/sec
[  4]   7.00-8.00   sec  61.3 MBytes   515 Mbits/sec
[  4]   8.00-9.02   sec  83.7 MBytes   693 Mbits/sec
[  4]   9.02-10.00  sec  96.5 MBytes   821 Mbits/sec
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bandwidth       Retr
[  4]   0.00-10.00  sec   857 MBytes   719 Mbits/sec  8182             sender
[  4]   0.00-10.00  sec   856 MBytes   718 Mbits/sec                  receiver

iperf Done.

Yikes, less than half the maximum theoretical performance at best, and well under a third average... I'd return that adapter and look for one using the RTL8156BG which apparently fixes this problem (see https://www.cnx-software.com/2022/0...8156bg-usb-3-0-to-2-5gbe-dongle-to-the-rescue).

For future readers, there are at the time of this writing three revisions of Realtek's 2.5GbE-to-USB chip: the original 8156, the improved but still not great 8156B, and the current 8156BG which seems to be the "fixed" version.

 

[TheLostSwede]

Yikes, less than half the maximum theoretical performance at best, and well under a third average... I'd return that adapter and look for one using the RTL8156BG which apparently fixes this problem (see https://www.cnx-software.com/2022/0...8156bg-usb-3-0-to-2-5gbe-dongle-to-the-rescue).

For future readers, there are at the time of this writing three revisions of Realtek's 2.5GbE-to-USB chip: the original 8156, the improved but still not great 8156B, and the current 8156BG which seems to be the "fixed" version.

Yeah, I've contacted the company first to see what they say, but there's no way to tell what chip you get unless the device vendor lists it specifically.
Jean-Luc obviously only cover Linux stuff, but I can't find anyone that has tested both directions on a Windows system, as Serve the Home only found issues with older drivers on the dongles they tested, but they only appear to be testing in one direction, not both.

 

[Assimilator]

Yeah, I've contacted the company first to see what they say, but there's no way to tell what chip you get unless the device vendor lists it specifically.

Ugh, I hate this. Probably best to go with Amazon reviews then.

Jean-Luc obviously only cover Linux stuff, but I can't find anyone that has tested both directions on a Windows system, as Serve the Home only found issues with older drivers on the dongles they tested, but they only appear to be testing in one direction, not both.

Maybe hit up STH and ask them if they'd be interested in doing those tests? I see they've tested a few of these dongles so maybe a "group test" would be a decent article for them to put up.

 

[TheLostSwede]

Ugh, I hate this. Probably best to go with Amazon reviews then.

Amazon in Taiwan? Yeah, no.

Maybe hit up STH and ask them if they'd be interested in doing those tests? I see they've tested a few of these dongles so maybe a "group test" would be a decent article for them to put up.

It does seem like they missed this issue, which is a bit weird, but they're seemingly only testing their switches one way as well.
I don't know any of the guys over there though.

 

[Assimilator]

Amazon in Taiwan? Yeah, no.

UPDATE YOUR LOCATION IN YOUR PROFILE REEEEE

 

[TheLostSwede]

UPDATE YOUR LOCATION IN YOUR PROFILE REEEEE

I'm going back to Sweden next month... We'll see for how long, but the plan is for the long term.

Still not heard anything back from j5Create's support team, it's been over 24 hours...
Got another five days until I need to return it to get my money back.

 

[mechtech]

I have extensively tested different Realtek and intel chips, and have very different results. (It is dutch forum, but you can use translate button if screenshot is not clear)
I think there is your problem:
Interrupt moderation disabled.
Large send offload disabled.
Power saving features disabled.
I have all 3 enabled, especially interrupt moderation is the most important.

You did not specify what chips exactly you use, there are many different ones. Latest Realtek chips do not use CPU excessively.
Older intel chips (i210) have special setting “DMA coalescing” that needs to be enabled. It has amazing effect on CPU and power consumption.
@chrcoluk

From all your testing, which network card do you use most/all of the time?

 

[TrantaLocked]

I have extensively tested different Realtek and intel chips, and have very different results. (It is dutch forum, but you can use translate button if screenshot is not clear)
I think there is your problem:
Interrupt moderation disabled.
Large send offload disabled.
Power saving features disabled.
I have all 3 enabled, especially interrupt moderation is the most important.

You did not specify what chips exactly you use, there are many different ones. Latest Realtek chips do not use CPU excessively.
Older intel chips (i210) have special setting “DMA coalescing” that needs to be enabled. It has amazing effect on CPU and power consumption.
@chrcoluk

I ran some iperf3 tests with my Ryzen 5 7600 and Realtek RTL8125BG and the results are incredible. None of the power saving or even offload settings changed CPU usage, but the one setting I identified, as you did, that made a difference was "Interrupt Moderation". At gigabit transfer, 15% CPU total usage in task manager with it OFF, to 2% CPU total usage with it ON. WHAT????

However, the CPU usage differential isn't at an appreciable degree until over 100mbps. Under that it's <=2% usage on or off, and while there is still a usage difference, it's tiny and would make no appreciable difference in performance or power usage. As an average user I don't see speeds that high very often even with multiple 4K streams! Maybe for direct downloads for games or movies and you have a fast connection. It could be helpful if you're doing CPU heavy work while doing home server transfers or downloading through Steam and the like.

I'm also not sure about the latency implications on gaming. With Interrupt Moderation on and off, latency is <1ms to my router. It could still make a difference in the flow of things, but I really do wonder how much of an effect it has in a game like LoL where your ping to the central server is usually well over 10ms anyway. Perhaps with it on it would reduce ping spikes when playing video in the background?

I did not find much of a difference in LatencyMon until 900mbps+, at which point interrupt latency dramatically turns up to ~400us average with Interrupt Moderation OFF. Below that though, it's pretty stable at under 100us average, on or off. I did run a few 500mbps tests where there were rare spikes to 200-300us with it OFF but it was hard to repeat consistently.

For this specific Realtek 2.5G RTL8125BG chipset with driver version 10.71.312.2024, very low CPU usage is achievable at ultra high transfer speeds as long as Interrupt Moderation is enabled. I'm not sure how my numbers compare to chrcoluk's since they did per core usage and with different model CPU cores. Maybe they could report on task manager numbers for comparison to mine?

It would be cool if Linus did a video on this and brought back Shroud and other gamers to find out if settings like this actually affect perceived online multiplayer game smoothness.

Anyway, I'm leaving it off for now since I rarely even use more than 50mbps let alone over 100mbps. For people who do CPU intensive work while doing large transfers or downloads, the CPU usage and interrupt latency reduction at very high transfer speeds could be worth enabling it for.