USB to 5GbE Adapters the Good Bad and Ugly

STH Merch on Teespring: the-sth-merch-...
STH Top 5 Weekly Newsletter: eepurl.com/dryM09
In our round-up, we compare three Marvell/ Aquantia AQC111U adapters that convert USB 3.0/ 3.1 to 5GbE. Although these three units from StarTech (USG5GA30), Sabrent (NT-SS5G), and TRENDnet (TUC-ET5G) all use the same controller chip, we see differences in performance, stability, physical specs, and pricing. In this video, we discuss what we found when comparing the three units.
--------------------------------------------
Individual USB 3 to 5Gbase-T Adapter Reviews
--------------------------------------------
TRENDnet TUC-ET5G review: www.servetheho...
StarTech USG5GA30 review: www.servetheho...
Sabrent NTSS5G review: www.servetheho...
--------------------------------------------
Other Referenced USB 3 to 5Gbase-T Adapter Reviews
--------------------------------------------
TweakTown's review of the TRENDnet unit where they saw the similar performance, but not the dropping issue (trying to show both sides of this): www.tweaktown....
ZDnet's "review" of the TRENDnet unit: www.zdnet.com/...

[@epobirs]

The moment I saw USB 3.1 Gen 1 on the chart the problem was obvious. The decision to have this Gen 1 / Gen 2 stupidity attached to USB 3.1 is a source of endless annoyance.

[@kojetono5853]

Yes, I still dont get why didnt they just go with 3.0 for 5gig, 3.1 for 10gig, and 3.2 for that gen 2x2 nonsense.

[@epobirs]

@@kojetono5853 There are some actual differences, mostly in the power management modes that came with the Type C connector, so you could have a laptop that did everything including power through the USB ports. But most of it is more esoterica than anything else for people who just want to know if higher speed is supported. Personally, I think it would have been better if the spec treated the 10Gb/s mode as a requirement for certification as USB 3.1. It would have saved consumers a lot of confusion and probably accelerated adoption of the 10Gb/s and later 20Gb/s (requiring Type C) modes.

[@spx2327]

It is always a joy listening to Patrick, no matter what he is reviewing or saying.

[@supremelawfirm]

Patrick's boots-on-the-ground experience is VAST!

[@nekron75]

The guy oozes enthusiasm and passion.

[@teadott]

Engineers: Boss we did it 5gbps QA Team: we have a small issue, its not consistent Boss: Don't worry about it, most people wont even notice. STH: USB to 5GbE Adapters the Good Bad and Ugly Boss: Oh no, oh noooooo

[@JarrodsTech]

Thanks for the comparison, I'm looking at getting some of these to speed up my laptop testing, pretty annoying none of them are using USB 3.1 Gen 2 as that's 10gbps, at least then in theory these 5gbe models would get full speed. Like you say though, I'll still take the 3x+ boost over gigabit.

[@javaguru7141]

Isn't USB half-duplex though? It would still prevent it from running at full speed both ways.

[@camerontgore]

11:55 and that's why it don't read ZDNet articles...

[@AlanShieh]

Great video! With USB3.1gen1 you have 20% overhead just from the bit encoding scheme, which brings the max data throughput down to 4Gb/s, before any protocol level overheads. Gen2 has a 5% or so encoding overhead, and supports store and forward in the hub when talking to slower speed devices, so you should be able to jam a lot of high performance devices (even previous gen USB3) on a single root controller. That’s something I have been playing with for 4k video capture cards and external drive arrays as a flex. Might be a fun video or review to do on mini micro systems.

[@supremelawfirm]

Glad you mentioned these "frame" sizes: 8b/10b = "legacy frame" e.g. 5G / 10 = 500 MB/sec 128b/132b = "jumbo frame" e.g. 5G / 8.250 = 606 MB/sec (to compare) 128b/132b = "jumbo frame" e.g. 10G / 8.250 = 1,212 MB/sec Thus, 10/8 = 25% encoding overhead 132 / 128 = 3.125% encoding overhead (USB) 130 / 128 = 1.563% encoding overhead (PCIe 3.0)

[@aydngulgun8936]

when you mentioned backpacking if the concern is potability/mobility detachable cable is a very big plus, that cable can be used for lots of things

[@ServeTheHomeVideo]

I was stretching for why one may spend $102 on the StarTech.

[@morosis82]

Not sure why you'd need 5Gbps ethernet while backpacking. I've done a couple years backpacking and never had a time I've even needed 1gbps. As far as I'm aware the only real reason for faster ethernet is to connect to a server on a local network. For fast storage on the go, an SSD in a 3.1 case would be far preferable.

[@miguelagueda3928]

I can only imagine someone wanting a lighter version with a fixed cable for something like a tiny mini micro project where they're going to be hanging from the computer all the time, but you might as well use a unit with a longer detachable cable and have it lay on the table/floor underneath

[@AudreyRobinel]

@@morosis82 You could have a laptop to work, with only GBE. At home you have your docking station, but at work, moving in the building you'd have your usb to ethernet dongle to connect at full speed wherever you work... In my workspace, i work in different places, it would come handy. However, i'd take the metal unit anyway : removable cable means i can use the same cable for multiple stuff, the cable won't bend and kink rendering the device unusable; i would be less worried about the unit being damaged; and anyway, what is the weight difference between those two units? As for the workspace, another solution would be to settle for less speed (2.5GB), and buy a few, have them located in the sports where you usually work, permanantly, since they are about 20$.

[@aydngulgun8936]

@@morosis82 backpacking here is not like traveling the world with just a backpack but more like needing to move around in different job sites etc.

[@ChrisBurnes]

Been waiting on this video. Good stuff. I just bought the 2.5gbe Sabrent adapter, still glad I did. It was about $30. Even if the $60 Sabrent 5gbe adapter is the best deal of the 5gbe adapters... When compared to the 2.5gbe adapters, it is 2x the cost, and definitely NOT 2x the performance.

[@JGoodwin]

If you buy this for say work, then for a person earning just $10/hr would break even on 3 hours of extra I/O bound activity that would have been used for useful work. If you are on a LAN with large files, let's say this happened for just 1 minute a day, then the break even would happen at 9.5 months of regular working days. If you earn $100/hr with the same workload, the break even would happen after 1 month of use.

[@miguelagueda3928]

It makes a lot of sense that is a usb limitation, given that 3.5 Gbit equates to a bit over 400MB/s which is the max transfer speed most SSDs over USB 5Gbit reach as well

[@HakonBroderLund]

Always showing off your newest camera gear in the left corner? This time the RF 70-200 2.8. Such a workhorse of a lens!

[@ServeTheHomeVideo]

:-) Often the new camera gear gets staged on the table and gets added to the shot. I thought it would hide the iPad.

[@HakonBroderLund]

@@ServeTheHomeVideo As a photographer and general tech nerd I always notice these showpieces. Great add haha. Products on review on camera right, and casual show of new tech on the camera left. Like it!

[@ServeTheHomeVideo]

@@HakonBroderLund I wish I had the skill to put these lenses to use! Since you may appreciate this, the main video was C70 w/ 15-35 f/2.8. The photos are all on the R5 w/ 28-70 f/2.0.

[@HakonBroderLund]

@@ServeTheHomeVideo Much appreciated! Looking to get that exact setup myself. Upgrading from 5D3 for stills and BM UMP and P4K for video with a bunch of EF lenses to your setup and RF lenses. Computer is a 3960X threadripper, more or less similar to your editing machine. Working on a TrueNAS build for the footage. Your page is a huge source of excellent info for that!

[@DavidTrejo]

You sly dog 😈

[@Chris-rm1pn]

I wonder if there are any USB 3.1 gen 2 adapters and how these would perform in linux/bsd.

[@kilrahvp]

Unfortunately nobody makes a gen2 chip yet...

[@Chris-rm1pn]

@@kilrahvp I've recently discovered that even thunderbolt is limited to 10Gb

[@ermacstudios]

@@Chris-rm1pn Thunderbolt 3 is limited to 20Gb of actual data speed, and because it's essentially encapsulating PCIe you don't end up with the USB conversion. Thunderbolt 3 10GbE adapters run at (roughly) full speed.

[@roppongi765]

@@ermacstudios Thunderbolt 3 bandwidth is 40 Gbit/s, not 20 Gbit/s (that was Thunderbolt 2)

[@ermacstudios]

@@roppongi765 You are correct that the theoretical bandwidth is 40Gb/sec, but because of Intel spec you can only get 22Gb/sec of actual throughput because of protocol overhead and bandwidth that's reserved for the DisplayPort 1.2 channel that runs in parallel. See the benchmarks at egpu.io for more info.

[@Rad_Travels]

Thanks for the review. I had issues, similar to what you reported, with TRENDnet and my MacBook Pro.I got QNAP QNA-UC5G1T instead.

[@PWingert1966]

Tell us about it! Is it working well?

[@uwilhavthat]

wow, I mean wow - i love how cavalier some people are about this video. This is a great in depth vid regarding real problems I am facing right now. My current issue - Jabra Panacast conference room camera. They give you a 5 ft cable. I have to get this camera feed to the conference room table and actually work. I would guess 35 - 45 ft (from camera, through raceway in floor, back up poke through, tie into hub mounted under table, then feed to computer on table. Sounds easy right, ha! Try getting the data needed to support video feed at the correct rate at that distance knowing the camera has a USB-C connection. Not easy unless you are ready to drop $2500 on a USB 3.1 over ethernet extender kit. I greatly appreciate your video and the labor you performed on all your testing. That is a lot of work and I am impressed!

[@AlexForencich]

Well, if you're limited to a 5 Gbps USB connection, you know immediately that you cannot possibly break 4 Gbps due to the 8b/10b encoding used by 5 Gbps USB. And then you have more protocol overhead on top of that.

[@ttww1590]

USB 3.1 is capable of 10 Gbps with Gen 2, so determining the reviewed devices are Gen1 limiting you to Gbps is a key step. When I was looking manufactures and retailers were obscuring this info.

[@bronsoneagle2438]

I use the Sabrent unit for work. No issues beyond not getting the full speed, and heats up after significant use over a long period. Our use case is for ingestion of engineering machines data. For backup or replication. We have also used it when needing to download large model libraries to engineer work stations.

[@Cooper3312000]

I have the QNAP usb-c 3.2 gen1 version around 3 Gbits/sec as well. It's been pretty reliable so far.

[@Mutation666]

Need more 10gbe ones

[@deechvogt1589]

Thank you for the comprehensive testing and thorough review of these 5Gb Ethernet adapters. Stay well!

[@prashanthb6521]

Your channel is a goldmine for info, thanks for all the videos sir !

[@guspaz]

In terms of if the price premium over 2.5 gig adapters is worth it: sometimes you just want to get the fastest networking speed you can on a machine that has no internal expansion or thunderbolt ports, and 3.5 is more than 2.5. On the other hand, if you do have thunderbolt ports, it appears that OWC's 10 gigabit TB3 adapter is down to $149, so that's actually more gigabits per dollar.

[@catminister3327]

This (10Gbe Thunderbolt 3 Adapter) gets extremely hot and i wonder if the rubber casing adds to the heating problem - after it gets too hot it stops working until it cools down, also needs to be plugged in via Ethernet to a primary internet connection (not daisy chained switches) drivers can be a bit tricky to install......not meant for extended consistent use. Heatsinks do very little to dissipate the extreme heat.

[@MarkD26]

Please do the third one and add the results in on a card or in notes. It would be interesting to see any data points that you can gather.

[@The80shilling]

Based on your review, I bought the Sabrent 5Gb USB to Ethernet adapter. I just got AT&T 2Gb fiber (it's actually 2.5Gb/s fiber) at my house for our Internet connection, and wanted to take full advantage of that speed and bandwidth, so built out a 10Gb environment as a mini-backbone. Attached the AT&T gateway/modem to the 10Gb switch via the built-in 5Gb Ethernet connection on the back, then purchased two Engenius EWS377AP Wi-Fi 6 access points that have 2.5Gb Ethernet POE ports, and they also are attached directly to the 10Gb switch. I connect the Sabrent directly to the 10Gb switch as well. I'm getting 1.2Gb/s to my PC, via Wi-Fi 6, through the Engenius APs, and 2.3Gb/s through the Sabrent 5Gb adapter, so even though it can't get much more than 3.5Gb/s in your tests, it completely fulfills it's purpose in my system.

I'm glad I haven't bought one of those (I was tempted to buy the QNAP one) and went just with a 2,5Gbe instead.

[@ElijahPerrin80]

Now that was nerdy, nice. Thank you

[@ArchMnemonic]

Wish I knew about the Sabrent adapter sooner. Might have saved maybe $15-20 per adapter over the Qnap 5gig USB3.1 adapters. Wondered about the not-quite 5gig transfer rates. Makes sense about the usb overhead. Very informative review.

[@jeanvincent4977]

I bought a Trendnet TUC-ET5G and notice the problem but never took the time to investigate, but after seeing the video I contacted TrendNet and created an RMA for it. Let see when the new unit arrive

[@AndyMolenda]

Will you be reviewing any TB3 10G (rj45 or SFP+) units? I'm interested in the Sonnet Solo 10G Thunderbolt 3 to SFP+ 10 GbE adapter

[@ServeTheHomeVideo]

If there is interest in this we can work on 10GbE units as well.

[@juri14111996]

why tb3? i hate tb3, its just a lot of problem, usb is way more user frendly

[@thatLion01]

I am also very interested in 10gbe choices. The thunderbolt ones and also official plug in copper ones. What switch did you test with?

[@AtomicEnd]

I highly recommend the QNA-T310G1S this is the SFP+ version (not the Rj45 one, that's a very different experience). I have had 3 of them running off 3 of my HP elitedesk minis. They don't need a fan, and only run slightly warm at max load. So far running for over 4 months no issues (all 3 are proxmox)

[@jamesosborne6489]

@@AtomicEnd Are you plugging the QNAP adapter into a USB C port or a Thunderbolt port? A quick look at the specs of the Elite Desk mini shows it has USB C gen 3.2 ports, but might also be able to have an additional Thunderbolt port?

[@savagedk]

15:45 because ever so often we need 5Gbps ethernet when backpacking... :p

[@mrmotofy]

Backpacking doesn't have to be remote wilderness. It can be a hiking journalist through Europe where their internet is way better and one needs to upload pics and video of their trip or the raw vid files of wilderness porn

[@savagedk]

@@mrmotofy Let me know of any places in EU a backpacker would visit that has a 5Gbit RJ-45 plug in the wall... Ahhh none? Ok then!

[@mrmotofy]

@@savagedk Europe and Asia tend to have higher and more widespread Internet speeds than the US. Back in 2019 there was the the Manhattan Loft Gardens apartment building in London being offered 10Gb internet for example. Photographers are traveling the world these days.. Data dumps are becoming more and more common. Remember anything over 1Gb requires a higher capacity NIC. Lots of ISP's are alreadt offering 2Gb or 2.5Gb Internet as the next step after 1Gb. Remember backpacking doesn't exclusively mean wilderness travel never returning to society. Backpackers travel all over and are now especially more connected. Being mobile like that can increase the need to do a data dump to NAS somewhere kinda thing. Don't project YOUR lack of need on others.

[@savagedk]

@@mrmotofy so none...

[@savagedk]

ps: im in europe

[@alexatkin]

Wish the Sabrent had been available when I got my Startech. To me it comes down to anything with a built in cable is likely to die eventually where the cable goes into the case, whereas a USB-C port probably wont. Oh and yes getting the 5Gbit adapters is a good thing vs the Realtek, as they stopped supporting new Linux kernels instead telling you to rely on the stock USB Ethernet support which doesn't do Jumbo frames and spams the logs with link/up down messages. I mean it works, but the Realtek driver worked better and just abandoning it leaves a bad taste in the mouth.

[@ZacharyBerger]

Can you review the TB3 10GbE adapters? There are entries from OWC, Sonnet, QNAP, and CalDigit, and I haven't found a comprehensive comparison

[@dksmar]

Happy new Year STH! Love your work and detailed reviews! Please try to review QNAP one

[@truckerallikatuk]

For travelling, don't ignore the fact that the Sabrent has replaceable cables. A cable is FAR easier to replace in a hurry in a strange city than a whole NIC is. So if the cable fails on the Startech, you need to replace the whole thing, vs popping into a mobile phone store and buying a cable.

[@calmeilles]

"This is much longer than that." - the sort of review content we need.

[@joenicklo]

PLEASE add chapters so we can dispense with the pleasantries and get to the content we're here for.

[@robertc574]

Thank you bro I really appreciated the knowledge and understanding 👍

[@typeer]

Thanks guys

[@thatLion01]

After watching your written reviews. I was really hoping you'd a youtube video for all these 5gbe adapters. Thank you Patrick you are amazing. And hope 2021! Will bring even more video reviews. !! Happy new Year!!

[@robertpayne4505]

I could not believe your test results with the Sabrent NT-SS5G. Clearly 3.1 Gen 1 is the problem, but this can be fixed to a certain extent. I have a hybrid 2.5/10 Gbs wired network of TP Link ER-707 M2 router and two NICGIGA 8x10Gbs switches, 4 x570 AMD Ryzen 9 5900 and 5950 desktops (10GBS NIC's), along with 4 Acer Laptops with PCI 4 throughout and USB 3.2 Gen 2. The laptops all have 2.5 Gbs Realtek adapters and 2 M2 2280 slots which carry 4TB each, hence the need for 5 Gbs Ethernet. You need to use jumbo frames at 16 K, set receive and transmit buffers to the max (4k), and switch off everything else that you can, with interrupt moderation set to adaptive. Also, make sure the ethernet/usb line ends at an unmanaged switch and not in a router, or you just add to flow control overheads. I managed to get 4.2/4.25 Gbs laptop to desktop, and 3.9/4.0 GbS laptop to laptop. CPU overhead was around 7.5-8.0%, what you would expect with 12 /16 cores on the desktop and 8 on the laptop. This equates to 80-85% of notional throughput; the old 2.5 Realtek USB ethernet units deliver around 290/292 MBs (93% of notional) and desktop to desktop transfers around 1.15/1.16GBs (92.5%). Why? Well, the problem is that all interrupts and data derived from USB ports need to travel over the PCIE bus, and with NVME drives and GPU's hooked directly into the processor they become third class citizens. 2.5 Gbs Ethernet hooked into USB 3.1Gen 1 will work at rated speeds (there's 50% unused headroom), but while the Sabrent connects at 5Gbs, the PCIE bus then has no headroom at all, certainly not on a laptop. Having USB 3.2 Gen 2 on the laptops does speed things up, allowing the slower 3.1 devices rather more bandwidth. USB devices are not often latency critical, but ethernet traffic is, and suffers as a result. One has only to consider the lamentable performance of some USB 3.1 HDD backup devices for tangible evidence of this. Always use powered units; Seagate Expansion drives (based on DM004 4TB video HDD's) are the best that I've come across. The software overhead for network file copies is around 7-8%; even on high performance networks with no shortage of host resources, the lack of priority on PCIE 4 bus will degrade performance by another 12% at a minimum based on my results; I suspect that you may have tested a few PCIE 3 devices in your surveys, in which case we can say that what should notionally be 92% of maximum throughput has been reduced to 70%. Lastly, if you do buy one of these, then buy the Sabrent. The metal case is a useful heatsink (why I bought the 2.5 Gbs Realtek units), it has two decent sized cables, and it's also the cheapest.

[@Jorge2222]

Great review. Pros/Cons, the Sabrent units seems the best but my only concern is a cable being pulled might undo the connection as it has more failure points if that happens, would still buy it instead of the others for the cable options. It would be a cool dongle to have when troubleshooting servers in the data center for the far off failure of a NIC and a quick way to get data/config off such a server.

[@noxlupi1]

I cannot help but notice, that these test where done out of the box, with Windows 10. Which actually, lately, has a notorious 3.5Gbit's cap, even on 10Gbe. Not sure if these can run on Linux out of the box, but that would be one way of getting around this. Another could be to try and disable Interrupt Moderation in the Network Adapters Advanced settings. Just to make sure, Windows isn't the bottleneck.

[@wildmanjeff42]

Thanks for the info, I actually would have bought the trendnet model, I have had good luck with their products in the past.

[@MarkRose1337]

This was useful. Thank you!

[@ciscowifiguy]

great content.. much appreciated

[@rsporsche]

I like the detachable cable design of the Sabrent. My understanding is the the design of the USB-C connector puts the parts subject to most wear on the plug rather than the socket as the plug is typically part of a replaceable cable and the socket part of a more costly device?

[@AmitChaudhry27]

Love sth always great stuff :)

[@inetkid]

Actual physical speed of 5GbE is 4.688 Gbps minus assume 10% encoding and header loss come to about 4.2 Gbps. Furthermore there is USB lantency. The reason for gradual slowing down is because fragmentation, enable jumbo frame on your network adapter and switch would solve this problem.

[@jeffreyplum5259]

I have another alert for those looking for 10Gig USB thruput All the single lane PCIE cards you find are limited to 5Gig per second speeds. Only the 4 Lane PCIE cards can give full 10Gig/sec thruput. I will be looking at the 2.5 Gig/second units. My use case is for pure server to storage point to point links, bypassing a gigabit switched LAN with many 100Mbit devices on it.

[@davidschadlich4256]

What if you try running the Trendnet on Linux ?

[@slowtrigger]

Awesome content. Have a great year Patrick!

[@UnicronPlanetMode]

Very helpful. Thank you!

[@i0ntempest]

I have a QNAP one and I do want to see a review on it. I use it on a Hackintosh and it sometimes drops connection, but I don't know if it's a problem with the Hackintosh USB patch or the adapter itself.

[@danielvas8909]

I have a pair of those and performance is excellent, but every time i reboot my nas i have to unplug re-plug it from the nas to make it work again, once it works it is reliable for months or whenever you need to reboot the system

[@squelchedotter]

Considering this seems to be driver related, I'd be curious how this works out on macos and linux. I wonder if anyone has tried that?

[@stonent]

Also they didn't go into detail about using the other drivers. You can right click and force the wrong driver on things which sometimes works on these hacked products where they put their own name on the drivers.

[@excitedbox5705]

Those metal cases cost $1-2 and there is no reason to buy the more expensive one. The one company is obviously trying to deliver a product at a decent price while the other is trying to maximize profits based on what someone might be willing to pay. The plastic case is a cost cutting measure to squeeze every bit of profit out of a product instead of making a tiny bit less money while delivering a MUCH more durable product. In most cases given the choice between paying $59 or $60 or even $61 as a customer I would be willing to absorb the cost of the more expensive case. I wish more companies understood that instead of these cost cutting measures. You see companies doing that all the time where they decrease quality in order to save $0.10-$5 where the customer would be happy to pay the extra cost. Instead they often you see them price gouge that extra with a 5-10x mark up. Paying $1-2 more for a metal case is worth it but $10+ is NOT.

[@excternal]

I have many 2.5Gb USB Ethernet adapter. theres have device down(gone and came back shortly in time) problem on commnication is busy. i diagnosed that caused by overheating because that problem occurs pattern. and i disasembled for installation heatsink on chip. it works for long lasting on busy to network. but is did not perfectly solution. chip is so much tiny and too much hot as i think. this heatsink is so heated on idle with unsetted for energy efficiency options and more heat up on situation of NIC is busy. in feeling temperature by finger contact. and another aluminum cased 2.5Gbe NIC is have same problem. but i am no longer using that becase didn't to found way for disasemble clearly with no damage. this problem is may be occurable cheaper 5Gbe USB NICs more.

[@thetj8243]

Thanks for this interesting test. I hope there will be some USB 3.1 gen 2 or higher options soon with 10gbit USB you should be able to get 5 gbit out of the ethernet side...

[@benjamintrathen6119]

Cheers for the gen 4 optane info on STH site, very cool.

[@lalalu7803]

Can you please add in the review for qnap too. Thanks

[@ServeTheHomeVideo]

Ordered two QNAP NICs so we could repeat the process for them as well. These will likely just be on the STH main site in a review not on YT.

[@scottcorley]

Curious if any of those adapters, particularly the Sabrent works well with pfSense. If you could use the built in NIC + an adapter on a micro form factor system that'd be sweet!

[@umarjamil8512]

Great Content. Why not try it on a powered usb hub. It maybe the trendnet unit want more power than usb is capable of.

[@ServeTheHomeVideo]

We did. We even tried higher-power USB 3.2 Gen2 charging capable ports. The same thing happened.

[@spiceyfrenchtoast9421]

If you were backpacking why would u need a 5gbps nic?

[@studioviper3016]

Wow what a great review. First time watching this channel and im a fan.

[@GreyGoshawk]

With Thunderbolt being 40gbps, I wonder if a Thunderbolt 10GBe adaptor will give the full speed.

[@JeffGeerling]

I've been testing the OWC TB3 adaptor, and it does get 10G speeds (over a 40 Gbps Thunderbolt 3 port).

[@tombu4484]

I have been using the a pair of the QNAP 5GbE unit on my MicroServer Gen10 Plus and EC200a. It had no problem with either Windows or Proxmox VE, but the QNAP unit will cause Arch Linux to hang on boot if it was plugged in. I have not managed to pinpoint what's going on so I moved one of the QNAP NICs that was for my laptop to my EC200a (to speed up rsync backups between the two servers). Maybe I should grab a pair of the Sabrnet cards because the QNAP cards also use a customized firmware/driver stack, and maybe that's why it was hanging Arch Linux. I've had bad luck with Realtek 2.5GbE adapters, they constantly drop connections (very severe on Proxmox, better on Arch and Windows), and the Sabrnet one was not available back when I built my MSG10+, and the QNAP one was the best bang for the buck back then.

[@esra_erimez]

I would be very interested in seeing a video about USB-eth module in Linux (TCP over USB)

[@ChristianSamsel]

I have an 5gbit Aquantia NIC on my x299 Deluxe Prime II. While the performance is fine, at high incoming throughput it crashes. I invested weeks without success (drivers, firmware, cooling, config parameters ...). Got an Intel x550-t2 instead.

[@henfibr]

Are there are any other reasons besides cost to choose a Realtek-based 2.5Gbps adapter instead? For instance, what about latency, jitter, stability, driver compatibility (Linux, Freebsd, VMware), switch compatibility, CPU utilization, thermals, power consumption, TCP offloading/queues ?

[@mikebroom1866]

Comparison between these and the cheap realtek 2.5GBE adapters??? I'm running one on my xpenology system at home and have gotten some good results. Drivers on anything but windows are atrocious though.

[@kilrahvp]

I have a pair of the QNAPs, they're basically identical to the Sabrent, although a bit smaller apparently. Metal body, uses standard drivers, USB-C port with 2 (shorter) cables supplied and work fine. Was aware of the limitations but there's just nothing better since there are no Gen2 chips yet and TB3 10Gs are just too huge and expensive.

[@cyberlando]

Can you please do one for 10GBE base T as well!

[@alexatkin]

Pretty sure there is no 10Gbe NIC on USB as it can't output enough current, they are available for Thunderbolt but as that's effectively PCIe it should perform as expected.

[@mrmotofy]

@@alexatkin USB3.1, which obviously wouldn't be full 10Gb for the same reasons but could probably get 7.5Gb or so

[@zaur66]

Really well done review, thank you

[@Rodney.Thomson]

When you saw the Trendnet unit 'drop' were you still able to ping between units? We've had reliability issues with USB3 to 5GBE adapters (we use the QNAP UC5G1T on laptops to allow for high rate data export from servers) where after some time the unit simply stops receiving data (well, effective receive rate drops to < 100k/s) but the upload is not affected. You could still ping between systems. I initially suspected the issue was with the Netgear switch, but with further testing I believe the issue is due to the QNAP USB 5GBE. More info and full use case here: community.netgear.com/t5/Smart-Plus-and-Smart-Pro-Managed/XS512EM-with-Aquantia-10Gbe-stops-uploads/td-p/1925484

[@ServeTheHomeVideo]

No. Link was down when the units dropped.

[@imark7777777]

Not only can that last one we both what are the chances the cable on the other ones is gonna get broken and then render the whole unit useless with a detachable cable it could be replaced or extended. And also serves as a breakaway connection.

[@gorgonbert]

have you tried MTU 9000 (jumbo frames) for better performance?

[@brentboswell1294]

False hope. There's very limited instances where a higher MTU increases performance...been a systems/network admin for 20 years.

[@gorgonbert]

@@brentboswell1294 It CAN have massive impact. For example running ESX datastores over NFS, the CPU utilization came way down on the storage array when we switched to Jumbo Frames.

[@kilrahvp]

No point, again it's USB that is the limit.

[@brentboswell1294]

@@gorgonbert I guess I'm just used to working with network hardware that has good TCP/IP offload capabilities (Intel 710 series and Mellanox). Stack processing and data segmentation is handled by the Ethernet controller 🤔

[@gorgonbert]

@@brentboswell1294 Different strategies... my thinking is to try to avoid generating unnecessary load that then needs to be off-loaded by some intricate piece of hard- and software... your strategy is to keep it easier and more flexible because it doesn’t need to be specifically tuned... it depends on what you‘re looking for...

[@samiraperi467]

So... how well does the Sabrent work in Linux? I have an Aquantia 5G PCIe card that chokes when I try to upload anything large to my server (confirmed with iperf, it's not just samba or whatever I happen to be using then). It's a commonly reported problem and is probably an inter-frame gap issue but I haven't figured how to fiddle with that, so I'm wondering if these behave differently.

[@daslolo]

This is review cranked up way past Beast mode! Subscribed. Any 10Gbe usbc-gen 2 adapters that don't cost 500$?

[@niyaziugur]

Jumbo Ethernet packets(9000 MTU) can increase the throughput.

[@OVERKILL_PINBALL]

Awesome video.. Now I have to go check my ASUS 10GB adapter to see if it using the same chipset. EDIT: It has an AQC107.... and I am curious if this is also limited. Time for some testing.

[@CDArena]

Haven't watched past 5 minutes yet, but I can't help but notice you haven't (yet?) talked about the software side of things - what OS? What drivers? Windows OS does have high speed limitations on file transfer, which are known by high speed enthusiasts

[@ServeTheHomeVideo]

Driver discussion will happen later in the video. Just for reference, 5GbE speeds are very low for modern systems. 100GbE/ 200GbE (we do have a ConnectX-6 200GbE link up now between Windows systems) are more challenging. 1-10GbE is not exactly considered high-end these days.

[@fabiocolaiacomo5411]

maybe the limit of just over 3 Gb/s comes from the MTU at 1500 if you try to set an MTU of 9000 on all devices in the test network (NIC, Switch, etc.) you can reach 5 Gb/s

[@ServeTheHomeVideo]

We tried 9K Jumbo and it did not significantly change the results. Take a look at the full TRENDnet unit review linked in the description for an example.

[@gorgonbert]

@@ServeTheHomeVideo you are certainly aware that all devices need to be set to MTU 9000 for this to work... host1, switch and host2... right? The reason why I ask is that I firmly believe you should be seeing a difference (I’m pretty much 100% certain)... if you don’t see any difference in performance it is highly likely something wasn’t working right... might be that the driver ignores the setting... I’ve had issues with some Dell branded Broadcom 10G NICs that had difficulty setting the proper MTU size (that was a while back though)

[@gorgonbert]

@@ServeTheHomeVideo On second thought... i undervalued the fact that we’re talking about USB... so scratch my earlier post... you’re probably right with your hypothesis... USB sucks... 🤷‍♂️

[@AlexForencich]

"5 Gbps" USB is limited to 4 Gbps due to encoding overhead, and then there are protocol overheads on top of that that eat up more bandwidth. 3.5 Gbps is probably about the best you can expect, irrespective of MTU settings.

[@benjamintrathen6119]

God I am such a nerd also, I have been thinking about this topic for some time.

[@hikaru-live]

Somthing you can try: take the Trendnet one apart to reflash its configuration memory to something stock original, for example the configuration from another onein this review, and re-benchmark?

[@ServeTheHomeVideo]

We had talked about it. Given there are less expensive models out there, we felt the expectation should be it works as-is without those extra steps

[@hikaru-live]

@@ServeTheHomeVideo The intention of that test is to check whether it is the hardware implement being at the fault, or the driver optimizations. That firmware modification forces the Trendnet NIC to not use their own driver.

[@j.d.3269]

Maybe TRENDnet power save is on?

[@robertbash380]

I wish I had seen this video before I bought the Trendnet. It stopped recognizing USB 3.0 and ran only at USB 2.0 speeds after a couple of days. Returned that piece of junk.

[@BenKlassen1]

Great information. Great effort.

[@shephusted2714]

there will be 10g via usb4 on next rasp pi q3 cause why wouldn't there be - they need more pci lanes but with fast network and nvme the sbc segment could become a force - they may just want to pop in one more tier to the product line #pci-e v4x4 #dual usb4 comtrollers - if they can get that with new arm arch - they will get what they can but somebody in the sbc mkt will figure it out eventually - it won't threaten gaming cpu but will make clustering much more better - and who doesn't want a 32gb 400 buck cluster with fast network and native nvme - we aren't there yet but it will be a massive jump in performance #link agg for the net and 4 nvme in a card on the pci slot is the way of the future - 20gbe and 32tb raid0 nvme cluster node

[@pedro_8240]

This "it shows as 5Gbps but doesn't transfer at such speeds" is kinda obvious, and not only expected but also fine, considering the overheads of USB. It's why I understand and don't understand why you won't find USB 2.0 1Gbps adapters, only 100Mbps. While USB 2.0 won't saturate a 1Gbps link, it can transfer at at least twice that, and sometimes I would link to have some cheap "1"Gbps USB 2.0 adapters for devices that lack better connectivity.

[@microcolonel]

The Sabrent packaging/design I much prefer.

[@kylehazachode]

Test the qnap please

[@robertc574]

Please, Test the QNAP Thank you

[@miguelagueda3928]

If there's no extra changes on the trendnet unit, you might be able to manually install the standard Marvel driver and make it work fine. I wonder if it's posible, at least you have an alternative if you already own it or don't have the other models available

[@wskinnyodden]

There goes data link's communications protocols speed overhead going over everybody's head.... hehehehe

[@watcher333666]

Powerful start, followed by performance drop after device gets warmer?

[@LoneRiderz]

I have an ITX build and this may be the only way for me to gain higher network speed.

[@律師]

Try UGreen 5GBE-USB, they are lot cheaper (CM312) at $70 usd. And they are metal chassis.

[@TimBoundy]

If I could get a cheap 5Gbit switch, then I might consider the 5Gbit adapters. For now, I'll stick with 2.5Gbit on my QSW-1105-5T.

[@alpenfoxvideo7255]

Given the choice, would you buy a PCI-e internal 10Gbe NIC or a Thunderbolt 10Gbe?

[@minigpracing3068]

Would these have ranked higher if they locked them to 2.5gbe? Does the Sabrent have a thermal connection between the chips and the case to help with thermal problems? I ran into this a lot when trying to use a usb-gigabit adapter for pfsense, if I clamped then to a cool rack shelf, they would work longer between drop outs. Gave up on USB adapters for firewalls, but it would be REALLY handy to have a good solid device that didn't drop out for quick or cheap builds.

[@nUKmAll]

Please test the Qnap USB 5Gb adapter

[@johngonzalez3502]

some of those adapters are only usb 3.0 like the Sabrent. Did you also test those using a usb 3.1 gen 2 usb port?

[@ServeTheHomeVideo]

Yes. They use the same chip and so while we tried, Gen2 ports did not help