TRENDnet 10G PCIe Network Adapter, TEG-10GECTX, Convert a PCI Express Slot Into a 10 Gigabit Ethernet Port, 802.1Q VLAN Tagging, Windows and Windows Server Support

What can I say, it works as intended. I've got 2 of these: one is in an Ubuntu 22.04 server (an old 4770 i7 machine) and the second and Arch Desktop (a ryzen 3900x). Both recognized the cards immediately and set their speeds appropriately with auto-negotiate enabled. Both machines are connected through a 10g copper port switch connected to an OPNsense router (1520 XeonD) with a 10g port.People trying to speed test these with file transfers are going about it the wrong way ad will never see the total throughput because of drive read/write speeds and protocol limitations of the chosen share method (i.e. SMB).The easiest and truest method to test is to run iperf3 as a server on a machine with a 10g interface, on the second machine run an iperf3 test as a client to your machine. BUT that will not really give you the actual result you want, as standard iperf3 tests are run as a single process and will probably cap out around 2-4 gigs. You need to run the client with the -P option to enable more processes, I typically use 10.Example:on the server end:iperf3 -son the client end:iperf3 -c (ip of the server) -P 10iperf3 -c (ip of the server) -P 10 -RThat gives me the output that I would expect to see, typically in the 8.4 - 9.4 gig range. Also be sure to run it in reverse as well with the -R option the second time.The cards work exactly as intended for 10g, though admittedly I have not tried 2.5g or just standard 1g because I haven't the need to. And 5 gig is incredibly uncommon to find so I won't bother with it.I've attached 3 screenshots to show you the results of the 3 iperf3 tests I mentioned. The first pic is if you run iperf3 -c (ip of server) and caps out at 3.73gsecond is iperf3 -c (ip of server) -P 10 and caps at 9.34gthird is iperf3 -c (ip of server) -P 10 -R and caps at 9.37gBTW I should note that I'm not actually using Cat6a for any of these machines. I'm actually using standard cat5e, but the lengths are less than 10 feet. If I had to go longer or started to notice issues, I'd up it to cat6a but it's not always necessary. Just an FYI

So at the time of my purchase, this card was a little over a bill. There were other options from Asus, ALFA, TP-Link & Rosewill just to name a few, all cheaper than the TrendNet card (Rosewill being the cheapest) and some of these may even be using the same Marvell AQTion controller, but for me, I just trust TREDNet a little better and this is the 3rd revision of the NIC so I'm hoping these guys worked out some bugs.Won't really try to go into too much technical details here, I just needed a nic that would get me a 10gbe.Cabling: I am using my existing Cat6. The runs are much shorter than the 50meter limitation of cat6's 10gbe spec. (Cat6a or higher can get you runs of 100Meters, though when using the SFP+ port, the run is limited to only 30Meters regardless of cable used).Switch: Zyxel XGS1250-12, Firmware v1.00. (B08TC3VJ61) This is a new switch that was released, I am still working on my review, in short, it rocks! I have tested using this NIC into one of the existing copper ports on the switch as well as using a Copper SFP+ transceiver (B075WMGBHX). I have had no issues establishing and maintaining a 10gbe connection.Installation: This is my only concerning point of this review, when I installed the NIC, I downloaded the "Latest" drivers from Trendnet's website, which is listed as version v1.00 with a date of 8/31/2020. However shortly after installing said driver and running Windows 10 updates (oh yeah, I'm running Windows 10 20H2), the driver was updated to version 2.2.2.0 with a date of 04/30/2020... If you go to Marvell's website, you can download their "latest" driver, version 2.1.21 with a date of 04/26/21 (The day I am writing this review is 04/25/21)I am unable to tell at this time if there is a higher version than 2.2.2.0. Regardless, its workingFinal Thoughts: So I have 2 machines, both running Windows 10 x64, where I installed these cards. (GA-z87x-ud7 TH & x99 Taichi). I had plenty of room in each system and I made sure there was good air flow. I see lots of concerns regarding anything 10gbe with how hot things can get. I'm attaching a screenshot from a simulated test I did between both machines. I found out shortly after upgrading my nics, my new bottle neck is the hdd array in my data server. I went from maxing out at 112MB/s (1gbe) to around 377MB/s (10gbe), though triple what I was getting, it wasn't quite 10x faster ;) So I enabled a RAM disk on both devices thus eliminating the drives array specs (ram disk gets around 23,552MB/s) and I would start seeing me hit the 10gbe max (around 900MB/s). My main work flow is photography, I'm now able to start moving my data to my storage server (which has multiple forms of data protection) but work on these files as if they were stored on one of my local drives on my main editing rig.Just verify that your system meets the min specs for this card, I am sure you will be satisfied.

I have a 2.5gb switch. Originally bought the TP Link TX401. It did not work and was stuck at 100/100 MBps. Went through troubleshooting and no luck. This TRENDnet worked at 2.5 GB out of the box. Windows 10 found the driver right away. Definitely a good product.

I'm using these Marvell AQC107 based cards in two Linux machines on my home network. Network throughput initially varied widely during large file transfers between these two machines. While troubleshooting, I learned that both cards were shipped with out-of-date/buggy firmware (v3.1.58) vice the latest/greatest firmware currently available on Marvell's website (v3.1.121a). I pulled the cards from my Linux machines, installed them in a Windows machine, updated the firmware via the included utility, and then reinstalled them in my Linux machines. Network throughput is now rock solid stable and only limited by the bandwidth of my storage devices.