TRENDnet TL2-F7080 Performance
This is going to look a bit strange to folks because it is going live out-of-order. For this, we are using a Keysight XGS2 chassis with the NOVUS10/1GE16DP card, and we are using eight of the sixteen dual PHY ports of the card for this. We are using an RFC2544 Quick Test for throughput with a few changes, for example, we extended the run times per iteration to 30 seconds and started at 10% of the full line rate, searching for frame loss up to 100% of the line rate. We also test the RFC2544 64, 128, 256, 512, 1024, 1280, and 1518 byte frame sizes, and then add the standard IMIX, the Cisco IMIX, and an IPSec IMIX, which introduces mixed frame sizes into the testing. What is slightly different here is that we also turned on the L1 rate counters since we are using a hardware-based custom testing solution, we have access to this without having to go far into software counters. You may see the L1 rates regress from a few future switches as there are 2-3 more that were supposed to build up to show this.
Why this matters is that we can now generate up to just over 1.7Tbps of network traffic and get awesome statistics. For example, we can use 64B frame sizes, as shown above, and we can see that we only get just under 61Gbps. That is because there is additional traffic happening at the L1 layer, and so the actual L1 throughput is 80Gbps, which matches what we would expect.
Just to see this in action, you can see the same 80Gbps L1 rate, but with 1518B frame sizes, the overheads are lower, so we get almost 79Gbps. At the same time, you can also see the latency jump.
The bar charts above start at 10% of the maximum theoretical line rate and then search for where frame loss happens as the links are scaled to 100% of the theoretical line rate. The above is testing a mix of frame sizes as part of an IPSec IMIX profile. We also get latency for the different frame sizes.
Here is a quick look at the loss and latency by the frame size or IMIX profile.
This is perhaps the first switch review where we can show that, even using high-end industry-standard test and measurement gear and IxNetwork, we are getting full line rate performance on these switches. A lot of folks have asked about whether something like the Realtek RTL9303 can actually handle high-speeds. This is a well-known type of test (albeit we did customize it a bit for our use), and using high-end gear where we can collect this low-level data. Somewhat to our surprise, even these sub $200 switches can now do 80Gbps of traffic across them.
TRENDnet TL2-F7080 Power Consumption and Noise
On the power consumption side, at idle, we were at 5.6W.
With a single SFP+ to 10Gbase-T module we were up to 7.0W.
Overall, the power consumption was about what we might expect at this level of switch. Some may say that it is higher than the 2.5GbE switches. It is, but that should also be expected since this offers more throughput. Just to give you some sense, our XGS2 chassis, which also had the 8x 100G NOVUS modules installed, was running at around 900W to test this ~17-22W switch.
When it comes to noise, this switch is far from silent. We are talking 42-44dba in normal operation when it is at idle. Frankly, this is too loud for a switch of this class. TRENDnet dropped the ball on the noise profile.
Note from Patrick: I wanted to double-check the noise and also update the firmware for the management section today. It is sitting on a desk with an AMD Threadripper system, three NVIDIA GB10 systems, the Minisforum MS-S1 Max and MS-02 Ultra, and both the QNAP QSW-M7308R-4X 8x 25GbE and 4x 100GbE Managed Switch, and QNAP QSW-M3216R-8S8T 16-port 10GbE Managed Switch. At 1 meter, all I can hear is this switch, and it is driving me nuts. I should not hear an 80Gbps switch over a 600Gbps switch. There is no reason for a device drawing 7.0W next to me to be this loud.
Final Words
All said, where are we on this one? To me, the biggest learning was that TRENDnet is a brand that has been around for some time, and a sub $200 8-port 10GbE switch is great. Some will prefer 10Gbase-T to SFP+, or perhaps a mix of the two. Just wait a bit, and we will have those reviews coming.
Still, the one item that really got to us was the fan. This should not need that much airflow, and a 42dba+ switch seems excessive when the device is operating at 7W. Some may think about a fan swap, but the other side is that this should be something that TRENDnet could just provide a better solution for. We review many of these switches, and the noise from this one would have us picking a different option if we wanted this next to a desk.
Overall, though, it worked well, and we were able to demonstrate, using high-end test equipment, that the Realtek RTL9303 solution is handling a lot of traffic without issue. That is a great finding for this TRENDnet.
If you were reading this page thinking, “What would happen if we looked at a Cisco Catalyst series switch? A Ubiquiti or MikroTik switch? Or others?” All I can say is stay tuned to STH. We are going to be filling out our The Ultimate Cheap 10GbE Switch Buyers Guide in 2026.
Where to Buy
Here is an Amazon Affiliate Link to what we purchased.
My understanding is that “web managed” means no console and no CLI. This switch apparently has both. So calling it that undersells it significantly.
Seems expensive once you factor in the cost of transceivers? I’d rather buy a 10GBASET switch with a couple SFP+ ports for uplinks. I’m looking forward to your reviews of other options.
I’m in awe at STH’s testing.
If you don’t know, this isn’t something you can just eBay and run for $10K. They’re doing well into $100k+ testing. Ixia uses FPGA based load gen hardware, so it’s many FPGA on a big card in a custom box, custom software. I’m wanting to see the 100G used but it’d be better if they could do 800Gb/s on it
one nice feature is having standard 4-pin connection for cooling fan. easly replaceble with quieter fan.
and if you want to keep sfp+ modules also cool, then you can install/put 2 extra small fans on top of sfp+ sockets.
I can’t understand why people would choose these over mikrotiks SFP switches? They are reputable, solid, and great p/p!
@Bill: I think there’s a major, major difference depending on which side you are approaching 10GbE from.
If you are doing the ebay-and-server-pulls route it’s basically all SPF+; probably-fine DACs are honestly more likely to cost money because of shipping than because they are a hot commodity; and anything but the more exotic and capable optical stuff is also pretty readily accessible; but old top-of-rack switches are relatively thirsty, relatively noisy and you probably don’t need that many ports. For that case all SFP or mostly SFP switches based on contemporary, low power, switch chips that let you go beyond point-to-point links or ‘flat’ unmanaged-only setups are fantastic.
If you are coming at it from the ‘new’ side; and mostly 10GBASET devices, maybe an optical run between wherever your ISP decided that the demarc point should be and where you actually want your switch; filling an SFP cage with copper transceivers is probably the worst available option. They are pretty expensive relative to the equivalent signaling witchcraft built into 10GBASET ports; and they tend to run hot enough that not all SFP cages can take it; if they were expecting optical and DACs. Always good to have a couple in the toolbox; just because you will encounter big fat all-SFP switches that need to have one UPS monitoring card or your laptop or whatever connected to them; but that’s mostly a bodge.
There’s probably a very narrow niche where they are the least-worst option: at one point 10GbE was Very Serious Stuff; so there are SFP switches floating around(though mostly loud and power hungry) built with all the bells and whistles and cost-barely-an-object niceties back when 10Gb was spendy enterprise kit; and it is unlikely that anyone will build the likes of those in 10GBASET, since that’s fundamentally a lower-end tech at this point; but mostly you want to choose the switch that matches your target devices.
Some kind of fan shroud to force the moving air through the heatsink and out the SFP+ ports may have allowed the fan to spin slower, reducing noise.
I initially thought this may be an interesting switch to put by my Netgate 7100 firewall/router, but as this is standing in my livingroom, when I saw the first notes about the noise level, and then especially Patrick’s note about the noise level, this switch isn’t an option for me.
I am a bit sensitive to noise @Dev_Mgr but this is one where I would have to do a fan swap or something. Stay tuned to STH since I think we have a decent option (targeting like 2 weeks until it is live since we did a video.)
On the 10Gbase-T vs. SFP+, there are two other notable wrinkles. First, the ability to breakout from QSFP+/QSFP28 ports is useful in some cases. Second, 10Gbase-T can be useful to deploy in servers because you also get easy 1GbE. If you look even at NVIDIA GB300 NVL72 racks, there are a ton of base-T ports used for management and monitoring. So 10Gbase-T allows you to have a faster interface, but also not have to do any conversion to go into existing 1Gbase-T switches.
Also, on the 10Gbase-T side, we are going to get more into the series tomorrow in a video.
@Ivo – I guess the reason we are doing this is to test that assumption, and to get more useful data than just running an iperf3 test. We are also going to the other end and gathering data for folks asking, “Why buy MikroTik (or Ubiquiti) instead of Cisco?”
Good to see. We need passive 8x 25gb/sfp28 next, hopefully from Mikrotik. Its ten years since they made the 10gb SOHO/condo targeted switches, and 8x is next step.
At work, 1G/10G is purely for mgmt these days. Very very rarely will we provision new 10G long haul at this point. So I view this product as a SOHO option, and in that light, that’s where I don’t understand the SFP+ cages.
Using surplus equipment from eBay makes sense though, I hadn’t thought of that!
This new testing is very welcome and definitely interesting. It’s nice to have a confirmation that the RTL9303 performs well, seeing as it’s used in both cheap AliExpress switches as well as a bit more expensive ones like the USW-Aggregation.
A comparison result from one of the AliExpress models would be interesting as those cost roughly half as much from brands like Horaco, XikeStor or Goodtop when talking L3 managed units. I would imagine that things such as the power supply and certain PCB components are worse in the cheaper models.
This is strange: in the photo captioned “TRENDnet TL2 F7080 Inside 7,” it looks to me like the manufacturer soldered only two of the six pins on the mainboard’s connector to the DC/power switch daughterboard.
Is that right? If so, that strongly implies that four of the pins are unused. But then all six pins are soldered on the daughterboard side. Curiouser and curiouser…
Does anyone have a decent guess as to why that might be?