Ubiquiti UniFi USW-Pro-XG-8-PoE Performance
To test this, we are using a single Keysight XGS2 with a NOVUS 1/10G dual PHY card. We also upgraded our IxNetwork 11 revision to IxNetwork 26. One of the advantages of using a high-end FPGA-based tester is that we can generate not just 64B packets at line rate and over 1.6Tbps on our setup, but we can also get solid metrics on latency and jitter. We are using the RFC2544 Quick Test template and adding extra time per step, along with some additional IMIX variants, to provide more data. Starting with the 64B line rate test, here is what we saw:
As you can see, we are hitting 100% line rate L1 traffic of 100Gbps (10 ports at 10Gbps) with no frame loss. That is only around 76Gbps of 64B packets since there is overhead. The scaling starts at 10% of the theoretical line rate, then increases to find how close the device can get to that theoretical rate without dropping packets. When you see the bars all the way to the right in green, on the 9th iteration, that means we got up to 100% line rate without dropping packets.
Something that caught our eye here was that the minimum latency and maximum latencies were close to the Sodola 12-port 10G switch that we just reviewed. What is different is that the maximum Jitter we are getting is much higher at 322ns in this 64B packet size test, whereas the Sodola, across 12 ports on the same Realtek switch chip, the maximum was 70ns.
Moving to the 1518B results, here is what we saw:
That followed a similar trend, with the Sodola having lower latency (both minimum and maximum) and lower maximum jitter at 1518B across 12 ports. It seems slightly odd that we are seeing better numbers on the generic switch with the same switch chip inside, but this is what we observed.
Aside from uniform packet sizes in 64B, 128B, 256B, 512B, 1024B, 1280B, and 1518B sizes, we are also using three standard IMIX profiles where the traffic generator is using different-sized packets in a mix. First, we have the standard IMIX profile:
Here you can see that the minimum latency and jitter look more like the 64B results, whereas the maximum latency looks closer to the 1518B result. Just to be clear, this is very normal since we are now using a mix of packet sizes. Even the Cisco Catalyst C1300-8FP-2G we just reviewed exhibits similar behavior.
On the subject of Cisco, they have their own IMIX profile, so here is the Cisco IMIX profile:
For those who plan to run more VPN traffic through their devices, we also have the IPSec IMIX profile running.
There are really two key takeaways. First, 10 ports of 10Gbps we just independently validated. You can see these are running with the same millions or billions of frames being sent and received with zero loss. That is exactly what we would want. We were a bit surprised to see the higher latency and jitter numbers. We are unsure why a 10-port Realtek RTL9313 switch would have more latency and jitter than a 12-port switch, other than if it had something to do with the different mix of 10Gbase-T ports versus SFP+ ports (8 PoE+2 versus 4+8) or if it came down to something in software. Still, it does not seem like most will notice the difference. It is just something we wanted to show since we found it and can now measure it.
Ubiquiti UniFi USW-Pro-XG-8-PoE Power Consumption and Noise
Since this is a PoE++ switch with an external power brick, we have a Ubiquiti-branded 54V 210W power supply.
At idle, we were getting around 19.6W. This is higher than some other non-PoE switches we have tested, but PoE switches tend to idle higher.
Plugging in one port, we get 21.3W.
Something notable was that when we plugged a SFP+ to 10Gbase-T adapter in, we only got 21.0W. Usually we see higher power with the adapter than we do on native ports so this was surprising.
Of course, the majority of power is dedicated to PoE++ downstream devices.
In terms of noise, after boot up, even while running the 100Gbps L1 line rate traffic we were at around 37dba in a 34dba noise floor studio. Just to give you some sense, when we only have one of the Keysight XGS2 chassis online, even at idle it is around 68dba at 1m.
Ubiquiti UniFi USW-Pro-XG-8-PoE PoE++ Testing
We hooked up both the Fluke LinkIQ-Duo (Amazon Affiliate) and the MicroScanner PoE (Amazon Affiliate.) Here you can see on the LinkIQ-Duo the 2.5GbE and 5GbE speeds being validated, as well as both instruments showing 51W PoE++. That is different from the Ubiquiti UCG-Fiber where the MicroScanner PoE did not detect PoE.
It may be hard to see, but the Yellow cable to the MicroScanner PoE is being lit by the switch at the switch connector. That is always neat to see.
Final Words
At $499, it is certainly one of the pricier 10-port 10G switches on the market. At the same time, there are customers who want all Ubiquiti UniFi features, including unified management, lit power ports, and, of course, PoE++ capabilities. The maximum latency and jitter are higher than those of the 12-port, low-cost managed switch we just tested, but most applications and buyers in this class will not notice the difference. It was just something neat to see using higher-end tooling.
For a lot of folks, this will be a perfect 10GbE switch. That is especially true if you want to be in the UniFi ecosystem and have relatively modest port requirements, at least for 10GBase-T. To be honest, this is one where we wish that there were more 8-port 10Gbase-T switches with both PoE++ and SFP+ uplinks. Ubiquiti did a great job on this one. We have tested many other Ubiquiti products, and this is a clear standout.
Where to Buy
If you wish to purchase one, here is a B&H affiliate link.
 and test it using a high-end traffic generator){kind=link}
Would be really nice if the L3 performance could be tested as Ubiquiti advertises with this.
The world waits while STH conjures up it’s next Ubiquiti review. Your network reviews don’t have any peers now.
That’s a really nice switch for desktop! There isn’t a lot of market for 8 port tx + sfp+ 10g switches, even the chinese switches are hard to come by like that. I was so disappointed by qnap I’ll never buy theirs again, but several network engineer friends of mine use Ubiquiti so it can’t be too bad.
I think the switch is a good deal in terms of cost per 10G port. The CRS304 has 4x 10G, and the cost per 10G is around $41~43. This got 10x 10G ports, and the cost per 10G is $49.9, slightly more than the CRS, but it gives you PoE availability.
The main thing about this switch is that it fits in a 10-inch rack. There aren’t many switches out there that have 10GBASE-T + SFP+ + POE.
I’m looking for a 10g switch for my 10-inch rack, so that’s why I’m looking at this one.
If anyone has a better suggestion, please let me know.
Yay you’re doing the good work. Best testing. Ubiquiti should have you review all they’ve got. I’m just waiting until you review from now on
Having an avg and max jitter are really not enough to tell you, for example, if your VoIP is going to annoy users. The average is 3ms, ok, great, but does that mean:
– most packets 2, some packets 20? or
– most packets 2, occasional packets 70? or
– something else?
The former is probably fine for VoIP, the latter will cause sound artifacts.
So when you say you saw unusual jitter on this box, that would be the first question I’d want answered: Does VoIP work well? Even under heavy load?
But more generally, you should be able to characterize jitter in a more interesting way, and I’d be surprised if your fancy new testing box doesn’t have software to do that already. Show a jitter CDF graph, or at least tell us the %age that comes in with latency <40ms, <50ms, <60ms since dejitter buffers are usually in that range. Alternatively you could how a jitter histogram.
BTW, "max jitter" is mostly useless info. 95th percentile is far more interesting.
Grr. “…show a jitter histogram”. Sorry.
@justsomeguy, the average jitter seems to be 3ns, not 3ms. So like 6 orders of magnitude lower. The max is <1.2us (microsecond). I don't think any human can discern such variations.
Wow. Yes, obviously, ms jitter in a 10G switch would be ridiculous. Clearly I need to focus a little better. Sorry.
…though the jitter info would still be a little interesting.
What is the black protection around SFP+ ports ? It-is in plastic or in metal ? If it in metal it can help a little bit to dissipate heat
I have this model sitting on my desk at home, powering our main floor AP and providing connections for a pair of desktops (yes, they’re on 10Gbps connections because I’m silly) and a printer, and I’ve been quite happy with it. I will note one minor thing; it does not seem to support 10Mbps connections. The old alarm system in my home uses a 10Mbps connection, and would not establish a link with this switch. It probably won’t affect anyone looking to support multiple multi-gig connections, but it caught me off guard for a few moments.
@ Michael Butash. “I was so disappointed by qnap I’ll never buy theirs again”
What was your disappointment with the qnap switch you mentioned. Which model were you referring to and what was your problems ?
I just wish this little switch had the front panel LCD like some other Ubiquiti gear. Surprisingly handy.
155W of PoE++ is not much at all given that PoE++ can go all the way up to 90W per port but this switch tops out at 60W per port. That equates to two ports at the power level with a bit left over for another 30W or two 15W pieces of gear. Many of the higher end Wi-fi 7 access points are going toward 60W. Then there is the trend of supporting PSE pass through where 30W or 15W PoE devices can start to be daisy chained off of a 90W or 60W capable switch port. (Daisy chain PoE speakers are awesome.) Speaking of power, being able to support a second power supply would be nice.
The variable latency here could be explained by the PoE addition on the ports. There is logic here to account for power delivery that SFP+ ports don’t need to do. What would be interesting is if this switch also supports 802.1Qav or 802.1Qbv for deterministic networking. That amount of variable latency would be troublesome in that scenario.
Overall seemingly a good switch for the price, especially if you’re already a Ubiquiti user.
Ubiquiti actively evades sanctions to sell tech to Russia for use in war. Boycott.
@joachim source? If so, then yes, boycott.
I imagine that source would be the two recent Pablo Torre Finds Out episodes on same-name YouTube channel, about Robert Pera.
J’ai constaté quel que chose d’extrême ment grave appler moi de toute urgence je vous explique tout par téléphone SoS Tel :418-647-2222