Gowin Solution R86S-P2 and R86S-B3 2.5G Mini-PC
Table of Contents
In July 2022, I was contacted by Gowin Solution (also see their international website) to review their new mini-PC/firewall devices which have 2.5G Ethernet interfaces. Gowin is a newer Chinese company that is beginning to produce mini-PC/firewall appliances. The company seems interested in reaching out to bloggers like myself to generate interest in their products and to help provide guidance on software installation since their primary focus is on producing hardware.
Gowin has 3 main variations of the R86S: the “B”, “P”, and “G” series. Each variation has 3 configuration options B1-B3, P1-P3, and G1-G3. The “B” and “P” series have three 2.5G Ethernet interfaces while the “G” series has three 2.5G Ethernet interfaces and two 10G SFP+ interfaces. The P3 and G3 models come equipped with WiFi 6 and Bluetooth.
What peeked my interest the most was Gowin’s “G” series mini-PC model which contains both 2.5G Ethernet interfaces as well as 10G SFP+ interfaces. Personally, I think that is the perfect combination of interfaces for those of us home users who desire higher speed networks without using bulky, power hungry, loud enterprise gear or a custom built system.
I realize quieter, more energy efficient enterprise servers are available in a 1U half depth format, but I found those are typically newer generation options that can be pricier than older, used enterprise equipment.
I initially received the “B3” and “P2” models to review, but I plan to review a “G” series model at a later time once I receive it.
One thing to note is that both models that I received to review do not normally come with the WiFi module pre-installed. However, Gowin indicated that the WiFi module may be purchased separately if you decide later that you want to add it.
Unboxing
The Gowin R86S comes in a nice little black box which reminds me a little bit like packaging that Apple uses for their devices. There is a foam insert that holds the R86S and accessories in place for protection during shipping. The quality of the packaging is much higher than I expected it to be.
The contents include the R86S, the power brick, a screwdriver, and WiFi antennas if you purchase the models which come with WiFi.
Note
The screws on the case are tiny hex sockets so if you plan to open it, be sure to keep the included screwdriver unless you have precision tools available.
Build Quality and Features
When I first saw images sent to me by Gowin, the chassis appeared to be a smooth dark gray color but the actual chassis is black and has a slight texture on the coating that you can mostly only see in the light as demonstrated in the photos below. My first thoughts after opening the box is that the build quality seems nicer in reality than in the first images I saw. The R86S is heavier than I expected for its size.
On the front of the R86S “B” and “P” series models, there are three 2.5G Ethernet ports. It is interesting that there are only 3 interfaces instead of 2, 4 or 6 like most other mini-PC/firewall appliances. One reason I suspect is that Gowin wanted to make the R86S as small as possible. The R86S is small enough to fit in your pocket.
There is also a USB 3.0 port as well as the DC power socket on the front of each device.
The “B” series model has a less powerful CPU and is fanless like most mini-PC firewall appliances as you can see in the photo below.
However, since the “P” series model (and also the “G” series model which is not pictured below) has a faster CPU which consumes nearly double the wattage, it is not fanless. The box is virtually silent a short distance away but as you move your ear closer you will start to hear fan noise about a 1-2 feet away. The noise level is so low that you could have this device sitting on your desk and you would likely not hear it. If you want a completely silent/fanless device you may prefer the “B” series model rather than the “P” series.
As I mentioned earlier, the “P2” model does not normally come with WiFi, but my review unit came with that module so that I could try it out. I installed the antennas so you can see what they look like if you purchase a model with WiFi.
On the left side of the R86S, you will see another USB 3.0 port, a HDMI port, and a TF slot. A TF slot is basically another name for a microSD card slot. A microSD card reader may be convenient if you are using the R86S as a mini-PC since you could flash microSD cards to be used in Raspberry Pi’s or perhaps your phone/camera has data you wish to import on your PC.
The power button and power LED is also located on the left side of the R86S.
The right side of the R86S does not have any ports. If you are curious, neither does the back of the device (I did not take a picture since there is nothing of note other than the connections for the antennas which you can see in all of the pictures).
On the bottom, there is a removable plate that exposes the NVMe interface and the WiFi interface (or the WiFi module if it is preinstalled).
A standard size NVMe drive can be installed as shown below:
Above the NVMe drive is several ribbon connectors (it appears below the NVMe drive in this photo because I have the R86S updside down for the picture). For models with fans, they will be connected as well. I find it interesting that the CMOS CR2032 battery is attached to the bottom portion of the case. This computer is so small that the CMOS battery would take up too much space on the motherboard!
On the other side of the motherboard (which is the top of the R86S), you can see the CPU with the thermal paste applied. As expected the metal chassis is being used as the heatsink.
One important item of note is that the RAM is actually soldered on the motherboard to save space. I did not realize this until I saw the STH review of the R86S-G model since I had not fully disassembled the system. In the interest of having a thorough review, I took the device fully apart to take a look at it.
You may purchase the R86S with 8 GB or 16 GB of RAM, so I highly recommend you go with 16 GB since you cannot change unless you are on a tighter budget and you think 8 GB will be enough for the life of the device.
The R86S is a very compact form factor – even much smaller than most other mini-PC firewall devices as can be seen in the picture below. From left to right: the Qotom Q350G4, the Protectli VP2410, and the Gowin R86S.
In fact, it is not much larger than the Raspberry Pi 4 Flirc case! (affiliate link)
Hardware Specifications
The primary differences between the “B” and the “P” series models is the CPU. The “B” series has the Intel Celeron N5100 CPU while the “P” series has the Intel Celeron N5105 CPU. The N5100 is more power efficient than the N5105, but it also runs at a lower base clock speed. The burst CPU core frequency is nearly the same for both CPUs so it is likely the performance will be fairly similar (the single thread performance is 6.6% different according to cpubenchmark.net) but the N5100 is able to idle at a lower CPU frequency allowing for increased power savings. If you are willing to sacrifice a small amount of performance to increase power efficiency, you may prefer the “B” series model.
You may get the R86S with 8 GB or 16 GB of DDR4 RAM. The P3 and G3 models come with WiFi/Bluetooth. You can save costs depending on the configuration you choose.
One notable feature for all of the R86S models is that you can add a PCIe 3.0 x4 NVMe drive. You may use the drive as your primary disk to greatly improve performance over the built-in eMMC storage. Not all mini-PCs support PCIe NMVe so this is a nice addition to this tiny but mighty device.
GW-R86S-B3
| Hardware | Option |
|---|---|
| CPU | Intel Celeron N5100 @ 1.10GHz (4 cores, 4 threads) |
| Memory | Micron 16 GB LPDDR4X |
| Storage | 128 GB EMMC |
| Storage | NVMe PCIe 3.0 x4 (optional) |
| Storage | TF card (microSD) |
| Network | Intel i225v 2.5G |
| WiFi/BT | none* (optional WiFi 6/Bluetooth module) |
GW-R86S-P2
| Hardware | Option |
|---|---|
| CPU | Intel Celeron N5105 @ 2.00GHz (4 cores, 4 threads) |
| Memory | Micron 16 GB LPDDR4X |
| Storage | 128 GB EMMC |
| Storage | NVMe PCIe 3.0 x4 (optional) |
| Storage | TF card (microSD) |
| Network | Intel i225v 2.5G |
| WiFi/BT | none* (optional WiFi 6/Bluetooth module) |
Performance
Performance is often top of mind when considering a hardware purchase. Since the R86S has 2.5G network interfaces, testing out the performance on those interfaces will be one of the most interesting aspects because I know many users are migrating to 2.5G and faster home networks. The price of 2.5G interfaces have dropped in recent years so adoption has been increasing.
In order to test the maximum throughput of the R86S, I will perform the testing on my internal network, which has 2 machines with 2.5G interfaces. Both systems are connected to their own interface on the R86S with the interfaces configured as separate networks on OPNsense so that network traffic has to be routed across the interfaces. This will allow IDS/IPS performance to be tested since all of the packets entering the interfaces will be processed by the IDS/IPS service(s).
I am going to use simple iperf3 tests to get a general idea of network performance and how it is impacted by the IDS/IPS service(s) you have enabled. This type of basic network testing may not push IDS/IPS to its limits like heavy real world traffic, but it should provide a baseline of firewall throughput.
No IDS/IPS Enabled
With no IDS/IPS enabled, as you might expect, the throughput for the 2.5G interfaces on both the “B3” and “P2” models approach the theoretical maximum bandwidth. There is a little bit of overhead with the network interfaces so you will never see the full 2.5 Gbps being utilized, but it should be very close as you can see below.
The CPU utilization briefly spiked to 20-25% and would drop back down to a few percent. This would occur a few times during the iperf3 test. As you will see in subsequent tests below, the CPU overhead is much lower than using IDS/IPS, which is to be expected.
GW-R86S-B3
[ ID] Interval Transfer Bitrate Retr Cwnd
[ 5] 0.00-10.00 sec 2.64 GBytes 2.26 Gbits/sec 0 918 KBytes
[ 5] 10.00-20.00 sec 2.64 GBytes 2.26 Gbits/sec 0 1.32 MBytes
[ 5] 20.00-30.00 sec 2.63 GBytes 2.26 Gbits/sec 0 2.12 MBytes
[ 5] 30.00-40.00 sec 2.64 GBytes 2.26 Gbits/sec 0 2.12 MBytes
[ 5] 40.00-50.00 sec 2.64 GBytes 2.26 Gbits/sec 0 2.12 MBytes
[ 5] 50.00-60.00 sec 2.64 GBytes 2.26 Gbits/sec 0 2.12 MBytes
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-60.00 sec 15.8 GBytes 2.26 Gbits/sec 0 sender
[ 5] 0.00-60.04 sec 15.8 GBytes 2.26 Gbits/sec receiver
GW-R86S-P2
[ ID] Interval Transfer Bitrate Retr Cwnd
[ 5] 0.00-10.00 sec 2.64 GBytes 2.27 Gbits/sec 0 717 KBytes
[ 5] 10.00-20.00 sec 2.64 GBytes 2.27 Gbits/sec 0 868 KBytes
[ 5] 20.00-30.00 sec 2.64 GBytes 2.27 Gbits/sec 0 868 KBytes
[ 5] 30.00-40.00 sec 2.64 GBytes 2.27 Gbits/sec 0 868 KBytes
[ 5] 40.00-50.00 sec 2.64 GBytes 2.27 Gbits/sec 0 1.26 MBytes
[ 5] 50.00-60.00 sec 2.64 GBytes 2.27 Gbits/sec 0 1.89 MBytes
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-60.00 sec 15.8 GBytes 2.27 Gbits/sec 0 sender
[ 5] 0.00-60.04 sec 15.8 GBytes 2.27 Gbits/sec receiver
Zenarmor Enabled
With no IDS/IPS the performance of the “B3” and “P2” models are practically identical. However, when Zenarmor is enabled, the story starts to change.
GW-R86S-B3
The “B3” starts off strong but tends to quickly drop in throughput unlike the “P2”. I noticed the temperatures reaching 65-67C so perhaps there is some thermal throttling occurring or it is just reaching the limits of the slower N5100 CPU. Some of my tests would reach only 1.7 Gbps so I do not think you are going to break past 2 Gbps with this model with Zenarmor.
[ ID] Interval Transfer Bitrate Retr Cwnd
[ 5] 0.00-10.00 sec 2.60 GBytes 2.23 Gbits/sec 138 1.34 MBytes
[ 5] 10.00-20.00 sec 2.57 GBytes 2.20 Gbits/sec 2 1.35 MBytes
[ 5] 20.00-30.00 sec 2.03 GBytes 1.74 Gbits/sec 196 1.29 MBytes
[ 5] 30.00-40.00 sec 2.16 GBytes 1.86 Gbits/sec 90 1.87 MBytes
[ 5] 40.00-50.00 sec 2.10 GBytes 1.81 Gbits/sec 456 1.36 MBytes
[ 5] 50.00-60.00 sec 2.16 GBytes 1.86 Gbits/sec 53 1.69 MBytes
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-60.00 sec 13.6 GBytes 1.95 Gbits/sec 935 sender
[ 5] 0.00-60.05 sec 13.6 GBytes 1.95 Gbits/sec receiver
GW-R86S-P2
You will notice below that there is a very slight dip in network performance when Zenarmor is enabled. The throughput graph is not quite as flat as when no IDS/IPS enabled, but you are unlikely to notice the real world difference in throughput. However, there is a hit to the CPU utilization. It hovered around 50% utilization so 2 CPU cores were fully utilized when running the iperf3 test.
[ ID] Interval Transfer Bitrate Retr Cwnd
[ 5] 0.00-10.00 sec 2.64 GBytes 2.27 Gbits/sec 0 2.73 MBytes
[ 5] 10.00-20.00 sec 2.61 GBytes 2.25 Gbits/sec 88 1.29 MBytes
[ 5] 20.00-30.00 sec 2.60 GBytes 2.24 Gbits/sec 0 1.43 MBytes
[ 5] 30.00-40.00 sec 2.64 GBytes 2.26 Gbits/sec 0 2.02 MBytes
[ 5] 40.00-50.00 sec 2.63 GBytes 2.26 Gbits/sec 319 2.12 MBytes
[ 5] 50.00-60.00 sec 2.60 GBytes 2.23 Gbits/sec 86 1.29 MBytes
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-60.00 sec 15.7 GBytes 2.25 Gbits/sec 493 sender
[ 5] 0.00-60.04 sec 15.7 GBytes 2.25 Gbits/sec receiver
Suricata Enabled
To determine the worst case performance with Suricata, I enabled all of the rules for both LAN interfaces. Normally you would only enable rules which apply to the types of hardware, software, and network services you have running in your network.
GW-R86S-B3
There is a significant drop in performance when Suricata is enabled. You may be able to get higher performance by enabling less rules, but the CPU is likely too weak to handle Suricata with a high level of performance.
[ ID] Interval Transfer Bitrate Retr Cwnd
[ 5] 0.00-10.00 sec 513 MBytes 430 Mbits/sec 81 727 KBytes
[ 5] 10.00-20.00 sec 521 MBytes 437 Mbits/sec 56 638 KBytes
[ 5] 20.00-30.00 sec 508 MBytes 426 Mbits/sec 22 737 KBytes
[ 5] 30.00-40.00 sec 432 MBytes 363 Mbits/sec 8 766 KBytes
[ 5] 40.00-50.00 sec 440 MBytes 369 Mbits/sec 23 563 KBytes
[ 5] 50.00-60.00 sec 440 MBytes 369 Mbits/sec 21 585 KBytes
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-60.00 sec 2.79 GBytes 399 Mbits/sec 211 sender
[ 5] 0.00-60.06 sec 2.78 GBytes 398 Mbits/sec receiver
I must point out a more accurate representation might be to only have Suricata enabled on 1 of the 2 LAN interfaces since Suricata is often used on the WAN interface and not on any internal interfaces. When I only enabled Suricata on 1 interface to simulate traffic over a WAN interface to a LAN interface, throughput more than doubled. It is still much less than the maximum 2.5 Gbps throughput.
[ ID] Interval Transfer Bitrate Retr Cwnd
[ 5] 0.00-10.00 sec 1.14 GBytes 976 Mbits/sec 131 697 KBytes
[ 5] 10.00-20.00 sec 1.13 GBytes 970 Mbits/sec 33 783 KBytes
[ 5] 20.00-30.00 sec 1.13 GBytes 973 Mbits/sec 67 648 KBytes
[ 5] 30.00-40.00 sec 1.08 GBytes 929 Mbits/sec 44 687 KBytes
[ 5] 40.00-50.00 sec 1024 MBytes 859 Mbits/sec 45 626 KBytes
[ 5] 50.00-60.00 sec 1.03 GBytes 881 Mbits/sec 44 602 KBytes
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-60.00 sec 6.50 GBytes 931 Mbits/sec 364 sender
[ 5] 0.00-60.04 sec 6.50 GBytes 930 Mbits/sec receiver
GW-R86S-P2
Performance was better than I expected for the “P2” model. It is very possible that basic iperf3 tests do not strain the system like real world traffic. I did not notice a significant drop in throughput when running the test 2-3 times. However, I notice an interesting issue (see the warning below). I am not sure the root cause of this issue.
Warning
After running the test 2-3 times, the throughput would drop significantly. If I restarted the Suricata service, the throughput would go back to its max throughput for a few test runs. I do not know if this issue is related the fact the network interfaces are Intel i225v, which have been known to have lots of driver issues (or perhaps flaws in the actual hardware). I even tried running half as many rules and still encountered the same issue. I do not believe it is due to thermal throttling since the temperatures (especially on the “P2” model) look reasonable.
This issue does not seem to manifest with Zenarmor which you think might happen if is caused by issues with the i225v chipset. I wanted to mention this issue as a potential word of caution for anyone considering using Suricata on the “P” series models.
[ ID] Interval Transfer Bitrate Retr Cwnd
[ 5] 0.00-10.00 sec 2.64 GBytes 2.27 Gbits/sec 0 962 KBytes
[ 5] 10.00-20.00 sec 2.64 GBytes 2.26 Gbits/sec 0 1.00 MBytes
[ 5] 20.00-30.00 sec 2.64 GBytes 2.26 Gbits/sec 0 1.00 MBytes
[ 5] 30.00-40.00 sec 2.64 GBytes 2.26 Gbits/sec 0 1.00 MBytes
[ 5] 40.00-50.00 sec 2.64 GBytes 2.26 Gbits/sec 0 1.00 MBytes
[ 5] 50.00-60.00 sec 2.64 GBytes 2.26 Gbits/sec 0 1.00 MBytes
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-60.00 sec 15.8 GBytes 2.26 Gbits/sec 0 sender
[ 5] 0.00-60.04 sec 15.8 GBytes 2.26 Gbits/sec receiver
Below is what happens when Suricta is running on both LAN interfaces after running iperf3 several times:
[ ID] Interval Transfer Bitrate Retr Cwnd
[ 5] 0.00-10.00 sec 517 MBytes 434 Mbits/sec 75 721 KBytes
[ 5] 10.00-20.00 sec 545 MBytes 457 Mbits/sec 24 655 KBytes
[ 5] 20.00-30.00 sec 545 MBytes 457 Mbits/sec 30 561 KBytes
[ 5] 30.00-40.00 sec 545 MBytes 457 Mbits/sec 24 725 KBytes
[ 5] 40.00-50.00 sec 542 MBytes 455 Mbits/sec 30 641 KBytes
[ 5] 50.00-60.00 sec 545 MBytes 457 Mbits/sec 37 547 KBytes
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-60.00 sec 3.16 GBytes 453 Mbits/sec 220 sender
[ 5] 0.00-60.06 sec 3.16 GBytes 452 Mbits/sec receiver
If I only run Suricata on 1 of the 2 interfaces instead of both (when throughput drops), just like with the “B3” model, performance would more than double. However, the performance is still below 1 Gbps.
[ ID] Interval Transfer Bitrate Retr Cwnd
[ 5] 0.00-10.00 sec 1.15 GBytes 989 Mbits/sec 90 701 KBytes
[ 5] 10.00-20.00 sec 1.15 GBytes 988 Mbits/sec 31 805 KBytes
[ 5] 20.00-30.00 sec 1.14 GBytes 979 Mbits/sec 58 665 KBytes
[ 5] 30.00-40.00 sec 1.15 GBytes 988 Mbits/sec 55 762 KBytes
[ 5] 40.00-50.00 sec 1.15 GBytes 984 Mbits/sec 56 615 KBytes
[ 5] 50.00-60.00 sec 1.15 GBytes 986 Mbits/sec 50 727 KBytes
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-60.00 sec 6.88 GBytes 985 Mbits/sec 340 sender
[ 5] 0.00-60.04 sec 6.88 GBytes 984 Mbits/sec receiver
Zenarmor + Suricata Enabled
For this test, I enabled Zenarmor on one interface and Suricata on the other interface since they both cannot run on the same interface at the same time due to limitations with netmap. Typically you would run Suricata on the WAN and Zenarmor on the LAN.
One thing that is interesting about running both Zenarmor and Suricata is that the CPU stays pegged around 50% which indicates that both Zenarmor and Suricata are competing for the same 2 cores (the Zenarmor team has indicated to me that is likely what is happening based on what I am seeing). Because of this, throughput will take a hit when running both services. It is too bad that the services cannot be distributed better across the cores because it is possible much higher throughput could be available even for these lower power CPUs if the entire CPU was being fully utilized.
GW-R86S-B3
The performance of the “B3” is only slightly lower than when I tried running Suricata on only 1 of the 2 LAN interfaces. Zenarmor does not seem to impact performance as much as Suricata does.
[ ID] Interval Transfer Bitrate Retr Cwnd
[ 5] 0.00-10.00 sec 1.00 GBytes 860 Mbits/sec 93 776 KBytes
[ 5] 10.00-20.00 sec 1.02 GBytes 879 Mbits/sec 27 732 KBytes
[ 5] 20.00-30.00 sec 882 MBytes 740 Mbits/sec 33 747 KBytes
[ 5] 30.00-40.00 sec 875 MBytes 734 Mbits/sec 20 771 KBytes
[ 5] 40.00-50.00 sec 871 MBytes 731 Mbits/sec 23 792 KBytes
[ 5] 50.00-60.00 sec 794 MBytes 666 Mbits/sec 69 775 KBytes
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-60.00 sec 5.37 GBytes 768 Mbits/sec 265 sender
[ 5] 0.00-60.05 sec 5.36 GBytes 767 Mbits/sec receiver
GW-R86S-P2
Although the “P2” model seems to handle Suricata and Zenarmor well by itself, when both are running performance starts to drop below 2 Gbps. However, the performance is pretty respectable since the overall throughput is still slightly more than double a 1 Gbps interface.
[ ID] Interval Transfer Bitrate Retr Cwnd
[ 5] 0.00-10.00 sec 1.54 GBytes 1.32 Gbits/sec 954 1.81 MBytes
[ 5] 10.00-20.00 sec 2.07 GBytes 1.78 Gbits/sec 1 1.41 KBytes
[ 5] 20.00-30.00 sec 2.18 GBytes 1.87 Gbits/sec 1269 2.17 MBytes
[ 5] 30.00-40.00 sec 2.53 GBytes 2.17 Gbits/sec 0 2.65 MBytes
[ 5] 40.00-50.00 sec 2.64 GBytes 2.26 Gbits/sec 170 2.34 MBytes
[ 5] 50.00-60.00 sec 2.62 GBytes 2.25 Gbits/sec 21 2.09 MBytes
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-60.00 sec 13.6 GBytes 1.94 Gbits/sec 2415 sender
[ 5] 0.00-60.04 sec 13.6 GBytes 1.94 Gbits/sec receiver
Just like with Suricata enabled by itself, after running a few tests, the throughput would drop significantly on a consistent basis (something with Suricata is causing this issue).
Wireless
In OPNsense, the wireless on the R86S was not recognized so if you plan to use the built-in wireless for OPNsense, you should not purchase the WiFi model unless you think you may use the R86S later for other purposes.
To test WiFi performance, I installed Ubuntu which recognizes the wireless hardware and connected to my wireless network. Then I performed the same iperf3 test as I did for the wired interfaces. With iperf3, I got around 421 Mbps on average, but I have seen it fluctuate lower or higher than that. Wireless performance is not as consistent as wired performance.
The R86S supports WiFi 6, but I currently only have WiFi 5 at home so I am not able to test how much greater WiFi 6 speeds are compared to WiFi 5.
The performance on your network may be better or worse depending on the wireless devices you are connecting to and the environment where you live since you may receive more or less interference from your neighbors than I do.
[ ID] Interval Transfer Bitrate Retr Cwnd
[ 5] 0.00-10.00 sec 489 MBytes 410 Mbits/sec 99 2.07 MBytes
[ 5] 10.00-20.00 sec 504 MBytes 423 Mbits/sec 0 2.31 MBytes
[ 5] 20.00-30.00 sec 499 MBytes 418 Mbits/sec 143 1.64 MBytes
[ 5] 30.00-40.00 sec 518 MBytes 434 Mbits/sec 0 2.04 MBytes
[ 5] 40.00-50.00 sec 492 MBytes 413 Mbits/sec 295 1.69 MBytes
[ 5] 50.00-60.00 sec 511 MBytes 429 Mbits/sec 0 1.83 MBytes
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-60.00 sec 2.94 GBytes 421 Mbits/sec 537 sender
[ 5] 0.00-60.05 sec 2.94 GBytes 421 Mbits/sec receiver
Storage
The R86S comes equipped with eMMC storage which is faster and more reliable than microSD storage used by the Raspberry Pi and other devices, but eMMC is not quite as performant as a standard SSD as you can see in the benchmark below. SSDs can reach peak read speeds of approximately 550 MB/s. The eMMC reached a peak around 320 MB/s for reading and 255 MB/s for writing so it is slightly less than 2 times slower than a standard SSD.
However, the eMMC storage is still quite a bit faster than typical microSD storage as you can see below. The microSD card barely reads and writes above 20 MB/s. There may be microSD cards which are faster or slower than the card that I used so your results may vary.
The R86S supports PCIe 3.0 x4 NVMe storage so I added a NVMe drive to see how much faster it is compared to the eMMC. Of course, NVMe should be much faster, but I wanted to find out if the full PCIe 3.0 x4 bandwidth can be utilized. I tested the same Samsung 970 Evo Plus NVMe drive (affiliate link) on my Proxmox server, and it was able to reach the advertised speeds for reading (~3,500 MB/s) and nearly the advertised speeds for writing (~3,100 MB/s).
In the screenshot below, you can see it does indeed utilize essentially the same amount of PCIe 3.0 x4 bandwidth as a desktop PC, which is a huge performance boost for your storage. I would recommend taking advantage of this additional performance especially when 500 GB NVMe drives can be relatively inexpensive.
Power Consumption
For basic power consumption, I used an energy monitoring smart plug which I have connected to Home Assistant so I could monitor the power usage over time in a nice graph. Depending on the accuracy of the energy monitoring circuitry, this may not necessarily be the best way to test power consumption. However, the values I saw seemed very reasonable considering the TDP of both the Intel N5100 and N5105 are 6W and 10W respectively.
GW-R86S-B3
The TDP of the R86S-B3 is 6W, but interestingly, I noticed when compared to the R86S-P2 that the power consumption is not much lower. When I booted up the R86S-B3, the wattage peaked around 12W. The idle wattage seemed to be around 9-10W which is actually very similar to the R86S-P2, and I even added NVMe storage which adds a small amount of additional power consumption. While using the box to open applications, run WiFi performance tests, etc., I did not see the wattage exceed 16 watts. If any extra peripherals are used, energy usage will increase.
GW-R86S-P2
The power consumption on the R86S-P2 seems to be around 9-10W when idling and when testing throughput with IPS/IDS enabled it jumped to 12W with 50% CPU utilization. When I booted it up, the wattage peaked around 14-16W. While using the box to open applications, run WiFi performance tests, etc., I did not see the wattage exceed 16 watts. If any extra peripherals are used, energy usage will increase.
While the energy usage of both models is not as low as the Raspberry Pi, the performance is still very impressive for the amount of power being utilized. The R86S (like other mini-PC firewall systems) is energy efficient enough to be used as a 24/7 home router. If you are using this device as a low end PC, you will save a lot of energy compared to a standard desktop PC.
Temperature
On the OPNsense “Dashboard”, I added the thermal sensors widget so that I can view what the onboard temperatures are reporting (I had to set the thermal sensor hardware to “Intel Core CPU on-die thermal sensor” on the “System > Settings > Miscellaneous” page to see the CPU temperatures). I do not currently have a heat gun to measure the outside temperature of each model.
GW-R86S-B3
During testing, I saw temperatures between 49-67C with the average idle temperature around 50-52C. Since the “B3” model is fanless, the temperatures are higher than the “P2” model. The “B3” model was reasonably hot when I touched it after performing CPU intensive tasks. However, it is not so hot that you could actually burn your hand. The case is acting like a heatsink and drawing a lot of the heat to the outside of the system.
I did notice that the temperature does spike a lot more than the “P2” model which has a fan. Although the normal running temperature is in the low 50s, I would see it peak to 65-67C when running a heavy load on the system. If you plan to have heavier workloads, you may wish to get the models with the fan to ensure the system stays cooler (and you will also have a faster CPU as well).
GW-R86S-P2
For the “P2” model, I noticed the temperature range was between 41-52C. On average, the temperature stayed between 42-43C when mostly idle, which is not too bad. The fan on the “P2” model definitely helps keep it cooler than the “B3” model even though the N5105 CPU of the “P2” model has a higher TDP of 10W vs 6W for the N5100 CPU.
When I physically touch the “P2” model, it does feel slightly warm but not nearly as warm as the fanless “B3” model. Both models are noticeably warmer to the touch than the Protectli VP2410 that I previously reviewed even though the CPU temperature of the Protectli is about the same or slightly higher than the R86S. I credit the difference to the heatsink design of the Protectli which makes the outside of the box feel a lot cooler even though it does not have a fan. There is a tradeoff between a compact form factor and cooling.
I do not foresee any issues with overheating with either model unless perhaps you locate it in a more confined space such as a structured media enclosure found in many homes (some do not have holes to allow airflow into the enclosure). However, the “B” series model may have the greatest risk for overheating since it does not have a fan and generally runs hotter than the “P” series. There is a lot of power packed into a small form factor so the addition of the fan in the “P” series models is very beneficial for keeping the temperature at a more reasonable level.
Where to Purchase?
Gowin does not currently have the devices for sale on their international website at the time of this writing. The boxes are actually available for purchase on AliExpress (affiliate link) , but Gowin expressed that those are from resellers without authorization to sell their products (at the time of this writing).
Whenever Gowin officially sells their products on their website, I will update this page with the relevant information. I plan to include affiliate links so I will be able to earn commissions for any purchases made by my readers, which would help me fund future projects and reviews! All support by my readers is always greatly appreciated!
Also, if you do purchase a R86S through AliExpress or through official channels, be sure to check out my OPNsense installation guide tailored specifically for the R86S (“B” and “P” series models).