The Rockchip RK3568-powered NanoPi R5S SBC with two 2.5GbE ports, one Gigabit Ethernet port, and M.2 NVMe storage is now available for $59, or $75 with a metal enclosure.
As previously mentioned, the mini router board is equipped with 2GB RAM, 8GB eMMC flash, two USB 3.0 ports, as well as an HDMI output for people wanting to make use of the Rockchip RK3568 processor’s multimedia capabilities, or simply have a user interface on a monitor.
NanoPi R5S specifications:
- SoC – Rockchip RK3568 quad-core Cortex-A55 processor @ up to 2.0 GHz with Arm Mali-G52 MP2 GPU, 0.8 TOPS AI accelerator, 4Kp60 H.265/H.264/VP9 video decoder, 1080p60 H.264/H.265 video encoder
- System Memory – 2GB LPDDR4X
- Storage
- 8GB eMMC flash for OS
- Key M socket for M.2 2280 (PCIe 2.0 x1) NVMe SSD support
- Optional SPI flash for network boot
- MicroSD card socket
- Video Output – 1x HDMI 2.0 port up to 4Kp60, or 1080p120
- Networking
- LAN – 2x 2.5GbE RJ45 ports (via 2x Realtek RTL8125BG PCIe controller) tested up to 2.35 Gbps (Rx) and 1.85 Gbps (Tx)
- WAN – 1x Gigabit Ethernet RJ45 port (via Realtek RTL8211F) tested up to 941 Mbps (Tx and Rx)
- USB – 2x USB 3.0 ports
- Expansion
- 16-pin 1.27mm pitch GPIO connector with 1x SD 3.0, 1x I2S (2x SDO, 3x SDI)
- 12-pin 0.5mm pitch FPC connector with up to 1x SPI, 3x UART, 4x PWM, 8x GPIO
- Debugging – 3-pin UART header
- Misc – Mask key for eMMC flash update, RTC battery connector, 5V fan header, 4x user LEDs
- Power Supply – 5V/9V/12V USB Type-C port (USB PD support)
- Dimensions – PCB: 90 x 62 mm; enclosure: 94.5 x 68 x 30 mm
- Weight – 57.5 grams without the case, 260 grams with the metal enclosure
- Temperature Range – 0℃ to 70℃
We did not have information about software support last time around, but now we do with the company providing FriendlyWrt 22.03 (64-bit) and FriendlyCore Focal Lite images based on OpenWrt 22.03 and Ubuntu 20.04 respectively, but featuring Linux 5.10 LTS and U-boot 2017.09. The company also highlights support for Docker, notably showing instructions to install Jellyfin multimedia applications. You’ll find additional technical documentation in the Wiki.
FriendlyElec does not typically include WiFi in their router boards, and instead relies on customers making use of USB WiFi dongles some of which can be used as access points.
The NanoPi R5S offers an upgrade path to either NanoPi R2S (Rockchip RK3328 quad-core Corex-A53) or NanoPi R4S (Rockchip RK3399 hexa-core Cortex-A72/A53) devices, and one consequence of using Cortex-A55 cores is that the RK3568 mini router operates more efficiently and with the metal enclosure, the CPU temperature never exceeds 50°C based on measurements made by FriendlyElec.
I’ll soon be able to test all that myself, as I’m expecting a package from DHL with two boards and enclosures later today.
Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.
Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress
If there’s one thing I really like with FriendlyElec, it’s their metal enclosures! IMHO they’re, and by a large margin, the best you can find in the SBC world. They’re heavy because the metal is directly etched in a solid block, but as a result they’re extremely efficient thermal-wise. In addition they make their boards particularly sturdy; you don’t fear the device falling on the floor when searching in your computer bag. Retrospectively the 3D-printed plastic versions for the NEO and NEO2 were a joke compared to this 🙂
Looking forward to the test, but this is as I keep insisting, not a router SoC.
Yep. Not router class hardware / something for serious fast routing, but IMO good enough for home / house on the beach / SOHO use cases.
Can it even saturate a 2.5GbE-link in any practical situation, though? iperf and other synthetic benchmarks are one thing, but e.g. real NAS-type usage and such are a different thing.
Based on RK3568 data available (e.g. this funny ‘benchmarking gone wrong’ exercise) I would expect NAS transfer rates above 150 MB/s with a PCIe attached SATA controller.
Would be ok-ish for my ‘home use case’ (using R5S as NAS + ‘switch extender’ combo with all 3 NICs in bridge mode and the 2.5GbE ports connected to those 1 or 2 2.5GbE capable clients that benefit the most from higher NAS transfer rates).
Yeah prob a great PiHole on steroids like board but maybe not the same as TLS but I also think the greatest thing about that board is not as a router but with a x5 sata m.2 and x1 on board you still have room to aggregate those 2x 2.5gbe if you can get a 5gbe phy but seems they can only source from realtek.
As it makes a much better NAS than a router.
As a general purpose SBC its a bit mhew IMO 🙂
> Looking forward to the test
IIRC Jean-Luc lacks the proper equipment to do some serious testing (various 2.5GbE thingies known to achieve max throughput and min latency to test against). Also with those four lame A55 at less than 2.0GHz some initial efforts might be needed to tweak SMP/IRQ affinity, RPS and the like… and while testing constant observation of detailed CPU utilization needed to identify/eliminate bottlenecks.
I have a working 2.5GbE USB dongle from RealTek now, and I still own the UP Xtreme i11 mini PC, so it should be OK.
> this is as I keep insisting, not a router SoC
Personally for a ‘business use case’ I am really interested in packet forwarding/blocking rates (to use this thing as ‘border firewall’ or border packet filter) with full control by the kernel (and not some proprietary SoC ‘engine’). Most probably for this all boils down to internal interconnect latencies…
That seems like a more reasonable use case, if the hardware delivers.
Hi tkaiser, how would you test the forwarding/blocking rate?
I’ve also got a R5S and I want to benchmark it against a USB-C to 2.5G adapter that the reviews claim it can do 2.5G just fine.
That’s a really good question. Currently thinking about lazy mode using iperf3 (first against R5S, then with R5S in between just routing, then with some rules) but usually while benchmarking you realize that something went wrong, your numbers are trash and you need to start over with a different test setup anyway (at least that’s what I do all the time as part of my day job).
I will try to start where I left the last time with the R4S after your comment. Currently, I don’t have a 2.5G switch and I don’t plan to get any tbh, because I don’t have a use for it; therefore I will only do p2p benchmark, which pretty much I expect to be the best case scenario. On the other hand, since it has 2 ports it makes total sense to benchmark both ports L1<->L2 and then WAN<->L1&L2. I think for that I’ll need another USB-C adapter. Maybe I’ll do this later. As you’ve proposed, after the initial benchmarking… Read more »
Unfortunately no basic monitoring happened while you were testing.
And we can’t trust into RK3568 clockspeeds (for whatever reasons they’re almost always lower than what cpufreq driver reports) and since all of this is based on a new Android BSP kernel (though using 5.10 and not 4.19 like last year) I would expect silly things happening.
Actually, I was monitoring the CPU freq and I didn’t notice a drop while polling the /sys values.
But of course, I can’t be sure if this value corresponds to the actual freq.
cpuinfo_max_freq is just a fixed device-tree entry, scaling_max_freq can be adjusted (limiting cpufreq driver), while cpuinfo_cur_freq or if missing scaling_cur_freq is what should print actual cpufreq according to driver.
But reality is different. 14 times measured (my sbc-bench results collection) and only a single time real clockspeed was at 2.0GHz: http://ix.io/3ZaI
OK, that’s embarrassing I thought I was monitoring the real-time freq lol. I need to fix that.
Even if you monitor the correct sysfs node real clockspeeds might still differ (that’s why sbc-bench is measuring them instead of trusting in some number).
Anyway: all this testing is IMO rather useless without monitoring the system. Curious whether sbc-bench -m works on the FriendlyWRT image? If not it’s time to switch to performance cpufreq governor and then run ‘iostat 5’ and atop/htop in parallel to see what’s going on.
I’ve now seen your tool. I’ll keep it for future reference.
So, it fails to run on the OpenWrt, but I’ll test on the rk3568-sd-friendlycore-focal-5.10-arm64-20220526.img.gz image, which runs on the SD card and post you the results if succeeded.
FYI, here’s the output.
http://ix.io/3Zb1
Thank you! RK3568 only clocked at ~1870 MHz while pretending to run at 1992 MHz 🙂
So far the slowest clockspeed reported (asides running with mainline w/o cpufreq support).
grep ‘SoC guess’ *.txt | grep RK3568 | cut -f1 -d’:’ | while read ; do echo -e “$(tail -n1 “${REPLY}” | awk -F”|” ‘{print $2}’)($(grep ‘^Linux’ “${REPLY}” | head -n1 | awk -F” ” ‘{print $2}’)): $(grep -i ‘measured’ “${REPLY}” | head -n1)”; done | curl -s -F ‘f:1=<-‘ ix.io -> http://ix.io/3Zb4
Can I please share your list in my post?
Of course, I’ll refer you as I’ve already done.
Sure, but IMO you should mention this other blog post (and comment thread) as well since I think we’re getting some further insights along the way 🙂
Was your R5S shipped to Europe? If so in which time and by China Post or DHL? Currently thinking about ordering one since wanting to do the necessary optimizations and with this thing on my desk less time wasting.
Yes sure! I haven’t seen that Jean-Luc also released his review. I’ll add a link to that of course. The device was posted in Germany with DHL, it took 2-3 days to arrive, but I also paid ~43€ for customs. I didn’t pay for the shipping, so I don’t know the price.
I’ll also follow the other post now.
What would be different about a “router soc”? I use an r4s as a router and it can easily saturate the 1gbit ports. Also, it’s CPU is WAY better than what any internet provider router has. (Like Fritz Box)
Any idea if the uSB-C port is data enabled or just power?
It’s just for power as I understand it.
Just power. They do say it quite clearly on their website, if you’d just go and take a look.
I don’t find all of the Ethernet ports useful. The optimal config for me would have been a single Ethernet and spot for an optional, good wifi module. Switch the ports around to include two PCIe sockets — one for storage, one for a 2×1 for a dual Coral module. Also MIPI display instead of HDMI (both possible?). And convert one of the USB into a SATA port.
Router not that useful. I want storage plus AI.
What do you mean by “storage plus AI”?
The NVR app this chip was designed for. AI to process camera video. Rockchip supplies a sample NVR app for it. I have not spotted a board yet with the connectors I’d like to see.
I came across the Mrkaio AIO-M68S by accident (another RK3568 that can’t clock at 2.0 GHz but remains at ~1.9GHz). They claim SATA and MIPI-DSI but seem to ignore PCIe completely.
For the AI camera NVR use case…
PCIe E key with 2×1 for a dual Coral AI board.
https://coral.ai/products/m2-accelerator-dual-edgetpu/
M.2 for PCIe SSD
One Ethernet
Good wfi (for people with no Ethernet)
A SATA port if you want slow back up.
MIPI DSI for option LCD inside case
HDMI for external screen, neither screen is required.
Then run Home Assistant Frigate on it.
https://docs.frigate.video/
Use case differ a lot between people. I know some field network engineers who will be very happy to finally be able to make a reasonable network tap with this. You put two ports in a bridge, no IP, capturing to the SSD, and managing via the last port. The last one I found to decently match that use case was the OpenBlocks AX3, but it was bulkier and required a power plug, while this one only needs to be connected to a USB port of the server it’s capturing the traffic from.
Needs that rk3588, where are those chips 😤
I wish, would make for a really nice home router with enough power to run a few other things like PiHole, WireGuard, bandwidth monitoring, QoS, etc. Maybe even some light weight services. I’d pay at least 2x for the Rk3588 and twice the ram.
A downgrade compared to r4s. They should’ve used the RK3588…
How so? Better CPU (rk3568 vs rk3399) and more ports.
The CPU in RK3399 will be faster, but R5S offers 2.5GbE, while R4S only has Gigabit Ethernet ports. R4S might still be better for CPU-limited applications.
> CPU in RK3399 will be faster
RK3399 scores ~30% faster than RK3568 with single-threaded tasks and ~40% multi-threaded. And all of this is pretty much irrelevant if no sensible fine-tuning of settings happened (eliminating artificial bottlenecks like ‘all IRQs on cpu0’ and the like). Without appropriate settings faster CPU cores might just ‘idle harder’ while a slow CPU core is at 100%…
The point is that on Aliexpress all shops offer 2G and 4G versions, while Freiendlyelec is saying only about 2G version so far. Shall I rely on that sellers?
You can see on friendly elec website an image with 2Gb/4Gb LPDDR4X written on it. Probably both will be available. But only 2GB for now. A lot of Aliexpress sellers where taking Pre-Order. So a think they are probably right but you will likely have to wait longer if you choose 4Gb
What is the purpose of the WiFi button (?) on the housing above the USB-C?
It’s not a button, it’s an opening for a WiFi antenna. There’s just a black plug in it by default, since the device doesn’t ship with any WiFi.
Do you know of this would work with a wifi pcie card? That way I can make a small wifi 6 ap with 3 Ethernet ports
I’m trying to mate an Ubuntu capable router with solar controllers and PoE gear, like WAP or security cameras / lights or fire alarms. It really is a huge disadvantage to power by USB-C not PoE. To get this to talk to anything in the real (PoE) world needs a midspan or an injector each to power even puny PoE devices, a USB-C supply just to power it that probably can’t talk to it over the same connection, the other USB-C to talk to the controller chain. Meh. If it was powered by one GbE PoE and output power on… Read more »
while not best option, there are adaptors for 802.3af towards USB
(e.g. .www.tekta.store/en-gb/product/dslrkit-gigabit-usb-type-c-active-poe-splitter-48v-to-5v-ieee802-3af-power-over-ethernet-for-raspberry-pi-4b/96758206746790828)
Poe to, adaptors flood aliexpress, from cameras to whatever use. Just check the specs meet your needs.
Could you please test the ‘tun’ driver works to create a working OpenVPN connection please with OpenWRT? I could not get it working with a RK3568 station P2 which is very frustrating. Cheers
I was planning to test Wireguard instead since it appears to the be recommended VPN solution now. Is there any specific reason to use OpenVPN instead?
Ahh no that is just the configuration I was trying with no luck on any of the images the oem provides and thought it would be a popular configuration. I will look at wireguard
iperf3 results are a bit disappointing out of the box:
https://www.cnx-software.com/2022/06/02/nanopi-r5s-router-review-part-1-unboxing-openwrt-and-iperf3-benchmarking/
This board with 4Gb and case is actually a better buy on AliExpress than it is direct from FriendlyElec. Same ship date and Friendly wants a boatload of shipping for the 2Gb model. What gives?