Radxa E52C is a compact router based on Rockchip RK3582 hexa-core Cortex-A76/A55 SoC and featuring two 2.5GbE ports, a USB 3.0 port, and a USB-C port for serial console access which will make some readers happy…
It’s an update to the earlier Radxa E20C “Mini Network Titan” router, which was limited to two Gigabit Ethernet ports and a USB 2.0 port, and powered by an entry-level Rockchip RK3528A quad-core Cortex-A53 SoC. The Radxa E52C is better in every way with a much faster CPU, higher maximum memory and flash storage capacities, faster networking, and support for USB 3.0 storage or wireless dongles leading the company to call it an “Enhanced Mini Network Titan”.
Radxa E52C specifications:
- SoC – Rockchip RK3582
- CPU
- Dual-core Cortex-A76 with up to 2.4 GHz
- Quad-core Cortex-A55 at up to 1.8GHz
- GPU – None (all good since we are talking about a router here…)
- AI accelerator – 5 TOPS NPU
- CPU
- Memory – 2GB, 4GB, or 8GB LPDDR4
- Storage
- 16GB, 32GB, or 64GB eMMC flash
- MicroSD card slot
- Networking – 2x 2.5GbE RJ45 ports via Realtek RTL8125BG controllers
- USB – 1x USB 3.0 Type-A port for data and flashing firmware
- Debugging – USB-C port for serial console access
- Misc
- User button
- Maskrom pinhole
- WAN, LAN, and System LEDs
- Power Supply – 5V DC via USB Type-C port; 3A recommended when connecting peripherals to the USB 3.0 port
- Dimensions – 72 x 72 x 28.7 mm (Aluminum alloy enclosure)
On the software side, it’s pretty much the same as for the E20C router with the E52C board running Debian Linux, FLippy OpenWrt, or iStore OS, another fork of OpenWrt with a user interface that aims to simplify the configuration process and also allows the user to easily configure the system as a simple USB NAS. Radxa also provides a hardware access/control library for Linux. But none of that appears to be available right now as the Download section on the product page only contains links to a product brief, schematics (PDF), a components placement map, and 2D drawings at this time.
You can already order the Radxa E52C with 2GB RAM and 16GB flash only on Arace Tech or AliExpress for about $55-$56. Other models are out of stock, but the 4GB/32GB version will go for $65, and the 8GB/64GB variant for $85. Radxa guarantees supplies for the Radxa E52C until at least September 2034. Note that’s the second Rockchip RK3582-based product from the company, as the Radxa ROCK 5C (Lite) was launched a few months ago as a GPU-less Raspberry Pi 4/5 alternative.
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
That’s exactly what I’ve been looking for over the last year or so. Dual-2.5, native storage, debug port, enclosure. Nothing more is needed. Combine that with a $60 8-port 2.5G switch and you have an easily manageable 9-port firewall. I ordered mine. I’ll see if mainline starts to work on these chips or not :-/
mainlining is under review [RFC,v2,2/2] arm64: dts: rockchip: add Radxa E52C – Patchwork
Great, if the DTS is the only thing missing, that’s a detail! Thanks for the link! I’m essentially concerned by the fact that till now I never managed to boot my rock5 itx on mainline (last attempt was 6.11-rc4), it just hangs silently during device detection, and since nobody shares info about any success, it’s hard to know what should be expected.
BTW my rock5 ITX now boots and works fine on latest mainline (eMMC, PCIe etc) so I’m confident that the router above will work out-of-the box without having to rely on rockchip’s kernel.
I thought of you when I saw the UART, I figured you might be in for one.
What switch would you have in mind?
> What switch would you have in mind?
I’m using this one and another one with 2x10G + 4×2.5G: https://www.aliexpress.com/item/1005006470298678.html
But now there are even more options. If you don’t need that many ports, you can find 5-port switches as well for an even lower price, around $50.
Both are web-managed and support VLANs. It’s not always trivial to reliably figure which ones are really managed or not, what I found is that when they indicate VLAN support in the spec, then you’re sure they’re managed (otherwise you can’t configure them).
Thank you for the link!
I am wondering if the CPU can support 2.5 Gbit, even worse, 2 x 2.5 Gbit. Obviously switching won’t be hardware accelerated.
it depends what you mean by:
as even the cheaper rk3566/8 can support simple NAT/firewalling at 2.5gbps
but when you add pppoe in the mix (I know you shouldn’t but ISPs have their own agenda) things might become interesting.
Also doing QoS (like SQM) or IPS/IDS at line rate might most likely be too much for this little box
I’ve pulled 25-26 Gbps bidirectionnal from the Rock5B which is only slightly bigger (2 extra big cores). I indeed hardly see how that one would have difficulties at only 20% of that!
The most important is not the data rate but the packet sizes and the connection rate anyway. Just forwarding traffic is cheap, my old Guruplug (single-core armv5 1.2 GHz) was already able to forward 1 Gbps between the two ports.
If they had added a display connector this would have been a nice multi purpose Computer that could be used as a Router. USB-C DP mode would have been enough and licence free (hdmi is proprietary and require a licence for the connector).
This thing is built around the RK3582 which is a cheaper RK3588S(2) variant with broken or hidden GPU (silicone lottery). Something w/o a 3D GPU is most probably not best suited as a ‘multi purpose Computer’ with display connector.
Agreed, and one of the problems many SBCs face precisely is that in order to maximize sales, their vendors try to make their boards as versatile as possible and in the end they’re just average in everything (including cooling sometimes). At least with a few vendors like Radxa, FE and a few others who are also able to make multiple models, it’s possible to have hardware optimized for a given use case. And that’s great. There are sufficient other options for a “multi-purpose computer that could be used as a router”, such as FE’s NanoPI R6S which has 3 network… Read more »
The GPU of the RK3588 cant be used with just free software. So its already useless in the therms of FOSS. If you take a 6 or 8 core CPU it does not change that much. You do not need a 3D GPU for Video playback and also not browsing websites. Also not for writing a book in LaTeX. Also writing code, for example writing OpenWRT code, would be a usecase. Run full OS, git clone, modify, compile and then run on same machine the compiled image. Just have to connect keyboard, mouse and screen and you can have this… Read more »
What you describe is just basic frame buffer (the same stuff that I use BTW). Sadly, there were some initiatives a long time ago to make VGA-compatible chips with high resolutions and very little acceleration precisely for such use cases, and these never took off because aside developers, nobody wants to buy them. Nowadays graphic chips need to support the latest whatever video decompression algorithm to show videos and I-don’t-know-what 3D acceleration stuff to display funny screen savers… With that said I’ve already found small mini-pcie video cards that could possibly match your common needs. Also maybe a usb-based video… Read more »
I know what match my need when its about PCIe slot and FOSS. The whole RK3582 device we are talking here about is smaller then such a PCIe slot. Talking about such a irrelevant PCIe Adapter is useless. The RK3582 can run 8k screens and have active decoders and encoders. Its not the basic thing you know since many years where the GPU is fully missing. Please read the datasheet first before telling wrong things: 1.2.4 Video CODEC Video Decoder Real-time video decoder of MPEG-1, MPEG-2, MPEG-4, H.263, H.264, H.265, VC-1, VP9, VP8, MVC, AV1 MMU Embedded… Read more »
With RK3582 you don’t know which cores you are given. Out of the 2 encoder and 2 decoder cores, two are disabled. But you don’t know in which combination, so in the best case you have 1 each and in the worst case you have 2 dec or 2 enc broken. If you’re lucky you have a super powerful jellyfin server, if unlucky, you can basically make a NAS/firewall.
I don’t understand your point. You’re talking about having just enough graphics to develop on the device, i.e. you want to display a terminal, so I suggested possible small alternatives (mini-pcie based, e.g. like https://www.86duino.com/?p=8586 or https://www.innodisk.com/en/products/embedded-peripheral/display for example, or external ones USB-based) and you start digressing about the device size compared to a totally irrelevant PCIe slot (why?), and enumerating tons of features implemented in a device that you previously rejected as non-free, and which offers many features that you previously confirmed as not needed since you want the device just for development. So my first interpretation is that… Read more »
While I don’t know, whether a full FOSS stack exists at present for using whatever display-related circuits are left in this device’s SoC, I agree with the sentiment, that an opportunity was missed here by not having display output support on the USB-C connector. I also don’t see the necessity for 3D accelaration hardware for my typical usage scenarios of web browsing, text-writing and the consumption of audio or video content. This device is already rather expensive, considering the availability of 4-W-consuming, Atom-based, used thin clients in sub-1-liter enclosures for less than €15 (alas with sub-2.5-GbE speeds), but its dimensions… Read more »
> This device is already rather expensive … less than €15 (alas with sub-2.5-GbE That’s precisely the point. Here you’re paying for a small powerful device capable of offering great connectivity for modern SOHO use cases. And it’s way cheaper than the equivalent x86 devices, which are often found around 300 EUR (RAM+storage included) when they have at least two 2.5G ports, and they suck much more power. The equivalent single-port board with video is the Rock 5C. Maybe they should also make a dual-port + GPU but I don’t really see the point. It’s quite rare to watch videos… Read more »
Since the USB-C ports are already there, they could at least have made it an option just like the options for more RAM or Flash they have.
[ With increased performance&capability SBC hardware it’s astonishing, there’s almost no cheap and small form factor device for PCIe (at least a standard 3.0, ~>x4(-x16), version) hardware development available. (maybe there’s some alternative with rpi6? and/or USB4 getting more affordable) Most popular PCIe devices are probably graphic cards, storage expansion, network upgrades and AI&ML accelerators. It seems ‘multi purpose computers’ are still a x86 domain(?), at least if it’s about standard interfaces&peripherals and multi core variety, especially with x86 lowering standby&idling consumption (for mainboards) within recent years and available mainline support. HDMI (while being the more versatile and available) “There… Read more »
The problem remains the low connectivity of general purpose chips in the Arm world. Take a marvell or NXP and you’ll have plenty of PCIe lanes. And I’m not even speaking about Altra’s 128 Gen4 lanes! Take an amlogic (often none) or a rockchip (often just a few). But that’s much better than what it was a few years ago. We’ve already moved from the single-lane PCIe2.1 to 2 lanes in a low-end chip. It’s not that bad! The difference with x86 precisely is there. A low-end intel N100 comes with 9 Gen3 lanes, or almost the double of the… Read more »
A little bit bizarre, the product main attraction is a chip lottery 🙂 “The way I read this is that the RK3582 is an RK3588S with defects. Rockchip must have noticed they had a non-negligible number of RK3588S that did not pass QC (quality control) due to a few Cortex-A76 cores not working, the GPU failing, or the NPU not working up to expectations, so the RK3582 was born. That’s why you may get four Cortex-A76 cores or the GPU working in some RK3582 parts, but likely not both of them, unless the chip was downgraded from being a RK3588S… Read more »
What’s so bizarre about selling a chip with a reduced guaranteed feature set at a lower price instead of binning it? At best, you get more than what you paid for.
This approach is perfectly suitable for enthousiasts, typically the same people who buy RPis and are happy to get more than what they paid for. However for consumer products you want something stable, but my understanding is that defective cores are properly marked so that the OS can disable them. In this case it means very little operations are needed in the factory (i.e. probably just flashing a fuse during testing), and limited work needed on the product side (integrate the code that disables these faulty cores). This allows to sell good performance CPUs at a very low price. It’s… Read more »
By the way, it’s a bit comparable to the DVFS/PVTM stuff where the operating system takes the silicon quality into account to limit frequency scaling.