Mekotronics R57 is a fanless edge AI mini PC powered by a Rockchip RK3576 octa-core Cortex-A72/A53 SoC with a 6 TOPS NPU, 4GB LPDDR5, a 32GB eMMC flash, and features such as RS485 and dual gigabit Ethernet that makes it suitable for industrial automation.
I initially thought it was a cost-down version of the Mekotronics R58 mini PC that I reviewed with Android 12 as my first ever Rockchip RK3588 device. But while the mechanical design is similar and some of the ports are in the same position, there are some changes with the new model equipped with only one HDMI output, one HDMI input, two GbE jacks, and a terminal block with RS232, RS485, and digital inputs and outputs.
Mekotronics R57 specifications:
- SoC – Rockchip RK3576
- CPU
- 4x Cortex-A72 cores @ 2.2GHz, four Cortex-A53 cores @ 1.8GHz
- Arm Cortex-M0 MCU at 400MHz
- GPU – ARM Mali-G52 MC3 GPU with support for OpenGL ES 1.1, 2.0 and 3.2, OpenCL up to 2.0 and Vulkan 1.1
- NPU – 6 TOPS (INT8) AI accelerator with support for INT4, INT8, INT16, BF16, and TF32 mixed operations.
- VPU
- Video Decoder: H.264, H.265, VP9, AV1, and AVS2 up to 8K @ 30fps or 4K @ 120fps
- Video Encoder: H.264 and H.265 up to 4K @ 60fps, (M)JPEG encoder/decoder up to 4K @ 60fps
- CPU
- System Memory – 4GB LPDDR5
- Storage – 32GB eMMC flash
- Video Output – HDMI 2.1 port up to 4Kp120
- Video Input – HDMI port
- Audio – 3.5mm audio jack, digital audio output via HDMI
- Networking
- 2x Gigabit Ethernet RJ45 ports
- WiFi and Bluetooth
- Optional 4G LTE/5G cellular connectivity via mini PCIe socket and SIM card slot
- 4x antenna holes
- USB – 1x USB 3.0 port, 2x USB 2.0 ports, 1x USB Type-C port with DisplayPort 1.4 alt mode
- Expansion – Terminal block with RS232, RS485, 2x digital inputs, 2x digital outputs, 12V, and GND
- Misc
- Power button
- H/L switch for the GPIOs to select “high effective or low effective”
- Power and Wakeup LEDs
- Power Supply – 12V DC via power barrel jack
- Dimensions – 186 x 106 x 33mm
- Weight – 362 grams
- Temperature Range – -10 to 75°C
Mekotronics will provide support for Android 14, Ubuntu, Debian, and Buildroot, and the system also supports AI frameworks such as TensorFlow, MXNet, Pytorch, and Caffe through the RKNPU2 toolkit like other recent Rockchip SoCs.
The Rockchip RK3576 fanless edge AI mini PC is available now, and the company told CNX Software that the sample price would be $140 for the default configuration (4GB/32GB). Additional information may be found on the product page. That’s the second hardware platform with RK3576 we’ve covered, the other being the Banana Pi BPI-M5 Pro single board computer.
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
This is awesome. The RK3576 is a 8nm SoC that can run with free software also on the GPU (not like the 8nm RK3588 that needs a closed source firmware to run the GPU).
This is a computer people could put on their desk and use is and still have all possible connectors that could be in need.
> The RK3576 is a 8nm SoC that can run with free software
While my parents taught me respect for religious believes I think you need to understand that your focusing on FOSS is a bit silly since the control ‘FOSS’ would have over hardware vanished over the decades anyway.
And GPU BLOBs or not are the least relevant part of this.
> would have over hardware vanished over the decades anyway.
Then simply tell how the solution for the future should look like. In those upcomming decades would the people then build their computers in some reverse engeneered FPGA’s to get a free as in freedom computer to be able to do whistleblowing like Snowden did?
QubesOS do not help when the Hardware is not free like Joanna Rutkowska explained.
You are talking about ‘religion’. I am talking about a free world where people like Snowden can blow the whistle and have no technical probelems to do so. Just provide the solution to all and i can go on to other topics i like working onto.
Then if it’s only a matter of drivers and firmware, how does that differ from x86 which just requires a regular mainline kernel ?
On x86 you cant run modern amd or nvidia gpus with free software. Also intel ARC require closed source software to run. On a desktop computer with pcie slot you have the freedom to get a nvidia 780ti and run it with fully free software.
But you also cant run free software on modern intel or amd cpus. Take a look at the reason why there is for example Libreboot and what hardware it supports.
Also RK358* require closed source raminit code that someone should reverse engeneer to make it possible to run a RK3582 (dont have a GPU for decoding) with fully free software.
Then it depends where the software starts for you. What matters to me is that I can boot on free software. I don’t care that much about the BIOS/SPL/early boot code as long as it lets me choose the kernel I want to boot and easily pass arguments to it, or call another boot loader. Regarding GPUs, I’m perfectly satisfied with the one included in the CPU so I don’t have to experience that old pain of drivers that break all the time or that prevent you from switching to a development kernel.
Regarding the low-level early init code such as the DDR training code, I don’t think we’ll get anything of acceptable quality level any time soon, just because there’s simply no need nor value for the hardware vendors to do that, and the perceived value for users is very low as well. What matters to the vast majority of users who care about software is that the pre-flashed boot loader is able to boot whatever ISO image they load into the device and that it works fine from this point.
“get anything of acceptable quality level any time soon”
You seem to have found issues on coreboot’s open raminit. What issues do you have? Could you fill up a bugrepport here? https://ticket.coreboot.org/issues
Huh ? no, I never even implied anything like that. I never decided to risk to sacrifice a machine to try it yet.
No, I’m just speaking about what’s found in SBCs nowadays. When you see that vendors themselves manage to fail to get their own training code to work and revert to lower frequency with the effect of lowering performance for their customers, you can just imagine what would come of from enthousiasts trying to reverse engineer some obscure code without any datasheet available!
I started to see DDR training fail on DDR3 with Atoms, and remember one board that lost some obscure setting that was not part of the configurable ones in the BIOS, and that suddenly failed to hot-reset with a certain type of RAM. Adding some large capacitors to the power lines was the only way to fix it with the provided hardware, and changing the memory sticks for all customers was the only solution to save them from going in-person in the data-center to power-cycle the machines after a software upgrade. When facing such extremely trick problems coming from vendors themselves, I don’t expect to get any reasonable quality code from a reverse-engineered solution.
> no, I never even implied anything like that.
You wrote quote: “get anything of acceptable quality level any time soon”
This is absolutely telling, that every single opensource raminit out there is not acceptable quality for you. Please fill up the bugreport on coreboot to help the developer get the quality level you expect.
> No, I’m just speaking about what’s found in SBCs nowadays.
This was not part of yor message before. On SBCs you in general dont find anything. A SBC is delivered in most cases with no software. So you are talking that the quality of “nothing” is not ok for you?
If you are talking about the available opensource raminit for SBCs, then please tell what problems you have with what opensource raminit on what SBC. You said “SBCs”. So you have issues on different SBCs with the different opensource raminits there.
You also have issues with the opensource raminit for RISC-V SBCs written by Daniel aka CyReVolt because you said that all opensource raminits dont have acceptable quality level. Please get into detail there.
I cant get any productive and usefull information out of your text and asking now the second time about the same to get hopefully now some usefull information that could help fixing the quality issues you have with opensource raminit on SBCs.
> Please fill up the bugreport on coreboot to help the developer get the quality level you expect
Huh. Very strange conversation. I just told you in the message you replied to that I do *not* use coreboot. I’m not going to spam their issue tracker saying “I don’t know what I’m posting about but user foss on cnx absolutely wants me to write garbage here”. That makes no sense.
> > No, I’m just speaking about what’s found in SBCs nowadays.
> This was not part of yor message before.
Sure it was, that’s the topic of the product presented here, I asked you who you compared such products with x86 and even mentioned “SPL/earlyboot” (sbc world) etc in addition to BIOS (x86).
> You also have issues with the opensource raminit for RISC-V SBCs written by Daniel aka CyReVolt because you said that all opensource raminits dont have acceptable quality level. Please get into detail there.
No I’m not having problems with products I don’t even know exist. Please stop ridiculing yourself by placing words in other people’s mouths, that’s nonsense and you’re not serving your cause.
Also it seems you can’t read because you’re claiming that I said “all opensource raminits blablalbla” while I said “I don’t think we’ll see … anytime soon”. See the difference between perception and accusations ? It’s precisely what you seem to be missing, and that might skew your world.
> I cant get any productive and usefull information out of your text (…)
Then better stop here because I can’t get anything useful from your ridiculous accusations that are not even substantiated by the text you’re replying to. Have a nice day.
> In those upcomming decades
…the people will be used to mass surveillance and be fine with it (they already are, see this fascination about all this creepy ‘AI’ crap). When Snowden revealed the bizarre surveillance practices of the US/NSA back then there were at least some spaces that were not completely wiretapped.
Today that’s different (Alexa, Cortana, Hey Google, Siri, whatever else 24/7 surveillance program exists) and stupid parents equip their children at an early age with ‘smartphones’ for the sole reason being able to track/surveil them. Those children when grown up won’t know a life w/o total surveillance.
Also the Snowden revelations led to agencies across the globe wanting to get the same powers the NSA has. And the population? Mostly doesn’t give a shit. Julian Assange is still in prison and his only ‘crime’ was exposing war crimes the US commits regularly. Are there mass demonstrations against that? Nope. Does this interfere with concepts like democracy (freedom of press and educated citizens as a prerequisit for this)? Sure.
The lack of FOSS GPU drivers is IMO rather irrelevant when looking at things from your perspective.
> the people will be used to mass surveillance
Yes, and like in china some people fight against that and put with this fight their life in danger to make everything bettet for everyone.
> and be fine with it (they already are)
No, stop trying to talk about all people at once. I am for example not fine with it and i know many other people that are also not fine with it.
> Today that’s different … Those children when grown up won’t know a life w/o total surveillance.
This is completely not true. Those children learn how to use Tor instead to get around the paranting-jail. They also learn the bad parts of what it could look like at worst.
> Snowden … And the population? Mostly doesn’t give a shit.
This is not true. The GDPR is strong based on Snowdens work. There is even a documentary how this have been worked out.
The population now have to live the benefits if they want or not.
> Julian Assange. Are there mass demonstrations against that?
Yes there are. When you go to those, then they would be +1.
> The lack of FOSS GPU drivers is IMO rather irrelevant when looking at things from your perspective.
No, free software is the ground work of a free world. You cant have free press when there is nothing free in hardware to write it on. And also there is no tor network to communicate about press work when there are no free as in freedom servers running tor.
You seem to have given up all the important things you learned about. Then i am surprised you even run sbc-bench and not given up all else. I am awaiting the values of the RK3576 in your results.md .
I know we’re getting off-topic, but…
> > Today that’s different … Those children when grown up won’t know a life w/o total surveillance.
> This is completely not true. Those children learn how to use Tor instead to get around the paranting-jail. They also learn the bad parts of what it could look like at worst.
That’s not what I’m seeing. What I’m seeing is kids these days who have pre-filled logins/mails/passwords in their browsers to instantly create accounts to whatever requests one to access one info or download any file, discord and chatgpt obviously being the first ones. And the sad part is that when you say “wait, do not click there, we can continue to search”, they say “no worries, look, it’s easy, it’s just one click away” and they don’t even understand what you were trying to protect them against nor why you’re preferring the complicated way. I think there is a point of equilibrium where their lives are made so comfortable with everything one click away even if this means giving all their lives to these platforms, that they’re not willing to listen to the implied risks for their future. We probably sound like the old ones saying “don’t use a car, it’s dangerous, you can die in an accident”, which was factually true but the risk vs benefits moved in favor of the majority accepting to take the risk and have the benefits.