SBC / Low Power boards general - Raspberry Pi and what not

  • 🐕 I am attempting to get the site runnning as fast as possible. If you are experiencing slow page load times, please report it.
I'd not bet on it ever being properly supported without huge caveats or support just being rugpulled in a few years. That's just the way the arm cookie crumbles. I'd wait instead. My incredibly well (for ARM circumstances) supported and ancient Allwinner A20 who doesn't even need firmware still can't do cpuidle. Nobody ever bothered implementing it. It'll always be easier to just buy an x86. It's just the way it is. This thing does have a BIOS, but that's not the whole story re: driver support. ARM SoCs rarely fail at the "gets Linux to boot" stage, the devil is later in the details.

Luggable News:
Almost done with the case, will go back to PCB design after. I decided to solder the switches directly on. Testing with different socket types (millmax/Kailh hotswap sockets) has shown there's a good chance keys may just straight fall out. This has opened the door to lower the entire design a little and also make it lighter, because a plate won't be strictly required.

I've used the Raxda Zero a lot recently, to the point that I kinda didn't really do the whole "internet" thing anymore. I spent a lot of time implementing things for emacs, which is basically my OS for that thing. It's quite fun. When you work with limitations like that system, you get a lot more creative in the coding department. The performance is a bit low though, even light gaming is somewhat of a hassle.
 
Remembered today that I had some kind of SBC laying around, way before I bought my Pi.
After some rummaging, I found the Odroid C1 I bought in 2015-ish and pulled it out of retirement to install direwolf to provide a network TNC for my APRS (ham radio thing) setup.
As the place I'm at is quite well covered by APRS repeaters, I didn't do a permanent APRS setup.

I use my Odroid C1 for:
  • Running daily and hourly scripts.
  • Seeding some torrents via qbitorrent-nox which is headless qbitorrent with a web interface. Network fuckery was a pain to deal with.
God bless this little thing, it's amazing.

I sold my old Raspberry Pi 2 to a fellow ham and still have a Raspberry Pi 3 I have no use for as of now (maybe a server, one day?).
 
Had no idea we even have a thread like this. My ARM autism will be well at home here. Speaking of, I'm kind of looking for opinions on a little schizo project I have going. Gonna repost this from the Linux thread:
Necroing this ancient post to say: not yet. I am saving up to get one. There have been staggering improvements since you made your post, it now runs an RK3588 (8-core with 4x 2.4 GHz Cortex-A76 + 4x 2.2 GHz Cortex-A55), up to 32 gigs LPDDR4, a standalone GPU and has an NVMe + 2.5" SSD slot. Not gonna lie, it looks sick. Pretty spicy price but imo probably worth the effort to make Moloch & Baal (Intel & AMD) tongue your asshole once you're rid of the PSP/IME shit. Not sure what is up with the TrustZone firmware on the Reform, AFAIK one of their head devs wrote somewhere that they are running some bespoke gimped version, with the only binary blobs being in DDR and GPU initialization but I can't find the exact post. Pretty cool stuff nonetheless. Guess this is about the best we can get until stronger RISC-V or POWER9 laptops become a thing.
Do any of you bajs have any experience fiddling around with old Chromebooks or CrowPi "laptops"? I kinda wanna try to get one for peanuts and see how efficient I can make it with Gentoo + mostly CLI apps. The CrowPi laptop kit seems really cool, though the price tag is a little spicy, especially since I want to get an M2 hat and NVME. If anyone's done stuff like this, any advice or thoughts are highly appreciated.
I'm considering getting an ARM computer instead of my current schizopad X230, mostly since I've come to appreciate ARM a good deal more after dealing with a bunch of IoT devices & running a Pi ZFS. The MNT Reform is pretty much the closest thing to what I'm looking for on the market right now, but the price tag is very salty. With shipping and all, getting a Reform with 32 gib ram and a 1TB NVME would run me up around 1800$, which is certainly not cheap. However, given the advent of the 16GiB Pi 5, I have started considering getting one of those instead, then running it through a hackjob CrowPi 2 with all of its pre-packaged internals scooped out to make room for a battery, fan and NVME hat. I've also considered getting some busted old laptop to use as a frame, but I am not confident enough in my skill with electronics to be able to adapt its screen, keyboard or battery. In total, the Pi-puter would run me up about 400$, so quite staggering price difference. Sure the CPU is a bit anemic and core count isn't the best, but either option beats out the X230 and scratches my firmware schizo itch (see IME/PSP vs ARM TrustZone). The idea is to use whatever I get as a daily driver with really minimalist ARM Gentoo running mostly CLI programs, which I have successfully done on something as slow as an old samsung chromebook plus, though it was a slog.

All in all the Reform looks way more appealing, but the price tag is making the choice really, really hard. The modularity, repairability and ease with which you can upgrade it is also very, very appealing. Then again, the Pi's CPU isn't stellar, but the Reform is more FOSS-friendly. Quite a tough sell either way.

Since you guys, especially @AmpleApricots, know a lot more than me in this department, any thoughts or advice would be very much appreciated.
 
Since you guys, especially @AmpleApricots, know a lot more than me in this department, any thoughts or advice would be very much appreciated.
I'll defer to Apricot man. The last ARM board I used was 8GiB Pi 4. That CrowPi2 appears heavily discounted. I remember people crowing about the price of setting up one of these Pi craptops.

 
  • Thunk-Provoking
Reactions: Ferryman
I'll defer to Apricot man. The last ARM board I used was 8GiB Pi 4. That CrowPi2 appears heavily discounted. I remember people crowing about the price of setting up one of these Pi craptops.

Interesting. The main reason I want this kinda thing is because A. it gives me an ARM machine with as little proprietary firmware on it as possible, B. it is modular, so I can reasonably upgrade it as necessary. Anything Intel or AMD automatically causes Probably gonna save up for a Reform though, since that does all of this and more.
 
I got a cheap deal for around 350 eurobucks on a 7940HS-based mini PC. What a powerhouse, really. Very low power for the performance it can dish out. My love quest of finding a small iGPU based system that genuinely can do everything I want to do is finally over. This thing doesn't even break a sweat with all the games I want to play. For ~200 more I could've gone with the successor model with a slightly faster NPU but the NPUs are genuinely not all that useful in both systems so meh. I suppose that'll take a few years and a few models down the road.

Interesting that the S922X Mali beats the Adreno. That's the main thing for me; the power-frugality of the smaller ARM SoCs and the fact that you can have a fully functional Linux machine in less than 9W or even 5W. When they try to do more powerful SoCs that are up to the desktop tasks of your average consumer (this usually means incredibly bloated webshit), I don't see the advantage over an often equally frugal x86 SoC. Apple's ARM offering kinda doesn't count in this calculation because Apple has full vertical integration and control. Even the x86 SoCs aren't always 100% supported in linux or work correctly. Sometimes the problem is even in the firmware. My quite frugal Thinkpad i5-7Y57 SoC for example can't reach the lower Pkg States for some reason. Most likely a firmware/hardware issue. There's just so many things to go wrong in the powersaving department if you don't have full control over the stack, you just can never know.

ARM computer
Usually with ARM, always consider when in doubt that things don't work because the drivers are not done. If you know the drivers are done, consider that they might be buggy or not actually support all features of the SoC, especially things like cpuidle or hibernation often plain don't work and these things can have a pretty dramatic effect on battery life. Somehow, kernel developers for ARM are really, really uninterested in making power saving functions work. Considering you said Schizopad: If you want to avoid firmware blobs, modern ARM is *not* the way to go, because it just requires blobs to even boot like the rest of them. Same with RISCV. Modern ARM SoCs also have embedded ARM cores that run proprietary software. If you want full control, you have to go old, e.g. older Allwiner offerings (A20, A64) are completely firmware-free and are basically in full hand of the OS, the FSF has a list. Of course though, a lot of these ancient ARM SoCs are weak. We're talking Pentium 3/4 performance in some tasks. Your X230 will look like a threadripper in comparsion. In general, you just cannot avoid proprietary hell without diving into extremely obscure and often quite slow things. Even completely ignoring all malovence (which certainly exists), this stuff is just too complex and expensive to develop and nobody's gonna give up their IP to the open source world for free. Embedded, propriertary OSes are now even a thing with $4 microcontrollers.

ARM is mostly interesting in the powersaving department, and there only in the lower end. Otherwise with modern ARM SoCs there are no differences to x86 in freedom IMO. As soon as a proprietary blob is involved, you effectively do not know anymore what runs on your system. Just the way it is. Even if there weren't blobs, you can't know if your CPU doesn't e.g. has some undocumented instruction to get instant ring 0 access to everything, like some old VIA x86 SoCs do. If you really want to go schizo and minimalistic, you need to go with old microcontrollers or retro computers. My Amiga has a proprietary OS too, but everything from the custom chipset to the OS is analyzed and documented so well down to the tiniest hardware quirk, it might as well be fully open source.

The MNT Reform seems really overpriced to what it is IMO. I also remember that there was quite a bit of drama about Rockchip kernel support and I'd really get all info what the current state there is before comitting to anything. With ARM you always also have to consider that contrary to x86, continued hardware support in the kernel can basically stop at any time for no reason. e.g. that one feature that is missing to make your SoC truly useful and is WIP might simply never end up in the kernel. Just the way it is with ARM, especially when newer SoCs are released and the one you use gets long in the tooth and people lose interest.

I'm currently building my own ARM luggable, a few posts above go into more detail. It's a fun project and my goal is a system with insane battery life, but I have no illusions that it's a particularily powerful computer that's useful for all tasks. If you have specific goals in mind, ARM is quite nice.
 
I got a cheap deal for around 350 eurobucks on a 7940HS-based mini PC. What a powerhouse, really. Very low power for the performance it can dish out. My love quest of finding a small iGPU based system that genuinely can do everything I want to do is finally over. This thing doesn't even break a sweat with all the games I want to play. For ~200 more I could've gone with the successor model with a slightly faster NPU but the NPUs are genuinely not all that useful in both systems so meh. I suppose that'll take a few years and a few models down the road.
I’ve got a laptop with a 7840HS, and apart from the outrageously short battery life the performance really is quite good. It’s only got an iGPU, but I’ve been bringing it along on business trips as a portable gaming computer and it really does handle itself well. Yeah I have to turn on FSR for any modern title, but on the relatively small screen (14”) that’s not a huge issue anyway. The CPU is strong enough to run Stellaris or Crusader Kings 3 at a reasonable clip, and games like Starfield or Cyberpunk are entirely playable on medium (and Cyberpunk doesn’t even look particularly bad with the graphics turned down).
 
ARM is quite nice
I do have an actual use case for proper ARM stuff insofar that a lot of my work includes IoT devices. Apart from that, its ideological purity. I know that proprietary firmware can't be avoided 100%, but the implementation thereof in ARM systems is much less egregious than in x86. From what I gather, the Reform Next only ships with a ~200kb DDR training blob as 100%, absolutely necessary proprietary firmware, with optional bits being WiFi/Bluetooth blobs that can be ignored through the use of an external bluetooth adapter.

I know that true 100% FOSSness is only doable on dinosaur systems and that's a fact. But I'll be honest, even with that in mind, the Reform is so cool to me in both concept and the philosophy of the people behind it that I'm still gunning to get one, inflated prices be damned.
 
~200kb DDR training blob
Are you sure? Not doubting you but genuinely curious, I was of the impression that the newer Rockchips need more blobs to run properly. Also I'm pretty sure the advanced onces have some "supervisor "ARM/RISC core tucked away somewhere. I get where you're coming from and I looked at the Reform myself before, but it simply did feel too expensive for me for what it ultimately is and the design wasn't too wild for me either. I actually think these people also make an ARM-based Amiga graphics card/acceleration board.

outrageously short battery life
Linux? I wonder how much of that might be the inability to reach lower package states because of broken firmware/broken Linux support. (I'm assuming linux here, correct me if wrong) I'm not sure if I can trust reporting I get out there of the SoC.

My trick is to limit framerates to 30 FPS. I think I started doing it many years ago because I used to own a GPU that'd sound like a jet engine if I didn't, and then it just stuck. It's not like I can't see the difference to higher framerates, I just don't really care. FPS just needs to be consistent for me, not necessarily high. Basically doubles what I can get out of any given GPU. I'm quite impressed of the games this thing can run, even more graphic intense titles like robocop and no man's sky run well, it does need a dash of AMD's FSR though, that is true.

Some people hate it, I don't really mind it either. It kinda has it's own vibe, reminds me of the various custom shaders for skyrim and such back in the day and the frame gain is quite dramatic, even on the quality setting. I think on a small, pixel dense screen the impact in picture quality is less big than on big, low DPI screens. Also I play such games very rarely to begin with and I'm just not gonna buy a 500€ graphics card that consumes 300W, heats the room I'm in and needs all kinds of huge components to merely play a video game. The games that need such an GPU would need to be a lot more impressive for that than they are. I mostly play indies and strategy games like the ones you named anyways.

It's also always worth it to play with the various graphic quality settings. Often there's either no difference in picture quality, no difference in framerate and sometimes both. I think it's because nobody writes their own graphics engines anymore. The game developers probably don't know themselves what these settings do in detail and how it impacts their particular game.
 
Linux? I wonder how much of that might be the inability to reach lower package states because of broken firmware/broken Linux support. (I'm assuming linux here, correct me if wrong) I'm not sure if I can trust reporting I get out there of the SoC.
Yes Linux. You may be right, but I see much better battery life in Linux than I did when I ran Windows on this thing. Particularly the sleep mode, Windows in sleep would drain the battery basically overnight, Linux sleeps all week.

Bear in mind my standard for laptops is the MacBook lineup. By Windows laptop standards I expect my 7840HS model is actually one of the longer-lasting ones. But it does die after a couple hours of use on battery, where the MacBook will last all day and still have some power to spare.
 
Are you sure? Not doubting you but genuinely curious, I was of the impression that the newer Rockchips need more blobs to run properly. Also I'm pretty sure the advanced onces have some "supervisor "ARM/RISC core tucked away somewhere. I get where you're coming from and I looked at the Reform myself before, but it simply did feel too expensive for me for what it ultimately is and the design wasn't too wild for me either. I actually think these people also make an ARM-based Amiga graphics card/acceleration board.
I stand corrected: you also need proprietary firmware for external displays through HDMI/eGP & power management. AFAIK the latter are a 20kib blob that only handles power and the former can be dodged by using VGA via a VGA to HDMI/eDP cable. And sadly, upon diving a little deeper, it also includes a ~20-100kib ARSIC controller, mostly for power management, but also with priveliged ring -1 access. That's quite sobering and somewhat disheartening, to be perfectly honest. Thank you for making me take a second glance at the Reform. I am strongly considering just sticking with my schizopads and working off a pi when I need to now. There goes my thunder *sigh*
 
The case fan (some custom job it looks like) of the Mini PC started failing after not even a month making some very horrid noises. To make it even worse, it's some custom jobbie, complete with custom heatsink for the NVMe drives. I tried disassembling and oiling it but it was already too late as that only helps for a time.

I could still send it back but since I don't think the next one will be any better and other Mini PCs look either like crap, probably have the same problems or just don't exist with the feature set at that price (it was a special offer), I decided on modifying this one with a different (much bigger) fan which also should be a lot quieter to begin with. Plenty of room in the case if you remove the heatsink and put heatsinks directly on the drive(s). I also tried skipping the case fan but that's a no-go, the RAM will get too hot and will report itself (via sensor reading) dangerously close to "critical" range. Man, DDR5 RAM does get hot. I did not know. I did try running it without that fan and interestingly, the computer is almost completely silent then even under heavy load, the blower-style fan for the CPU heatsink is VERY quiet. That whole custom heatsink assembly feels completely overengineered and just sticking a 80mm fan at the bottom would not only have been quieter while working just as well, it would have made replacements also a lot easier. I can't imagine this was a lot cheaper so I have no idea what drove them. Only if they use the same heatsink assembly for several models maybe, which granted, they probably do.

And if you get hungry, you can fry some eggs! My ancient AMD A4-5000 is still my fileserver. It would make sense, kind-of, to replace it with something newer but at this point I just really enjoy that that old computer still works and is useful in what it does. The power consumption difference isn't that dramatic either, especially since I wrote some custom stuff where it goes to sleep if doing nothing and wakes up by itself when required. (It also acts as a router for some other systems). I noticed people being very aware of their fileserver power consumption, yet they usually always run 24/7 even though in a home enviroment, that usually simply is not necessary.

I stand corrected: you also need proprietary firmware for external displays through HDMI/eGP & power management.

Yeah something like that. I remember HDMI output not being supported in mainline for rockchip not too long ago. Otherwise, these small ARM/RISC cores are just very useful from a design standpoint. They're even in your x86 SoCs, often running some proprietary RTOS or other to do some houskeeping tasks like you said, waking up the actual SoC. The modern SoC is a blackbox of IP blocks and proprietary blobs doing god knows what and I highly doubt any single given designer has a complete overview of it. The times where you could just get a 400-700 page book on that CPU in front of you and have a reasonable idea what it does are sadly long over.
 
I upgraded my Garage Pi to a Pi 500. I will say it's annoying that without surgery it can't take an on-board M.2 card. I opted not to do that and instead stuck a spare M.2 SATA in a USB enclosure. It's still pretty snappy especially compared to the Pi 400 it replaced. Now that I'm thinking about it I should have just stuck a Pi 5 in an enclosure so I could have the M.2 and maybe used a Pi or other small keyboard.

It's main purpose in life is to look up data sheets/schematics/other info and drive the (almost finshed) CNC router.
 
They're even in your x86 SoCs, often running some proprietary RTOS or other to do some houskeeping tasks like you said, waking up the actual SoC. The modern SoC is a blackbox of IP blocks and proprietary blobs doing god knows what and I highly doubt any single given designer has a complete overview of it. The times where you could just get a 400-700 page book on that CPU in front of you and have a reasonable idea what it does are sadly long over.
Orangepi rv2 just came out but still filled with blobs. Worst part is the GPU. Plus RISC-V is known to have security exploits. This one's from the Northwestern Polytech University of China and has been patched.
2024年4月,项目联合承研单位西北工业大学研究团队首次在RISC-V SonicBOOM处理器上挖掘出寄存器端口争用漏洞。该漏洞是国内漏洞数据库首次收录的RISC-V处理器设计上可远程利用的中危漏洞。本报告就该漏洞的机理及影响进行分析。
此次漏洞挖掘和验证工作由西北工业大学胡伟教授团队完成,该团队长期从事硬件设计安全验证、安全漏洞与恶意逻辑检测、密码应用安全、硬件安全设计自动化工具等方面的研究。
In April 2024, the research team of Northwestern Polytechnical University, a joint research unit of the project, first discovered a register port contention vulnerability on the RISC-V SonicBOOM processor. This vulnerability is the first medium-risk vulnerability in the RISC-V processor design that can be exploited remotely in the domestic vulnerability database. This report analyzes the mechanism and impact of the vulnerability.
This vulnerability mining and verification work was completed by Professor Hu Wei's team at Northwestern Polytechnical University, who has long been engaged in research on hardware design security verification, security vulnerabilities and malicious logic detection, cryptographic application security, and hardware security design automation tools.
If you really wanted to though, there's folks who do black market SBCs in China by borrowing fab lines when they're supposed be down. Not sure if they take Monero or just plain cash lmao. They're pretty hard to contact though (and lmao bulk orders only) since they escaped the great 2024 telegram dragnet op.
 
Back