Radxa 5b - where do I start? - 5 major problems?

I think many people here are assuming the Rock 5b is just a more powerful Raspberry Pi 4. It is not, nor is it being marketed as one. This board WILL NOT work perfectly out of the box. If you are new to Single Board Computers, and ESPECIALLY if you’re new to Linux, buy a Raspberry Pi instead. You will have a much more pleasant experience. If you are willing to tinker and learn, this board is great, but do not expect it to be plug and play. For me, this is probably the best SBC I’ve ever owned and I like it a lot more than my RPi 4 boards. For the average person who has never or rarely used an SBC, this board may not be what you’re looking for.

1 Like

Mine has been running consistently for several months now without any issues. Yes, NVMe boot doesn’t work out of the box, but it does work perfectly once you flash the SPI ROM. I’ve been using it and it’s worked reliably every time. Flashing the SPI ROM isn’t a difficult task for anyone with a little in-depth knowledge of Linux, by the way. Every SD card I’ve tried has worked without a problem. This is why we’re saying that this board is ok, because it is. I am certainly not being paid to say this, I am simply providing my own experience. For a beginner, this board isn’t and will never be great, which is fine for me, because I’m not a beginner. If you are (such as if you’ve only ever used an RPi), and you are not willing to do some research, learn, and spend a few hours tinkering, this board isn’t and will likely never be for you.

We have incorporated the suggested DMC tuning to our code. It should be available in our next ROCK 5B image release.


ah, well I honestly didn’t mind if it was a bit slower and lower idle power, it honestly is an amazing piece of hardware for low power I/O. Currently mine appears to sit at 1 watt idle which is kinda insane.

Also alot of problems seem to be getting fixed pretty fast with armbian anyway, the first image I used seemed to have a problem with network speeds in one direction (I wasn’t the only one I saw someone made a youtube video with the exact same speed limitation) but it got fixed the next time I tried a new image

The stuff about usb dongles it’s a bit weird you can’t always blame the soc for linux usb problems, especially with wireless. Not sure what problem you have there but mediatek stuff tends to work, if it doesn’t there’s always going for a kernel recompile with the mt76 driver from the openwrt github. I haven’t had any issue with SD cards either.

LOL! Armbian does close to nothing here. Quite the opposite, if you read ROCK 5B Debug Party Invitation again you will realize the Armbian guys only whining about HW details like USB PD and them not touching anything related to the 5.10 BSP kernel majority of Rock 5B users will have to use for many many more months or even years.

But since they rely on Radxa’s kernel repo since day one everything that gets fixed there is automatically inherited in Armbian after some time.

That was just an inappropriate PCIe ASPM default in Rockchip’s kernel. Reported as first issue/suggestion here in this active device review and fixed by Radxa within days. As such once Armbian updated their kernel packages the fix arrived there too.

Next issue Armbian will automatically inherit by using Radxa’s kernel and being from the very same aforementioned list of suggestions is this here. This will then also allow Orange Pi 5 running with Armbian to perform better since currently when using Armbian and not the crappy Xunlong OS images performance is harmed by the uptreshold parameter.

In fact, this is not fixed in Armbian, because it uses a different defconfig. I’m going to go ahead and file a fix…

Update: filed https://github.com/armbian/build/pull/4835
Update2: merged. See, you can easily contribute. :slight_smile:

Which is the Best OS for you to use in Rock 5B?

Since you’re familiar with Armbian’s build system you can do this to see how much I already contributed:

git log | grep -E -c "Author: ThomasKaiser|Author: Thomas Kaiser"

Most probably there’s a reason why I don’t want to be associated with ‘ignorance and stupidity’ any more?

No idea since me ignoring everything around GPU/NPU and so on. For (my) headless use cases it really doesn’t matter since all the OS images base on same BSP kernel anyway.

Armbian has some abandoned tweaks more or less by accident that are sometimes ‘better’ than Radxa’s defaults but that’s all. But these tweaks are abandoned, nobody at Armbian cares about this stuff and for example as such if the kernel/scheduler sends an I/O intensive task to cpu4 it will be fast with Armbian while when the scheduler sent it to cpu7 it will be slow. And they’re not going to fix this ever judging by status of this issue/task/whateverBS: https://armbian.atlassian.net/browse/AR-1262

And the most funny part about all of this is this excercise in ignorance: my suggested code that reflects the situation with hexa and octa core SoCs like RK3588 has already be copied into the Armbian script that does all these sorts of things: https://github.com/armbian/build/blob/master/packages/bsp/common/usr/lib/armbian/armbian-hardware-optimization#L78-L93 – But only for a fraction of devices, all the others use my old code that is sitting directly below. :rofl:

With Radxa’s OS images I/O will always be slow(er) since the usual amount of ignorance: https://github.com/radxa/kernel/commit/55f540ce97a3d19330abea8a0afc0052ab2644ef#commitcomment-79484235


For a layman, what are the steps to be implemented and by whom before we can see proper hw support?

The current Rockchip 5.10.110 BSP kernel provides ‘proper hw support’. You won’t get better ‘hw support’ from anywhere within the next 5 years since this BSP kernel (board support package) is the result of Rockchip software and hardware engineers working together to support their hardware as best as possible. In fact, ‘proper hw support’ will get much much worse the next months when running with an ‘upstream’ kernel.

The ‘layman TL;DR’ is: in general it’s fine.

This thread here started with 5 issues of which the 1st is just Youtube reviewers having no clue as almost always (‘Radxa 5b is slower than OrangePi5 and Mekotronic R58 despite same RK3588 CPU’) followed by the others that boil down to:

  • Radxa still hasn’t resolved USB PD powering issues (2 and 3)
  • not enabling drivers/modules in their kernel fork (4)
  • the board shipping in a shitty state if you want to boot from whatever storage (5): If they would’ve managed to flash the SPI NOR at the factory with something working to allow for booting from whatever storage as it’s default on x86 (where all of this is handled by a dystopian and horribly insecure monster called UEFI) then everybody would be fine and no more questions would be asked

I agree with the last point… simply cause I had random reboots on OpenFyde and once I did that spi clean flash magic happened. Anyway Kaiser… u seem to be very critic on all these boards u give us good advices but… why are you so critic about them… whats the pinacle of these boards to you? the pi foundation? are them the mighty ones who know all and make all things right?

Coming from a Windows user and as U could see im no Linux expert and all I wanted was a portable ChromeOS device being this a possibility once maybe openfyde bring us android apps for me Im fine… because even with all its bugs and issues for armbian on an optic user base webapp style… Im good and happy.

Idk whats everyone usage for these little boards… But I believe this is recent tech with lot to improve and I believe we can see new surprises in the future so Im happy with it.

Edit: I dont agree with some youtubers reviews… they tend to focus on lame titles like ‘the pi killer’ big stupidity right there… and then they lose the fun of reveiling and reviewing a nice product to basically… kill it.

1 Like

Do you understand what has happened here?

I’m really sorry I’ve made that last remark, I shouldn’t have said something like that, no need to repeat yourself twice. I have no idea what history you’ve got with the community, and who exactly you’re upset with the most (Rockchip/Radxa/Armbian/etc), and what exactly tipped you over the edge, but I’m sure your expertise could still help all of us a lot.
Again, I’m sorry for saying what I shouldn’t have.


weird I can’t think why it resolved for me then other than changing over to a switch instead of going nic to nic

If things you request can’t be addressed anytime soon, not withing this and next year, what do you expect? Who does like to generate or hear complains about all the time?

Perhaps you are been ignorant? I am sharing my private time every Wednesday to talk about anything regarding the project except complaining about technical issues. I am not working as a technical support and that is anyway done in public, answer can be provided by anyone. This is the deal we have with public: best effort support and you are the one that wants to rape that deal.

We mostly agree you are right in technical sense, yes, there are problems but project can only do what is possible. Just don’t make this another persons / projects problem. Sending a patch is already a problem as someone has focus into the work and review it.

No, we don’t do great engineering changes here. Nobody asked, nobody is paying for and nobody wants to invest time into the dead code. We are already generating too big loss, so making it bigger is not really smart. Perhaps this is why you have a feeling of being ignored, dear “customer”. Serving is limited and you can’t get all time slots there are.

Not everyone can do fancy YouTube or highly valuable QA testing with extreme tl;dr; and comparing what people you respect so little do for everyone. To generate one line fix, one can blow a week, while in that time you can easily generate 20 pages of technical documentation on testing and spam all forums with complains what doesn’t work and what you recommend to be fixed. Thank you, but most of problems are actually known, “just” time fixing them is a problem.

No matter how close to nothing we do, we invest 50-100 hours every day into open source. I know its not really much and certainly it looks nothing from your highly valuable position / perspective, but projects that invests 1% of our “nothing” into sales of out work, can easily looks to be on our level (some even convinced Radxa to pay them, while they paid us nothing). See, there are problems you simply ignore while accusing others of ignoring you. Just communicating with people (you would say idiots, because they are not on your technical level) to arrange things or to deal with internal technical issues I lost hours and hours every day. This is not possible to do alongside full time job.

You are right in technical sense - things are wrong and need fixing, but this require time / resource / interest. Even if all this matches, you can’t treat people like sh* and dictate people who are already working overtime on their day time jobs and private time to work more …

1 Like

Just to clarify, by support I mean for now I think it would be good to have mainline support where the sd card, network and usb ports work, I mean this is the standard initial goal for any soc going mainline so it can basically be run as a headless server. Seems like it’s almost there just the pcie 2.0 isn’t working so I assume that means no ethernet and no usb. I’m not a developer so don’t really know but as an example I bought a couple of rk3328 based boards and I have gotten great use out of them (great for libreelec or openwrt) but sometimes do find it odd at how progress stalls or DTS files don’t get updated/fixed or drivers like the usb3 phy don’t get mainlined.

Heh I have a celeron tablet i’m compiling stuff on, and it only has 8GB of ram and for some reason compiling clang caused it to go completely out of memory and grind to a halt. So I ran nbdkit memory on the rock 5b and made a 10G ramdisk to use as swap and clang compiled no problems haha. I had a go seeing how it handled x86 emulation but it seems like it would take a while to figure out other than just booting an existing disk/distro with qemu but qemu might be too slow.


It would be interesting to have somewhere a summary of what works/what doesn’t work. I read the forum from time to time, and I confess I’m confused about the real state of the RockPi5, which I’m thinking of buying one day.

1 Like

Well, I guess right now YMMV depending on what you what to get from this board.

  • The CPU is good, and having a real PCIe 3.0x4 might be a game changer if that’s what you’re after (e.g. for I/O-heavy tasks). As a headless server it’s quite a good one, in terms of performance per dollar, as well as performance per watt. With a Rock 5A or Orange Pi5, or anything else based on a RK3588S this applies too, except for the I/O part.
  • The GPU, on the other hand, is still waiting for the proper drivers, be it Panfrost or Mali, both are yet to deliver: it’s stuck at ~40% it’s potential, yet it can still deliver really good performance. Maybe in a a few months or years things will change, as it seems that Mali is finally releasing some proper Vulkan support for other chips, even if it’s closed-source; in the meantime, Mesa is working on its own PanVK which may take a while. So far, it’s a beast on a leash with a strict collar. It’s still good enough (as in, ~60fps) for GameCube/Dreamcast/Wii/PSP emulation, thanks to it’s sheer power, but it can do much better than that.
  • Speaking of graphics, you’ll really want to use Wayland and forget about X11, but I guess it’s a good thing anyway.
  • Mainline support is moving slowly (like, super-slowly), but that brings us back to the question of what you want to get out of that…
  • PD support is a real pain in the back, though can easily be solved with a dumb power supply.
  • Video encoding/decoding is a bit awkward at the moment, some things work nicely, e.g. you can actually get smooth 4K-8K HEVC/AV1 playback with HDR, but there’s a caveat of HDMI sound passthrough not quite being there for Linux in general, which is why I’m sticking with my Shield for media; otherwise, I’d probably switch in a heartbeat. Also, apparently GStreamer is the way, with FFMpeg being less of a priority (due to some licensing issues or something)…
  • NPU — can’t comment, haven’t tried. Once I get my hands on a 5A (btw, it’s “Rock 5”, not “Rock Pi 5”), I’ll probably try to bring up a Frigate server on that one.
  • Accessories aren’t there yet. There’s an acrylic open case (just two plain slabs), and a metal case that only supports passive cooling, plus some options for DIY 3D-printing, and that’s about it (KKSB is going to release a good metal case in 2-3 moths). Even the fan support is a bit awkward: despite having dedicated PWM pins on the board, you need a custom solution to make it work, e.g. a kernel modification that enables it, or a script/app from a third-party GitHub repo.
  • Speaking of accessories, WiFi/BT may give you a bit of a struggle. It works, mind you, but you may have to do a little dance to switch it on.

The bottom line is, pretty much all of the declared functionality is there (apart from Vulkan I’d say, though maybe it does work under Android; haven’t tried that one), but all of it has a huge “WIP” sign all over.

I’m personally happy with the board, as it does almost all I wanted it to do (except for the media part, I guess), but I still can’t shake off the feeling that it’s a powerful hardware being severely held back by the software part, and there’s no roadmap for when that’s going to be fixed.

Also, please treat this as my personal opinion (that of a non-educated casual user), as some more experienced folks may have their own strong one about some of the points.


Do you WANT vulkan Ginkage ??? Theres a thing… called Windows ! :see_no_evil: Nah im just joking Im in the same page. :face_with_hand_over_mouth:

ondemand tuning is added to rbuild. Thanks.