4 ms·

As somebody who just wasted too much of his life trying to figure out a software issue on another RK3399 powered SBC, let me tell you that software documentatio

As somebody who just wasted too much of his life trying to figure out a software issue on another RK3399 powered SBC, let me tell you that software documentation/support is worth so much more than features or power. If the software that can run on these are cobbled together semi-proprietary chains of mishmashed crap, these boards will continue to be mediocre at best. Which! For the record... makes me very sad. There's a reason I bought the board I bought. On paper it seemed great. But the boot process was entirely undocumented and I ran into so many issues with kernel compatibility and featureset for their custom images.

garaetjjte 4y ago
>If the software that can run on these are cobbled together semi-proprietary chains of mishmashed crap Like these weirdo Broadcom chips in Raspberry Pis?
- simondotau 4y ago
  This would be a fair point if you were stuck with proprietary software. But this, this is a perfect example of how an open source ecosystem changes the physics. In an open source ecosystem, it doesn't matter how "weirdo" any chain of mishmashed crap might be if it's exactly the same weird crap as a billion other people. Scale means the fundamentals will have already been solved (or, if you're an early adopter, will soon be solved).
rjsw 4y ago
What doesn't work? Getting NetBSD to work well on this class of SBC hasn't been too hard. I probably need to try to get a RPi of some kind though.
marcodiego 4y ago
That is why I refuse to buy anything that is not supported by armbian.
megous 4y ago
> If the software that can run on these are cobbled together semi-proprietary chains of mishmashed crap, these boards will continue to be mediocre at best. What part of SW needed for RK3399 is that? To me pretty much all from bootloader to kernel is mainline in various projects and FOSS. So far it was fairly straightforward use mainline APIs trying to make cameras work via ISPs, RGA for image rotation, Hnatro for JPEG encoding in HW, etc. etc. Very few small bugs here and there that can be easily fixed. I've seen much worse. TRMs are available. Various other PDFs from Rockchip can be found, too (even some obscure ones about camera calibration, and stuff like that).
- Limb 4y ago
  I have a question regarding "mainline" that I hope someone could help clarify. What is the difference (in terms of coding, etc.) between mainlining support for a chip into Linux versus pinning it to a specific kernel revision? What makes it easy to bring a chip up on a specific kernel version,but not easy to submit that to the mainline kernel so future versions of the kernel are compatible? Are there significant changes between kernel versions that require someone to be an active maintainer for the specific chip, making sure changes to the kernel stay compatible with the chip? And thus most manufacturer don't care enough to keep supporting future versions once they get on specific kernel version working?
  - garaetjjte 4y ago
    To have code included in mainline you need to have reasonably sane code in the first place, and then invest some time working with upstream maintainers to address their concerns. On the other hand vendor code dumps are usually haphazardly put together garbage that nobody wants to touch.
  - forty 4y ago
    Rebasing (possibly shitty) patches from the SBC vendor to a recent kernel can be non trivial in my experience (as someone who have 0 knowledge of the Linux kernel code, so I'm a good baseline to define what's trivial ^^). And yes most vendors are will ship one working kernel source and forget about it forever right after :) even vendors that are not the worse depend on drivers that are provided by the SoC vendor, and might be stuck because rebasing a GPU driver might even be hard for them I guess.
CameronNemo 4y ago
This has not been my experience with rk3399 SBCs. The boot process is not only documented, but entirely blob free (from uboot to tf-a to the kernel). Recently I even got suspend to ram working on my rk3399 powered pinebook pro, which was one of the few items I had not been able to get working with a mainline stack. Hardware video decoding does require a recent kernel and gstreamer, but again it is fairly well documented and entirely mainline. Don't the RPI boards require blobs to boot the GPU, which boots the rest of the board? And hardware video decoding requires paying extra and using a proprietary blob? That said, this device (and the RPi 400) seem really wacky to me and I doubt I would ever buy one. I would not want to tell my grandparents "sorry printer does not work because arm" and the keyboard is way too sucky for me. Also no track point? That would be great for this form factor.
- cevn 4y ago
  Do you have suspend to ram with working audio when it comes back? That's my final hurdle.
  - CameronNemo 4y ago
    Hmm thanks for the challenge. I tested it and on resume, audio did not play. I had to restart pipewire and then it worked again.
- yjftsjthsd-h 4y ago
  > Don't the RPI boards require blobs to boot the GPU, which boots the rest of the board? And hardware video decoding requires paying extra and using a proprietary blob? Yes, Raspberry Pi hardware is stunningly awful and succeeds only through sheer ecosystem size, community dedication, and a company that's willing to ship something that works and try to make the thing reasonable in spite of having built it on completely unreasonable Hardware.
  - II2II 4y ago
    In other words: support. I think it's fairly well known around these parts that the primary appeal of the Raspberry Pi is support and brand recognition. Both are qualities that people are willing to pay for, may that be in dollars or lacking specifications.
Matsta 4y ago
From the reviews I've seen over the years for the Rk3399, even when the software side is all working, it seems to be a bit of a disappointing chip. My guess with a lot of RK3399 devices popping up is due to a couple of things: 1. The limited supply of budget SoCs due to the chip shortage and 2. Rockchip trying to clear stock for when RK3588 boards finally start to ship. Of course, we'll be back to square one on the software part. I do see the Jasper Lake mini PCs getting cheaper and cheaper, and I see that as more of a competitor to SBCs as there doesn't seem to be as many stock issues compared to Rpi4's and by the time you buy a Raspberry Pi, SD card, heatsink and power supply it does end up being not much cheaper.
bsder 4y ago
> let me tell you that software documentation/support is worth so much more than features or power. This is precisely why I left the RPi ecosystem for the Beaglebone ecosystem. RPi is great for beginners. The moment you leave that level--the support is trash. The BBB is tougher for beginners. However, everything is documented in datasheets like you expect. When you need to step up and out of the ecosystem, you can.

garaetjjte 4y ago

>If the software that can run on these are cobbled together semi-proprietary chains of mishmashed crap Like these weirdo Broadcom chips in Raspberry Pis?

simondotau 4y ago
This would be a fair point if you were stuck with proprietary software. But this, this is a perfect example of how an open source ecosystem changes the physics. In an open source ecosystem, it doesn't matter how "weirdo" any chain of mishmashed crap might be if it's exactly the same weird crap as a billion other people. Scale means the fundamentals will have already been solved (or, if you're an early adopter, will soon be solved).

rjsw 4y ago

What doesn't work? Getting NetBSD to work well on this class of SBC hasn't been too hard. I probably need to try to get a RPi of some kind though.

marcodiego 4y ago

That is why I refuse to buy anything that is not supported by armbian.

megous 4y ago

> If the software that can run on these are cobbled together semi-proprietary chains of mishmashed crap, these boards will continue to be mediocre at best. What part of SW needed for RK3399 is that? To me pretty much all from bootloader to kernel is mainline in various projects and FOSS. So far it was fairly straightforward use mainline APIs trying to make cameras work via ISPs, RGA for image rotation, Hnatro for JPEG encoding in HW, etc. etc. Very few small bugs here and there that can be easily fixed. I've seen much worse. TRMs are available. Various other PDFs from Rockchip can be found, too (even some obscure ones about camera calibration, and stuff like that).

Limb 4y ago
I have a question regarding "mainline" that I hope someone could help clarify. What is the difference (in terms of coding, etc.) between mainlining support for a chip into Linux versus pinning it to a specific kernel revision? What makes it easy to bring a chip up on a specific kernel version,but not easy to submit that to the mainline kernel so future versions of the kernel are compatible? Are there significant changes between kernel versions that require someone to be an active maintainer for the specific chip, making sure changes to the kernel stay compatible with the chip? And thus most manufacturer don't care enough to keep supporting future versions once they get on specific kernel version working?
- garaetjjte 4y ago
  To have code included in mainline you need to have reasonably sane code in the first place, and then invest some time working with upstream maintainers to address their concerns. On the other hand vendor code dumps are usually haphazardly put together garbage that nobody wants to touch.
- forty 4y ago
  Rebasing (possibly shitty) patches from the SBC vendor to a recent kernel can be non trivial in my experience (as someone who have 0 knowledge of the Linux kernel code, so I'm a good baseline to define what's trivial ^^). And yes most vendors are will ship one working kernel source and forget about it forever right after :) even vendors that are not the worse depend on drivers that are provided by the SoC vendor, and might be stuck because rebasing a GPU driver might even be hard for them I guess.

CameronNemo 4y ago

This has not been my experience with rk3399 SBCs. The boot process is not only documented, but entirely blob free (from uboot to tf-a to the kernel). Recently I even got suspend to ram working on my rk3399 powered pinebook pro, which was one of the few items I had not been able to get working with a mainline stack. Hardware video decoding does require a recent kernel and gstreamer, but again it is fairly well documented and entirely mainline. Don't the RPI boards require blobs to boot the GPU, which boots the rest of the board? And hardware video decoding requires paying extra and using a proprietary blob? That said, this device (and the RPi 400) seem really wacky to me and I doubt I would ever buy one. I would not want to tell my grandparents "sorry printer does not work because arm" and the keyboard is way too sucky for me. Also no track point? That would be great for this form factor.

cevn 4y ago
Do you have suspend to ram with working audio when it comes back? That's my final hurdle.
- CameronNemo 4y ago
  Hmm thanks for the challenge. I tested it and on resume, audio did not play. I had to restart pipewire and then it worked again.
yjftsjthsd-h 4y ago
> Don't the RPI boards require blobs to boot the GPU, which boots the rest of the board? And hardware video decoding requires paying extra and using a proprietary blob? Yes, Raspberry Pi hardware is stunningly awful and succeeds only through sheer ecosystem size, community dedication, and a company that's willing to ship something that works and try to make the thing reasonable in spite of having built it on completely unreasonable Hardware.
- II2II 4y ago
  In other words: support. I think it's fairly well known around these parts that the primary appeal of the Raspberry Pi is support and brand recognition. Both are qualities that people are willing to pay for, may that be in dollars or lacking specifications.

Matsta 4y ago

From the reviews I've seen over the years for the Rk3399, even when the software side is all working, it seems to be a bit of a disappointing chip. My guess with a lot of RK3399 devices popping up is due to a couple of things: 1. The limited supply of budget SoCs due to the chip shortage and 2. Rockchip trying to clear stock for when RK3588 boards finally start to ship. Of course, we'll be back to square one on the software part. I do see the Jasper Lake mini PCs getting cheaper and cheaper, and I see that as more of a competitor to SBCs as there doesn't seem to be as many stock issues compared to Rpi4's and by the time you buy a Raspberry Pi, SD card, heatsink and power supply it does end up being not much cheaper.

bsder 4y ago

> let me tell you that software documentation/support is worth so much more than features or power. This is precisely why I left the RPi ecosystem for the Beaglebone ecosystem. RPi is great for beginners. The moment you leave that level--the support is trash. The BBB is tougher for beginners. However, everything is documented in datasheets like you expect. When you need to step up and out of the ecosystem, you can.