Is a stable set up possible?

#1

16GB Rock5B, SPI flashed, booting off a Samsung 980 NVMe drive, headless and headed in different tests. Powered by the recommended 30W PSU or a 20V/80W feed coming from a dock. Results are all the same, regardless of what image I’m using ( pick one of Debian, Ubuntu, Armbian, Batocera, Reborn ), within minutes or hours, the board will lock up. Headless and USB based video ( in that case, delivered via the dock I have on my desk ) are longer lived and I got over 20 hours of uptime in headless or USB connected video. As soon as an HDMI cable is connected to the board to either augment the USB video or replace it, all bets are off. A simple window move could lock the machine. This morning, a short video caused a lockup.

The machine is set up to use the panfrost driver. Can the machine ever be stable while that driver is in use or do I have to disable the GPU and stick with llvmpipe? The board was intended as a desktop replacement and when it works, it does a fine job. There’s no point having a desktop that locks up randomly however. One potential use of it is as a Pi4 replacement in my arcade machine but without the GPU, even 1980s video games struggle at 60fps so right now, a Pi4 has better overall experience to this Rock5b.

Do I have an unlucky board or is this expected?

Thanks

#2

unlucky board or unlucky hardware combo. I can tell that my rock5 runs quite stable, I have not had a lockup et at in over a month. When I got the board I tried armbian and I had frequent stability issues with it.(mostly coming from the gpu in that case) I have to admit that my board is running headless though, as a jellyfin server. I am still waiting for my second board to do more every-day tests. From my time in this community I believe that there is a lot of potential for instability from different hardware combinations used, ranging from the power supply to more often the used nvme ssd. try without/with a different nvme ssd and see if theres a difference. there is a thread here about which nvmes work, and AFAIR(dont quote me on it) there were issues with samsung drives.

1 Like
#3

I’ll use the eMMC and remove the NVMe for a while to do more testing. Headless appears rock solid with stress testing and long running compiles. Thanks for the reply.

#4

I suggest you try the latest Radxa official image to make sure it’s not a hardware issue.

https://github.com/radxa-build/rock-5b/releases/download/20230401-0040/rock-5b_debian_bullseye_kde_b32.img.xz

#5

let us know how things turned out. Btw, I boot the rock off eMMC and use nvme for storage only

#6

Will do and I’ll flash that to the eMMC storage. Thanks.

#7

Running on that image now ( looks beautiful BTW ). I have a stress -c 4 running while connected to HDMI which in the past hasn’t lasted long. NVMe removed, booting and running off the eMMC. Powered via the Radxa 30W supply. Cat5 connected to the onboard ethernet.

#8

I use a Rock-5B v1.42 16MB, Samsung 970 EVO NVME, passive cooling through radxa shell (the green one) with a self-compiled Armbian Bullseye+Cinnamon as my main Desktop machine. Runs rock solid for days, never crashed so far. Powered by an ARGON40 5V (4A?) brick. Had instabilities only with PD adapters, after switched to the ARGON all is fine. Only one bad thing: USB3 backpower disturbs NVME boot, so I had to remove the USB3 HUB if I need to reboot, so far only for kernel update.

What is not working:

  • HDMI over HDMI->DVI-D Adapter (and I tried several). xrand reports the right resolutions for the monitor and select the right resolution, still no image, tried 2 monitors. Armbian 5.10.110-rockchip-rk3588 kernel.
  • gave up pancsf driver installation (soft rendering is fast enough for all my needs)
  • HW mouse-cursor “disapear” after DPMS, switchet it off and live with the flickering
  • radxa A8 card shows as two(?) WIFI adapters, but works as expected, BT works, although not extensively tested
  • boot from SD-Card (I must remove the NVME drive… take ages until I figured that out)

Before I installed on NVME, runs fine from 128GB EMMC.

But that said, I have two boards, one seems to be broken …

#9

So far so good. Pulled down some snaps ( Chromium to get to the latest version among others ), ran a video call or two and all looks good on the eMMC and software renderer. I’ll try imaging one of the NVMe later to see if it’s the image, the drive or just staying within the llvmpipe world. Thanks for the pointers. At this rate, this can become my daily driver.

1 Like
#10

Mark @jack reply as answer to let know others about that :slight_smile:

1 Like
#11

I installed Armbian to the Samsung NVMe and it’s been fine. The instability comes from the GPU drivers. Once I blacklisted the BT modules as per the install instructions, everything is working fine and this board is well on it’s way to be my daily driver.

2 Likes
#12

What is the specific meaning of “HW mouse-cursor “disapear” after DPMS” you mentioned? I don’t understand, I also had some problems on my machine. I used Ubuntu20.04 when I connected the HDMI, the monitor could display mouse-cursor normally, but I turn off the monitor off, again turn on, the mouse-cursor disappeared and the mouse could be clicked, but the cursor could not be seen. Do you have a good solution to your earlier problem?

#13

Exactly, this is the effect. Once the monitor goes into power safe state (DPMS), the hardware cursor is gone.
The fix is to switch the hw cursor off. However, after I updated the kernel from 5.10.110 to 5.10.160, the problem is gone.

#14

OK,Thanks. I will update to Ubuntu 22.04 LTS😄