Orion O6 Debug Party Invitation

wilsonyan · March 5, 2025, 8:56pm

I’m booting the latest linux kernel that I compile myself, I boot via UEFI/ACPI. The latest Linux kernel has support for the onboard network.

[    3.881667] r8169 0000:31:00.0: Adding to iommu group 0                                                                                                  
[    3.881757] r8169 0000:31:00.0: enabling device (0000 -> 0002)                                                                                           
[    3.899038] r8169 0000:31:00.0 eth0: RTL8126A, 00:xx:xx:xx:xx:xx, XID 649, IRQ 112                                                                       
[    3.899048] r8169 0000:31:00.0 eth0: jumbo features [frames: 9194 bytes, tx checksumming: ko]                                                            
[    3.900222] r8169 0000:01:00.0: Adding to iommu group 1                                                                                                  
[    3.900375] r8169 0000:01:00.0: enabling device (0000 -> 0002)                                                                                           
[    3.919504] r8169 0000:01:00.0 eth1: RTL8126A, xx:xx:xx:xx:xx:xx, XID 649, IRQ 113                                                                       
[    3.919517] r8169 0000:01:00.0 eth1: jumbo features [frames: 9194 bytes, tx checksumming: ko]                                                            
[    3.921965] r8169 0000:01:00.0 enp1s0: renamed from eth1                                                                                                 
[    3.929365] r8169 0000:31:00.0 enp49s0: renamed from eth0

I made a crappy screenshot of an example of how you can still run gui stuff using x2go. Basic stuff anyway, I doubt video works. I run Openwrt in systemd-container (systemd-nspawn), i’m not sure how wise that is overall security wise but it seems to work for me.

wilsonyan · March 5, 2025, 9:15pm

Unfortunately for me this is the culprit of stability problems and there may be a performance hit associated with disabling it. I saw they updated their EDK2 stuff within the last few days, I had a go at compiling it myself but it seems to require some ancient toolchain and I couldn’t get that to work either just running ‘make’

willy · March 5, 2025, 11:16pm

i could build the entire edk2 on an ubuntu 22.04 by scrupulously respecting the README that’s deep in the CIX directory. In short, you edit build_and_package.sh to set the board to O6,you set two variables and you run it.

diff --git a/Platform/CIX/Sky1/PackageTool/build_and_package.sh b/Platform/CIX/Sky1/PackageTool/build_and_package.sh
index 697b746..0791990 100755
--- a/Platform/CIX/Sky1/PackageTool/build_and_package.sh
+++ b/Platform/CIX/Sky1/PackageTool/build_and_package.sh
@@ -12,7 +12,7 @@ export WORKSPACE=$PWD
 export PATH_OUT="${WORKSPACE}/output"

 export ARM_TOOLCHAIN_ELF="gcc-arm-10.2-2020.11-x86_64-aarch64-none-elf"
-export UEFI_PROJECT="Merak"
+export UEFI_PROJECT="O6"
 export UEFI_PROJECT_PATH="Platform/CIX/Sky1"
 export GCC5_AARCH64_PREFIX="${WORKSPACE}/tools/gcc/${ARM_TOOLCHAIN_ELF}/bin/aarch64-none-elf-"
 export IASL_PREFIX="${WORKSPACE}/tools/acpica/generate/unix/bin/"
diff --git a/Platform/CIX/Sky1/PackageTool/pm_config/opp_config_custom.h b/Platform/CIX/Sky1/PackageTool/pm_config/opp_config_custom.h

Then:

$ export IASL_PREFIX="$PWD/tools/acpica/generate/unix/bin/"
$ export ARM_TOOLCHAIN_ELF="$PWD/tools/gcc/gcc-arm-10.2-2020.11-x86_64-aarch64-none-elf"
$ ./build_and_package.sh

For me it worked. For native tools I’m using gcc-11.4.0, binutils 2.38 and make 4.3 in case that matters.
I was careful to respect their directory setup though, it probably counts. Please just check the build_and_package.sh script and install the various repos relative to their $WORKPLACE variable.

wilsonyan · March 6, 2025, 1:13am

thanks, ok I got my image built it’s time to try flashing it

wilsonyan · March 7, 2025, 9:00am

It looks I was wrong about the cause of the problems it’s looking like segfaults occur when setting the cpu speed to 2.6 instead of 2.5 otherwise it’s all good.

This is a kernel I compiled basically you copy the Image and initramfs-linux.img to /boot and copy the 6.14.x directory to /lib/modules , there’s an example grub.cfg there that usually goes in /boot/grub/ but you’d need to change the UUID that’s in there if you want to use it

bill88t · March 13, 2025, 6:58pm

BredOS coming soon very soon near your local bakery! (Full gpu acceleration!!!)

Uli_Beling · March 14, 2025, 5:25pm

Very keen to test.
Thanks in advance
uli

Itxaka · March 14, 2025, 8:22pm

I remember seeing in December back when I ordered that this had a TPM support.

Welp, Im unable to see it now either on the page or on the docs anywhere.

On APCI mode, tpm is not detected and I cant see it on the boards as well. I truly remember it saying tpm (desoldered) so it was a tpm module that you could remove and replace by others…

Anyone remembers that or is able to see the module somehere, maybe on devicetree mode?

If they drop that from the public materials…they totally screwed us as its the sole reason we bougth it, arm board wtih TPM and secureboot…

Also, has anybody been able to set secureboot in ACPI mode, on setup mode? I cant for the life of me get it working, unless I manually enable it and add the keys…

hannes-campidell · March 14, 2025, 10:01pm

Hello,
Here desoldered means that it isn’t on the board but pads are available so out-of-the-box there is no TPM

cutterjohn · March 14, 2025, 10:54pm

Have I been stupid and missed power usage on this board? Or if not cans omeone at least least mention min/max wall power w/at least the base board?

THANK YOU!

[EDIT]

Information is for my PERSONAL USE, so don’t get excited… or exercised…

[/EDIT]

rev · March 15, 2025, 12:05pm

Chip is rated 30W tdp, no clue about package / board power use tho.

rev · March 15, 2025, 12:13pm

https://dl.radxa.com/orion/o6/hw/radxa_orion_o6_v1.20_schematic.pdf page 32

tkaiser · March 15, 2025, 10:16pm

North of 10W in idle.

Which is an indication of “something being wrong”. There’s something seriously broken and I wonder when something will be released that fixes this. Until then the O6 will be collecting dust.

wilsonyan · March 16, 2025, 12:21am

Can confirm, basically it’s minimum 10w, using all cpu cores its pretty much under 30w, i’m not sure how high it goes with all cpu + all gpu. If they somehow find a way to address the high idle this soc would be huge but even if they don’t i’d say this soc is still very strong to use right now for i/o / cpu tasks. The main competitor you’d think would be intel based stuff and they idle at at least 7w or so ? It definitely competes really well there with the 2 x 5g ethernet, working pcie and strong usb. This soc is pretty much good to go for stuff that doesnt require the ai / gpu, the latest kernel works with a few patches.

tkaiser · March 16, 2025, 10:19am

Nope, an idle system that is designed properly doesn’t need to consume energy. The only non-server Intel hardware around here idles at 2.5W.

But it’s not about comparing numbers, the +10W in idle is a simple indication that there’s something seriously wrong (most probably at the firmware layer which is all closed source so nobody other than Radxa or Cix could have a look and they both don’t seem to care).

Bill_Fleming · March 16, 2025, 7:41pm

The Realtek NICs probably consume a significant amount of idle power. They probably stay hot to the touch constantly. If they didn’t it would be the first 5/10gb NIC I’ve seen that doesn’t put out massive heat.

tkaiser · March 16, 2025, 8:47pm

I tested this a while ago on Rock 5B but don’t find the numbers right now.

So tested this again one time measuring with two RTL8126 and one time with three (got a M.2 variant for 10 bucks back then some months ago):

This is with two NICs, two SSD and some peripherals:

sh-5.2# Netio=powerbox-1/3 sbc-bench.sh -m
Power monitoring on socket 3 of powerbox-1 (Netio 4KF, FW v4.0.5, XML API v2.4, 232.44V @ 50.02Hz)

Cix P1/CD8180, Kernel: aarch64, Userland: arm64

CPU sysfs topology (clusters, cpufreq members, clockspeeds)
                 cpufreq   min    max
 CPU    cluster  policy   speed  speed   core type
  0        0        0      800    2600   Cortex-A720 / r0p1
  1        0        1      800    1800   Cortex-A520 / r0p1
  2        0        1      800    1800   Cortex-A520 / r0p1
  3        0        1      800    1800   Cortex-A520 / r0p1
  4        0        1      800    1800   Cortex-A520 / r0p1
  5        0        5      800    2300   Cortex-A720 / r0p1
  6        0        5      800    2300   Cortex-A720 / r0p1
  7        0        7      800    2200   Cortex-A720 / r0p1
  8        0        7      800    2200   Cortex-A720 / r0p1
  9        0        9      800    2500   Cortex-A720 / r0p1
 10        0        9      800    2500   Cortex-A720 / r0p1
 11        0        9      800    2500   Cortex-A720 / r0p1

Thermal source: /sys/devices/virtual/thermal/thermal_zone9/ (thermal-zone4)

Time       cpu0/cpu1/cpu5/cpu7/cpu9    load %cpu %sys %usr %nice %io %irq   Temp      mW
04:30:59:  800/ 800/1500/ 800/1500MHz  3.91   3%   0%   0%   0%   0%   1%  25.0°C    16200
04:31:04: 1800/1800/ 800/1500/1500MHz  3.92   1%   0%   0%   0%   0%   1%  24.0°C    15800
04:31:09:  800/1800/1200/1500/1800MHz  3.93   1%   0%   0%   0%   0%   1%  24.0°C    15720
04:31:14: 1800/1800/ 800/1800/1800MHz  3.93   1%   0%   0%   0%   0%   1%  24.0°C    15690
04:31:19: 1800/1800/1800/1500/1800MHz  3.94   1%   0%   0%   0%   0%   1%  24.0°C    15680

Now adding the 3rd same NIC via adapter to the PCIe slot:

Time       cpu0/cpu1/cpu5/cpu7/cpu9    load %cpu %sys %usr %nice %io %irq   Temp      mW
04:34:21:  800/1800/1500/1500/1800MHz  3.53   1%   0%   0%   0%   0%   1%  26.0°C    17980
04:34:26:  800/1800/ 800/1500/1800MHz  3.56   1%   0%   0%   0%   0%   1%  26.0°C    17880
04:34:32:  800/1800/ 800/ 800/1200MHz  3.60   1%   0%   0%   0%   0%   1%  26.0°C    17730
04:34:37:  800/1800/1500/1500/ 800MHz  3.79   1%   0%   0%   0%   0%   1%  26.0°C    17670
04:34:42: 1500/1800/ 800/1500/1800MHz  3.81   1%   0%   0%   0%   0%   1%  26.0°C    17600

As such less than 2W per NIC unconnected. We’re still talking about ~5W too much in idle.

cutterjohn · March 16, 2025, 9:02pm

true, my n97 hardkernel h4 idles at less than 10W(wall), but CAN peak when unlimited at ~30W (wall, just 1 stick DDR5, nvme, have not stress tested w/multiple spinning rust drives, although I do have one up and running as a nas, but it is set to default power limit which is IIRC ~13W… that unit is an h4+(has 4x SATA ports and 2x2.5Gbps ether, non plus misses 1x 2.5Gbps ether and has no SATA popped, it’s going to be a router w/a 4x5Gbps ether netcard 3 running opnsense, I had THOUGHT of using it as a lightweight desktop, but… reasons… too light for my use case mostly, but only barely… I could use it if I had to, but I don’t so nope. I also tried an rk3588 similar config and it was about the same but with worse linux/sw support.)

Whoops, sorry, I guess that this should have been a separate thread. Thanks for all the info though.

[EDIT]

hardkernel netcard 3 has only heatsinks on the drivers for their 5Gbps ether, if you look at their site, they do do VERY DETAILED analyses, power temp and throughput.

n97 the DDR5 SODIMM uses several Watts and nvmes basically use similar power to spinning rust drives.

RK3588 board w/nVME and 16GB LPDDR5(can’t remember if x or not offhand) idles at c. 5W and peaked c. 13-14W. I think H4(non plus) peaked at the ‘13W’ of 23W watt power, but I need to recheck those numbers as I may be confusing it w/‘unlocked’ power which IIRC is c. 30W, plus I need to do more UEFI updates as hardkernel has released a few since I did the test running/2w use, but Im pretty certain that it will be an opnsense box w/netcard 3 and an eMMC boot, wifi router -> AP hanging off the side of it.)

[/EDIT]

Salvador · March 17, 2025, 12:46pm

This is not longer a phone or tv box SoC soldered into a board and call it a “SBC”. A full fledged desktop-class SoC is going to consume like one. You just cant do magic, all the I/O at SoC level uses power.

What may be happening here is that there is some bug in the firmware and some component is not put into sleep when not used, but i dont expect it to match other ARM boards.

Something like an 5600G(still inferior in I/O by a lot) uses no less than 20W at idle.

dominik · March 17, 2025, 2:48pm

Of course it’s not anything like that, but still it could be just better and closed firmware would not help much,
On Rock 5B I could get 2-3x times more power consumption on idle just by using right kernel. Hopefully we would get needed sources and some optimizations, 5W idle is possible on i/o rich x86, so Why not here? One component can ruin whole setup.