Orion O6 Debug Party Invitation

[tkaiser]

the BIOS setting the 2.6 GHz limit

This does not seem to work, at least you were running at 2.5 GHz: https://browser.geekbench.com/v6/cpu/11703943.gb6

Isn’t there an UEFI setting to bring back the little cores? The changelog.txt contains “Remove little core enablement option in BIOS Setup.”

Asides that when trying to test different firmwares/kernels/whatever (in general: most probably different settings) I don’t understand why incapable tools are used that do not measure both RAM bandwidth and latency individually. Since depending on type of benchmark different memory performance might matter more than CPU ‘horsepower’.

Both of those HPL scores are still lower than a 4-core A76 Raspberry Pi 5…

As such: benchmarking gone wrong confirmed? With reliable benchmarks the O6 with all 12 cores active scores three times higher than an RPi 5B. So how should the low HPL score be explained other than something went wrong? IMO a benchmark is reliable when it is not that much affected by dependencies like software libs, OS and kernel version (7-zip’s internal benchmark being a pretty good example)

[tkaiser]

Yes for the one based on CIX’s SDK.

On https://dl.radxa.com/orion/o6/images/debian/ something appeared few hours ago: b6 version of your Debian image. Is there any changelog or any instructions (for example UEFI update necessary to get low(er) idle consumption, maybe improvements with RAM and CPU clockspeeds?)

[RadxaYuntian]

I just created the final EDK2 change as the kernel is approved. The official EDK2 binary for this release will be built after this commit is accepted in CIX’s tree.

Here is the changelog from CIX for 202502 and 202503 releases (we skipped 202502):

V202502

2.2 What Fixed
Can’t find audio codec during cold boot. 10% possibility;
System auto reboot after execute command in terminal: sudo poweroff;
Fan speed can’t be automatically optimized when SoC temperature changes;
Display enters UEFI stage without cix logo;
system printed “rcu: INFO: rcu_preempt detected stalls on CPUs/tasks” during reboot & xPU stress
test. around 2% possibility;
HDA: Speaker switch to HP and do STR, codec HP Out Mute on;
Some U-Disk lost connection during read/write.
2.3 What’s New
Support camera0 or camera1
Support camera0 & camera1 in parallel
support exFAT filesystem
Support ISO install with U-Disk
2.4 Known Issues
Base on Radxa O6 board v1.2, below are known issue, will be fixed in 202503 Monthly release:
totern and chromium playback simultaneously, totern playback has probability stop at rewind;
ISO install Debian, apt install smplayer, then open smplayer, can’t play the media;
ISP: user Gstreamer command line capture photo, short horizontal lines in the photo;
GPU Performance gap(15%) between different DisplayPort (DP) ports when using glmark2-es2-
wayland.
Suggest use C1 as display output for better graphics performance.

202503

1.2 What Fixed
On Debian12 setting, set power button behavior as suspend，press power button over 5 sec, when
press power button, system went in standy, after 4 sec shut down. unplug U-disk, system reboot.
GPU Performance gap(15%) between different DisplayPort (DP) ports when using glmark2-es2-
wayland.
Fan is in full speed after system reboot.
After poweroff, Type-A power leakage
Attached with AMD graphic card, power off but failed to shutdown
System suspend failed if ddr-size is larger than 32G
While attaching BT/WIFI module and dual gbe net, system can’t power off successfully
1.3 What’s New
GO update
NOE SDK update
1.4 Known Issues
Base on Radxa O6 board v1.2, below are known issue, will be fixed in next release
Installed noe compiler, at X86 PC, execute cixbuild cfg/xxx.cfg, can’t find libaipu_simulator_x2.so
lib.
ISP: user Gstreamer command line capture photo, short horizontal lines in the photo

[RadxaYuntian]

i think it is just a re-upload of the first

It was actually a different build from CIX to fix some issues in the initial version. Don’t remember which exactly though. It was a long mail thread.

[RadxaYuntian]

I’m only seeing 8 cores total in Ubuntu

SystemReady BIOS needs to support Windows, which does not support different types of cores on Arm, so they are disabled.

[meco]

@RadxaYuntian is this new edk also fixing PCIe power not being turned off on shutdown?

[Uli_Beling]

Can we then get a upgraded Bios for Linux with all the cores?
Are you fooling us again?
First not the tec spec. now not all cores for linux available, whats next?
Seems we have bought the most expensive paperweight.

[nyanmisaka]

Join the radxa Discord Server!

[RadxaYuntian]

@RadxaYuntian is this new edk also fixing PCIe power not being turned off on shutdown?

Should be fixed in this one.

Can we then get a upgraded Bios for Linux with all the cores?

Don’t use SystemReady if you want all cores.

[tkaiser]

SystemReady BIOS needs to support Windows, which does not support different types of cores on Arm, so they are disabled.

Which is a bit strange since Windows boots flawlessly on something like a Snapdragon 8 Gen3 which combines the very same A520/A720 as in P1 but adds also a single X4 core to the mix.

And while all cores on each SoC share same features/flags they do differ between both SoCs. Asides the flags they both share (fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm jscvt fcma lrcpc dcpop sha3 sm3 sm4 asimddp sha512 asimdfhm dit uscat ilrcpc flagm sb paca pacg dcpodp flagm2 frint i8mm bf16 dgh bti ecv afp) these flags are unique to each SoC:

• Snapdragon 8 Gen3: ssbs
• Cix P1: sve sve2 sveaes svepmull svebitperm svesha3 svesm4 svei8mm svebf16 mte mte3 wfxt

Or is this an ACPI vs. DT thing?

[Mario]

See:

Orion O6 Debug Party Invitation

Got the latest insider build of Windows 11 (27818.1000) to boot with all cores. It required a small kernel patch, though disabling the little cores would also work. These recent builds assume that all CPU cores support SPE, but it turns out that the A520s don’t, which leads to a fault when trying to access the missing CPU registers. [Screenshot%202025-03-25%20012614] WSL2 (Hyper-V) also works (IORT needs to be disabled in ACPI for now): [image]

This could probably be worked around by making core 0 an A520, but ultimately should be fixed in Windows as SPE is just an optional profiling feature.

Though, the SystemReady band test suite was run with only 8 cores as well, so I suspect there’s more to it and the system couldn’t be certified with all cores enabled.

[tkaiser]

This could probably be worked around by making core 0 an A520

As it’s the case wird all those Snapdragons Windows runs on

From the Windows perspective: has this any disadvantages the 1st CPU core being a slow one? Bad affinity settings maybe?

I lost a bit overview (and interest) and have no clue whether Cix has been asked or said to implement a firmware setting ‘adjusting’ the core order so core 0 could be set to the first A520 instead of core 11 from the fastest cluster…

[washley]

There’s a similar problem on linux with regard to PMU counters. It does detection on CPU 0 which, being a big core, has 30 event counters available. But the little cores only have 7, so you can kernel panic it by trying to use more than 7 on the little cores.

[geerlingguy]

I don’t understand why incapable tools are used that do not measure both RAM bandwidth and latency individually. Since depending on type of benchmark different memory performance might matter more than CPU ‘horsepower’.

I did run tinymembench again, as well as my full “standard suite”—results are being compiled here: https://github.com/geerlingguy/sbc-reviews/issues/62 (I have moved the earlier results into a comment and am re-filling the original issue with SystemReady results).

[incognito]

How about just installing the module?

sudo apt install r8168-dkms or maybe r8125-dkms

[geerlingguy]

I wasn’t able to do that through the graphical installer with the Live CD ISO, at least. It would bail out since I didn’t have a network connection.

[incognito]

This seems to be Debian’s fault then. Maybe you can use a WiFi dongle or USB ethernet as a workaround.

[geerlingguy]

I mean… that’s fine… but I’m just following the Radxa Wiki recommendations, and pointing out what’s not working: https://docs.radxa.com/en/orion/o6/bios/boot-debian-official-iso

If Debian requires additional work to get installed then it would be useful to include that information in the Wiki since that’s much more accessible to end users (instead of reading through 50 pages of this forum thread ).

[incognito]

I understand your point, but that it would not install would be true for any x86 system with the same ethernet chip. So while pushing Radxa to make their docs better is a nice goal, this is not something specific to them, and any manufacturer should include these instructions (and then, where do we stop? Should every manufacturer include instructions on how to install any distro on any pc?)

[geerlingguy]

Most X86 manufacturers don’t specifically call out one ‘tested’ / ‘golden’ OS that they recommend in their docs

This being a novel motherboard and a novel CPU, any and all help towards setting people on a happy path is appreciated.

I understand that the rtl8126 driver is not included. And I could easily pop it onto a USB flash drive with the driver on it during Debian setup, and install it that way (though it’s inconvenient and I’m lazy, which is why I wound up switching tracks to Ubuntu).

But the whole purpose of this thread (IMO) is to smooth over the 300 or so remaining bumps so people who buy these boards after we initial alpha/beta testers (I’m including current purchases in that group) will not have to live with all the same struggles we’ve faced.