Radxa Orion O6 (Mini ITX)

[geerlingguy]

Basic information

• Board URL (official): https://radxa.com/products/orion/o6
• Board purchased from: ARACE Tech
• Board purchase date: December 18, 2024 (pre-order, 'shipping after the Spring Festival 2025', so early Feb)
• Board specs (as tested): Radxa Orion O6 16GB (32GB for the one I pre-ordered)
• Board price (as tested): $251 ($299.00 for the one I pre-ordered)

NOTE: I originally tested the 0.2.x firmware with the Debian 12 Device Tree configuration the first batches of the Orion O6 shipped with. Those test results are stored for posterity in this comment below. Because Radxa and Cix have released firmware which radically alters the performance characteristics of the board post-launch (April 2025), I am re-running all benchmarks and will list those results below.

Also, GitHub user @System64fumo is maintaining a list of all the features and tested hardware that works or doesn't work in mainline Linux currently: Orion O6 mainline support.

Linux/system information

# output of `screenfetch`
                          ./+o+-       jgeerling@orion-o6
                  yyyyy- -yyyyyy+      OS: Ubuntu 25.04 plucky
               ://+//////-yyyyyyo      Kernel: aarch64 Linux 6.14.0-15-generic
           .++ .:/++++++/-.+sss/`      Uptime: 7m
         .:++o:  /++++++++/:--:/-      Packages: 1589
        o:+o+:++.`..```.-/oo+++++/     Shell: dash
       .:+o:+o/.          `+sssoo+/    Disk: 9.8G / 234G (5%)
  .++/+:+oo+o:`             /sssooo.   CPU: ARM Cortex-A720 @ 8x 2.6GHz
 /+++//+:`oo+o               /::--:.   GPU: 
 \+/+o+++`o++o               ++////.   RAM: 1365MiB / 15229MiB
  .++.o+++oo+:`             /dddhhh.  
       .+.o+oo:.          `oddhhhh+   
        \+.++o+o``-````.:ohdhhhhh+    
         `:o+++ `ohhhhhhhhyo++os:     
           .o:`.syhhhhhhh/.oo++o`     
               /osyyyyyyo++ooo+++/    
                   ````` +oo+++o\:    
                          `oo++.    

# output of `uname -a`
Linux orion-o6 6.14.0-15-generic #15-Ubuntu SMP PREEMPT_DYNAMIC Sun Apr  6 14:37:51 UTC 2025 aarch64 aarch64 aarch64 GNU/Linux

Benchmark results

CPU

• Geekbench 6 (Ubuntu 25.04): (1228 single / 6009 multi - https://browser.geekbench.com/v6/cpu/11703943)
• Geekbench 6 (Windows 11): (1085 single / 5671 multi - https://browser.geekbench.com/v6/cpu/compare/11727381)
• 20.161 Gflops (0.76 Gflops/W) (geerlingguy/top500-benchmark result)

Power

• Idle power draw (at wall): 14.2 W
• Maximum simulated power draw (stress-ng --matrix 0): 24.9 W
• During Geekbench multicore benchmark: 27.3 W
• During top500 HPL benchmark: 26.4 W

Disk

Inland 256 GB PCIe Gen 3x4 NVMe SSD

Benchmark	Result
iozone 4K random read	51.67 MB/s
iozone 4K random write	215.64 MB/s
iozone 1M random read	1420.56 MB/s
iozone 1M random write	1452.81 MB/s
iozone 1M sequential read	1916.47 MB/s
iozone 1M sequential write	1683.79 MB/s

Network

iperf3 results:

• iperf3 -c $SERVER_IP: TODO Mbps
• iperf3 -c $SERVER_IP --reverse: TODO Mbps
• iperf3 -c $SERVER_IP --bidir: TODO Mbps up, TODO Mbps down

(Be sure to test all interfaces, noting any that are non-functional.)

GPU

glmark2

glmark2-es2 / glmark2-es2-wayland results:

NOTE: The 9.0.0 firmware and ACPI mode don't seem to properly expose the GPU for Ubuntu 25.04, at least. The iGPU will need later drivers and possibly other firmware to work correctly in Linux. See:

• https://forum.radxa.com/t/igpu-driver-status-and-dgpu-support/25128/16
• https://forum.radxa.com/t/orion-o6-debug-party-invitation/25054/409

1. Install glmark2-es2 with `sudo apt install -y glmark2-es2`
2. Run `glmark2-es2` (with `DISPLAY=:0` prepended if running over SSH)
3. Replace this block of text with the results.

vkmark

vkmark results:

1. Install vkmark with `sudo apt install -y vkmark`
2. Run `vkmark` (with `DISPLAY=:0` prepended if running over SSH)
3. Replace this block of text with the results.

Note: vkmark needs to be compiled from source on Debian 12 and earlier.

GravityMark

GravityMark results:

1. Download the latest version of GravityMark: https://gravitymark.tellusim.com
2. Run `chmod +x [downloaded_filename].run`
3. Run `sudo ./[downloaded_filename].run` and press `y` to accept the terms.
4. Open the link it prints, and run the Benchmark defaults, changing to 720p resolution and 50,000 asteroids.

Note: These benchmarks require an active display on the device. Not all devices may be able to run glmark2-es2, so in that case, make a note and move on!

Ollama

ollama LLM model inference results:

See: geerlingguy/ollama-benchmark#13

Memory

tinymembench results:

Click to expand memory benchmark result

tinymembench v0.4.10 (simple benchmark for memory throughput and latency)

==========================================================================
== Memory bandwidth tests                                               ==
==                                                                      ==
== Note 1: 1MB = 1000000 bytes                                          ==
== Note 2: Results for 'copy' tests show how many bytes can be          ==
==         copied per second (adding together read and writen           ==
==         bytes would have provided twice higher numbers)              ==
== Note 3: 2-pass copy means that we are using a small temporary buffer ==
==         to first fetch data into it, and only then write it to the   ==
==         destination (source -> L1 cache, L1 cache -> destination)    ==
== Note 4: If sample standard deviation exceeds 0.1%, it is shown in    ==
==         brackets                                                     ==
==========================================================================

 C copy backwards                                     :  11874.4 MB/s (2.1%)
 C copy backwards (32 byte blocks)                    :  11686.3 MB/s (1.4%)
 C copy backwards (64 byte blocks)                    :  11691.6 MB/s (1.4%)
 C copy                                               :  13160.3 MB/s (0.7%)
 C copy prefetched (32 bytes step)                    :  12196.7 MB/s (0.8%)
 C copy prefetched (64 bytes step)                    :  12154.2 MB/s (0.8%)
 C 2-pass copy                                        :  15616.0 MB/s (0.1%)
 C 2-pass copy prefetched (32 bytes step)             :  13435.1 MB/s (0.3%)
 C 2-pass copy prefetched (64 bytes step)             :  15813.3 MB/s (0.3%)
 C fill                                               :  40840.2 MB/s
 C fill (shuffle within 16 byte blocks)               :  40847.8 MB/s
 C fill (shuffle within 32 byte blocks)               :  40847.4 MB/s
 C fill (shuffle within 64 byte blocks)               :  40842.5 MB/s
 NEON 64x2 COPY                                       :  14618.1 MB/s (0.6%)
 NEON 64x2x4 COPY                                     :  15031.3 MB/s (0.2%)
 NEON 64x1x4_x2 COPY                                  :  14249.7 MB/s (0.2%)
 NEON 64x2 COPY prefetch x2                           :  14087.5 MB/s (0.5%)
 NEON 64x2x4 COPY prefetch x1                         :  15009.3 MB/s
 NEON 64x2 COPY prefetch x1                           :  15042.0 MB/s (0.2%)
 NEON 64x2x4 COPY prefetch x1                         :  15041.8 MB/s (0.2%)
 ---
 standard memcpy                                      :  11352.6 MB/s (1.3%)
 standard memset                                      :  48439.3 MB/s (0.9%)
 ---
 NEON LDP/STP copy                                    :  15467.9 MB/s (1.4%)
 NEON LDP/STP copy pldl2strm (32 bytes step)          :  13229.0 MB/s (0.2%)
 NEON LDP/STP copy pldl2strm (64 bytes step)          :  14515.6 MB/s (0.4%)
 NEON LDP/STP copy pldl1keep (32 bytes step)          :  15019.7 MB/s (0.2%)
 NEON LDP/STP copy pldl1keep (64 bytes step)          :  16052.8 MB/s (2.1%)
 NEON LD1/ST1 copy                                    :  15500.5 MB/s
 NEON STP fill                                        :  48873.6 MB/s (1.3%)
 NEON STNP fill                                       :  47780.1 MB/s (1.4%)
 ARM LDP/STP copy                                     :  14210.0 MB/s (0.2%)
 ARM STP fill                                         :  48795.8 MB/s (1.2%)
 ARM STNP fill                                        :  47495.6 MB/s (1.2%)

==========================================================================
== Memory latency test                                                  ==
==                                                                      ==
== Average time is measured for random memory accesses in the buffers   ==
== of different sizes. The larger is the buffer, the more significant   ==
== are relative contributions of TLB, L1/L2 cache misses and SDRAM      ==
== accesses. For extremely large buffer sizes we are expecting to see   ==
== page table walk with several requests to SDRAM for almost every      ==
== memory access (though 64MiB is not nearly large enough to experience ==
== this effect to its fullest).                                         ==
==                                                                      ==
== Note 1: All the numbers are representing extra time, which needs to  ==
==         be added to L1 cache latency. The cycle timings for L1 cache ==
==         latency can be usually found in the processor documentation. ==
== Note 2: Dual random read means that we are simultaneously performing ==
==         two independent memory accesses at a time. In the case if    ==
==         the memory subsystem can't handle multiple outstanding       ==
==         requests, dual random read has the same timings as two       ==
==         single reads performed one after another.                    ==
==========================================================================

block size : single random read / dual random read, [MADV_NOHUGEPAGE]
      1024 :    0.0 ns          /     0.0 ns 
      2048 :    0.0 ns          /     0.0 ns 
      4096 :    0.0 ns          /     0.0 ns 
      8192 :    0.0 ns          /     0.0 ns 
     16384 :    0.0 ns          /     0.0 ns 
     32768 :    0.0 ns          /     0.0 ns 
     65536 :    0.0 ns          /     0.0 ns 
    131072 :    1.0 ns          /     1.5 ns 
    262144 :    2.0 ns          /     2.8 ns 
    524288 :    5.8 ns          /     8.3 ns 
   1048576 :   21.4 ns          /    30.4 ns 
   2097152 :   27.5 ns          /    34.6 ns 
   4194304 :   32.9 ns          /    36.7 ns 
   8388608 :   38.3 ns          /    41.5 ns 
  16777216 :   48.5 ns          /    56.1 ns 
  33554432 :  123.8 ns          /   169.5 ns 
  67108864 :  185.7 ns          /   230.0 ns 

block size : single random read / dual random read, [MADV_HUGEPAGE]
      1024 :    0.0 ns          /     0.0 ns 
      2048 :    0.0 ns          /     0.0 ns 
      4096 :    0.0 ns          /     0.0 ns 
      8192 :    0.0 ns          /     0.0 ns 
     16384 :    0.0 ns          /     0.0 ns 
     32768 :    0.0 ns          /     0.0 ns 
     65536 :    0.0 ns          /     0.0 ns 
    131072 :    1.0 ns          /     1.5 ns 
    262144 :    1.5 ns          /     2.1 ns 
    524288 :    2.0 ns          /     2.7 ns 
   1048576 :   17.9 ns          /    26.5 ns 
   2097152 :   25.6 ns          /    32.8 ns 
   4194304 :   29.4 ns          /    34.8 ns 
   8388608 :   31.3 ns          /    35.5 ns 
  16777216 :   39.5 ns          /    46.8 ns 
  33554432 :  110.6 ns          /   155.1 ns 
  67108864 :  163.5 ns          /   206.7 ns 

sbc-bench results

https://0x0.st/8WAL.bin / ThomasKaiser/sbc-bench#115

Phoronix Test Suite

Results from pi-general-benchmark.sh:

• pts/encode-mp3: 8.724 sec
• pts/x264 4K: 11.95 fps
• pts/x264 1080p: 50.79 fps
• pts/phpbench: 611861
• pts/build-linux-kernel (defconfig): 776.221 sec

[ThomasKaiser]

Slightly/highly off-topic... but I thought I ask anyway since the question might interest some of your audience too (if you disagree simply mark my comment as off-topic).

To test 5GbE capabilties a TrendNet TEG-S762 just arrived (cheapest unmanaged passively-cooled switch with at least 2 5GbE ports I could find two days ago) and an RTL8126 based M.2 NIC is on its way (10 bucks taxes and shipping included, currently listed for a lot more again)

Since you recommended WisdPi equipment to me may I ask where you ordered and if from their store whether you got something like a commercial invoice? I'm thinking about ordering two RTL8157 based WP-UT5 you already reviewed to test the Onion O6 against Mac clients but need a 'real' invoice :)

[geerlingguy]

@ThomasKaiser - Their site uses Shopify, so it spits out the normal Shopify-style invoice (lists billing/customer address, items, shipping, tax—I presume it would also list VAT information in EU (not sure on that).

For my most recent order, it took 3 days from departing China to clear customs in the US, then four days to hit the regional mail facility here.

I also just noticed WisdPi is selling the WP-UT5 on Amazon now, too.

[MagicAndre1981]

I'm thinking about ordering two RTL8157 based WP-UT5 you already reviewed to test the Onion O6 against Mac clients but need a 'real' invoice :)

I bought a RTL8157 based adapter from Wavlink on German Amazon (your name may indicate you are German, too). On USB 3.0x Gen1 (5Gbit) you get 3.5GBit/s so around 400MB/s and full 5 GBit/s on USB 3.x Gen2 (10GBit) Port. But I use Windows devices, so I have no idea about Mac. You may need newer drivers.

[ThomasKaiser]

Shopify-style invoice (lists billing/customer address, items, shipping, tax—I presume it would also list VAT information in EU (not sure on that).

Since they listed 'included taxes' at checkout I simply gave it a try and edit this comment later when items and invoice arrived.

no idea about Mac. You may need newer drivers.

RTL8157 has pretty good support by stock macOS (confirmed to work out of the box with at least macOS 13-15). I checked the macOS driver package and it only contains two directories named '10.8' and '10.9-10.15' so this is only for ancient OS versions :)

[spx86]

Can you help test if this development board supports Arm CCA?
I also want to buy this model development board because I want to do some development work in Realm, so I need to know if orion o6 supports Arm CCA. i can hardly find any information about it!

[geerlingguy]

There's an Orion O6 Debug Party thread over on the Radxa community forum. I just got a special box in the mail today...

[ThomasKaiser]

'Mine' arrived as well:

Let's hope the aforementioned thread will bundle information needed at this point (and it doesn't happen in some Discord channels)

[hrw]

Do tests, share data etc.

Planning to order, but winter vacations first :D

[hrw]

Can you add BSA ACS to list of tests?

You run it from EFI Shell: bsa.efi for normal run + bsa.efi -v 1 for verbose one (and share both).

This checks for BSA compliance.

And my ArmCpuInfo as it shows better which cpu features are supported than running Linux and checking /proc/cpuinfo ;D

[mkesper]

If this thing gets mainline support it will be rocking!

[ThomasKaiser]

@geerlingguy @cnxsoft: latest sbc-bench v0.9.70 should now contain all adoptions/hacks needed for dealing with CD8180.

Especially the weird cluster setup is addressed and the mystery why cpu11 is not only faster than its cpu9 and cpu10 siblings from same cpufreq cluster but also faster than cpu0 with which the core shares the real CPU clockspeeds but not access to DRAM.

See commit comment for further details: ThomasKaiser/sbc-bench@6b0cd05#commitcomment-151945077

[geerlingguy]

@ThomasKaiser - Quite strange! I wonder if there could be any chip to chip variance here? Seems very weird to have one core behave with a measurable difference (especially wrt memory access). If that's as designed, what was the goal??

[ThomasKaiser]

If that's as designed, what was the goal??

Good question. I'm hoping @wtarreau as announced will soon start his detective work about the inner workings :)

(Though $dayjob might be in the way just as in my case right now)

[wtarreau]

Yeah I really want to, but indeed I didn't plan on receiving one so I haven't reserved time yet. But probably this week-end. I already downloaded the image to flash on it, and already have an SSD available.