Radxa Orion O6 (Mini ITX) #62
Comments
geerlingguy
changed the title
|
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 :) |
|
@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. |
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. |
Since they listed 'included taxes' at checkout I simply gave it a try and edit this comment later when items and invoice arrived.
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 :) |
|
Can you help test if this development board supports Arm CCA? |
|
There's an Orion O6 Debug Party thread over on the Radxa community forum. I just got a special box in the mail today...
|
|
'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) |
|
Do tests, share data etc. Planning to order, but winter vacations first :D |
|
Can you add BSA ACS to list of tests? You run it from EFI Shell: 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 |
|
If this thing gets mainline support it will be rocking! |
|
@geerlingguy @cnxsoft: latest Especially the weird cluster setup is addressed and the mystery why See commit comment for further details: ThomasKaiser/sbc-bench@6b0cd05#commitcomment-151945077 |
|
@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?? |
Good question. I'm hoping @wtarreau as announced will soon start his detective work about the inner workings :) (Though |
|
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. |
|
I was trying to get the OS loaded on an NVMe drive, but when I flashed the B3 version of the Debian desktop image, there's no .img file included in the Home folder as per the instructions: https://docs.radxa.com/en/orion/o6/getting-started/quick-start#3-write-the-image I was considering cloning the USB drive, but I have a 1TB USB drive and 256 GB NVMe installed, so that won't work either. I guess I can just try downloading the B3 img.gz file to the Live install USB drive, then flash from there... but I had assumed it would be included. Maybe a change from revision B1 to B3? |
|
I had the image on mine, it was the B3. I downloaded the "usb-preinstalled-something-b3" image, uncompressed it to a USB hard drive, the image was 20 GB. Then once booted, the NVME image was in the radxa user account. I just copied it over the /dev/nvme0n1 device, rebooted and it worked first time. Also, apparently the USB image and the NVME ones are not exactly the same. For example I noticed lscpu missing from the USB one while it was on the NVME one. Thus there's at least one difference. BTW be careful if you ever test openssl. There is an outdated version 3.0.1 in /usr/share/cix/something that is not correct, and unfortunately that path is before /usr/bin for the radxa user. Same for their libs, so you can quickly end up with tests reporting garbage numbers. And you can change the max CPU freq in the BIOS from 2.5 to 2.6, as indicated by Thomas (I did it, since better test it closer to the expected final specs). |
|
That's what I downloaded (https://dl.radxa.com/orion/o6/images/debian/orion-o6-debian12-preinstalled-desktop-b3.img.gz), and I flashed that download directly with Etcher, then booted off the USB stick. The home directory had no extra files in it (confirmed in Terminal and in the file browser, not just looking in Etcher)... so I just downloaded that entire file again on the Live booted image. Flashed it with Etcher, and it is running now. However, the fan was not going full blast in the Live CD boot from USB, nor during BIOS operations... but once I booted from the internal NVMe, the fan is now going at 100% continuously (and is quite loud!). Any way to force the fan to use PWM? CPU temp is reporting as 30°C so it's certainly not necessary to be going all out :D |
|
Yes, I got the same as well. I explained on the forum how I calmed it down, by echoing value 17 to pwm1 (at 16 it eventually stopped spinning). Found it, it's here: https://forum.radxa.com/t/orion-o6-debug-party-invitation/25054/126?u=willy Do not hesitate to read some contents there, as there's already quite a lot of data shared (CPU cores arrangement in clusters, RAM latency/BW etc). |
|
@wtarreau Thanks, and no worries, right after commenting, I finished reading through the debug party thread and found your comment on the PWM there too. I'll carry on with my testing here :) |
|
Power consumption during a top500 HPL run: 
|
|
I posted an initial pre-release blog post about the board, as I think it's important for Radxa to tone down some of the marketing hyperbole around the board, if they don't want the first wave of testers to lambast this board. It's good hardware, and I'd hate for it to be let down by early software issues. (And for the hundredth time, I wish Radxa would give more time between these 'debug parties' where tons of issues are found, and the public launch, when people who pre-ordered get boards that are pretty good but lacking in some essential features, or they have to do things like pin old software packages since updating would remove functionality they expect to have...). |
|
I usually take my time or do a preview (as opposed to a review) for new boards from Radxa. |
|
@cnxsoft - The tricky bit for that (in general) is when to actually do a full-on review, though. Do you do it when they start shipping boards to people who pre-ordered, or wait some time after that? From prior experience, I don't have a lot of confidence we'll be in a much different place in a few weeks, when these boards start shipping :( I want to be proven wrong, because I really like this board hardware, but as I say in the blog post, I think I need to temper my expectations! |
|
My method is quite basic. I review a few other boards first, and once I've done I switch to it. I try not to delay more than one month though. |
|
A couple updates over the past few weeks:
Also been monitoring the O6 Radxa forum, looks like one user has an RX 7600 running Skyrim at 4K, and another has Windows 11 up and running. |
|
The another is Mario known for Windows on R. And thanks for the link to the OpenBSD mailing list, great to see Jared McNeill in action (again, he did a lot of the ground work to boot any arm64 based distro on recent Rockchip platforms). |
|
@Civil - I'm currently running Ubuntu 25.04 with 9.0.0 BIOS, in ACPI mode. It has a newer kernel and later AMDGPU and Nvidia drivers without having to install backports or use Mainline Kernels to update. |
|
Testing CUDA cores and 'AI' inference on the 3080 Ti... Then I installed and compiled llama.cpp for arm64, compiling in CUDA support with Trying out some inference: That worked, and quite fast: Power consumption spiked from idle at 30W to full tilt 438W. More benchmarking information will be in this issue: geerlingguy/ollama-benchmark#13 |
The board power consumption becomes not even worth mentioning when the PCIe slot suddenly ramps up :D |
Sure, I just wanted to point out that the UEFI version you're running is solely meant to get this certification (by doing weird/nasty stuff like disabling the little cores for every OS to get silly Windows happy/booting) and there's another UEFI version on its way that is meant for end users then being able to also run any aarch64 OS version (maybe even Windows if Cix 'fixes' the CPU ordering). Asides disabling the A520 cores your testing also revealed that the SR UEFI version has other performance impacts. |
|
@ThomasKaiser - True... hopefully both will be fully supported though. If Nvidia ever decided to actually release the Windows on Arm drivers for their cards that I know exist... this board would immediately be the best value for Windows on Arm development (especially for games), and Qualcomm would be quickly in the dust for desktop use cases (IMHO). (Ampere is still a great option but those boards are kinda hulking and the power consumption is a bit higher too. But they have their uses for development.) |
|
Ampere boards are still quite expensive (count roughly $1000 for the naked board, and add at least as much for the CPU module). They're nice for enterprise usage but a bit too much for home. Also their cores are numerous but not that fast individually, so they only make sense for highly parallel workloads. |
|
@wtarreau - The Neoverse N1 cores in the Altra are about on par with the A720s, at least for most of the applications in my benchmarking. You just get a loooot more of them, and consistent performance between cores (instead of the odd hodgepodge of 'big' and 'medium' cores). I think of the Ampere chips as 'Threadripper' scale chips, versus the Cix being a low-end desktop or mobile chip. It is certainly a lot slower than Apple M-series if you just want a great desktop experience and a native Arm CPU. But if you want a good desktop experience and IO/Linux capabilities, this board is what I think will hit that sweet spot. The Amperes get some use in home environments too (just like a Threadripper), not everyone's needs are the same :D |
You can get a bundle with the CPU and motherboard for around $1500. |
Oh no, they're quite far from this, roughly half at same frequency. The N1 is exactly the same as an A76 (e.g. RK3588 or RPi5). They're arranged by pairs on L2 caches.
Yes for this I agree. I'm using daily a Q80-26 (80 cores at 2.6 GHz), "make -j" is a joy on this :-) Also their PCIe controller works very well, and the memory controller as well (I get something like 90% of theoretical BW, never seen this anywhere else, even on x86).
I'm precisely not convinced at all for a desktop due to each core not being that fast. Single-threaded workloads are slow. Typically the link phase of a build (particularly if you're using LTO) can be long. I guess that certain desktop applications can be slow (e.g. JS in a browser). Note that by "slow" I don't mean "ultra slow", Just not what you'd get on a low-end x86. But I agree that for anything multi-threaded/multi-process it's a bomb.
I really thought about it and figured that aside building kernels or bisecting haproxy I wouldn't really benefit from it. The Ampere One cores are said (by marketing) to be way more powerful. I have never tested them though, so I have no opinion on these. |
Yes but you just don't want to pick a low frequency CPU. Sadly they mix high core counts with high frequencies. There are 32 cores at 1.7 and 128 cores at 3.0/3.3. I think many of us would love to get a 16/32 at 3.3/3.5 instead of having to sacrifice the frequency that way to limit the TDP. In fact contrary to the x86 world you cannot choose a balance of cores count and frequency for a given TDP. On x86 you can find super high frequencies in 4/8 cores and super low ones in 64+ cores. That's what we're missing there. |
What I mean is when I'm doing my benchmarking of a 2.8 or 3.0 GHz Ampere Altra, and comparing that in end user applications and benchmarks to the 2.4 GHz (sometimes 2.5 GHz) A720s on this system, I see similar performance. Sometimes the Ampere wins, sometimes the Cix... it is dependent on how much it relies on cache, RAM access, etc. I don't spend as much time in the weeds of the theoretical performance, I just test how the cores perform with my workloads :) And there, this board is great... but does not live up to the potential I initially envisioned based on specs alone! |
You can save a bit. Altra m128-30 is usually sold for 6500-650$ used on ebay. Add ASRock altrad8ud there from Newegg (just mb is 829$) and you can get mb and CPU for 1.5k$ Though cheap CPUs comes and goes but every other month you can get one |
|
(totally off topic now, but) I really wish Ampere would sell CPUs DTC, or at least just have some available like that on NewEgg. I hate bundles lol. |
|
Also don't forget the RAM cost. The power of Altra comes from the many channels, you really need to fill them. On the Adlink model only 6 channels are populated, but a few other boards have the 8. |
I haven't tried buying from them, but apparently they're available on https://anafrashop.com/cpu-2?filtrManufacturer[]=2623 - though it looks like they're all on demand and not kept in stock. |
|
Sorry, what do you call "DTC" ? I couldn't figure what that means here. |
Direct To Consumer. |
Ddr4 rdimms are rather cheap nowadays if you are going for 16 or 32gb sticks. And you might save a bit if you'll go for ddr4-2933. It is slightly slower, but ampere doesn't seems to take advantage of extra speed. P.s. right now if you'll search for m128-30 on ebay you'll see a lot of used, working one shipping from us for 440$ |
|
Thanks Rebecca. I saw these CPUs for sale already but they were way too expensive for high frequencies, precisely due to the problem I mentioned above (i.e. lots of cores). |
Just wondering, did you try to compile the bench tools while specifing arm v9.2 march? it crossed my mind that the reason why some benchmarks are below what is expected is becuse it is running on generic arm v8 instructions. But i did not had time to test myself. Im not sure if i would even trust -march=native here, considering some linux distros and some tools misID it as armv8. |
|
I will have to try recompiling the top500 benchmark at least. I had hoped at least Ubuntu 25.04 being so new with 6.14 would identify Armv9 appropriately. |
It was only certified for a few distros (excluding Debian though). Others might have additional issues (for example the lack of RTL8126 driver).
By definition, SystemReady cannot be gated behind custom DTB and our custom ISOs. The test uses unmodified upstream ISOs, just like what an end user is supposed to use. We direly want to support SR + ACPI. Nobody wants to keep updating software forever. A standard compliant machine might finally make this a done deal. But until we have the source code, there is not much we can do. Cix is moving very fast on this though. We should have their initial ACPI BIOS code this month. |
Ampere is not too popular, so on the used market they tend to be available until the seller drops the price to ~400-600$ for M128-30. So if you don't mind used hw, just grab a CPU from eBay, and a brand new MB from Newegg. Cooler is a bit of a problem, but Noctua used to sell them if you ask their business support nicely, and Arctic has a mounting kit for their 4U-M cooler available for sale (that would be a cheaper option that you can grab from Amazon). |
|
The thing is, in my case since we already have a nice Q80-26, the gains to expect by upgrading to M128-30 are not that high to justify such a change. Plus the TDP of that beast is much higher (250W vs 150W) and ours just has the stock fan on it. That's why we never upgraded. But I agree with you on the used CPU approach, we did that already for many-cores EPYC and Xeons ;-) |
BLIS already detects the presence of SVE support just fine, but as I already told you elsewhere that is not necessarily a win in the HPL benchmark (given that the hardware vector size is still 128 bits, just the same as ASIMD/NEON, and that the workload is a boring FP64 one). I am somewhat skeptical that any other Armv9.2-A functionality would help much in that case (LSE/atomics could, but HPL is a multi-process instead of multi-threaded application, so I doubt it), unless @Shivansps has something specific in mind. |
|
Today I got an email from ARACE with the following, regarding my original order for a 32GB model with the AI Kit case:
I went over to the site to try a new order... 
From $299 for the original pre-order to $311.90... that I can stomach. From $311.90 to $1500.90? Not so much :( |
|
Posted a video and blog post this morning:
Still more testing to do, of course :P |
|
I heard today that tariffs could be almost dropped again for 90 days, maybe your should monitor the price variations in the forthcoming days. |
|
@wtarreau heh, tariff situation changes almost minute-by-minute now. Someone could probably make money being an arbitrator of when to actually 'cross the border' with a shipment of essential goods! |
|
Yeah I thought the same, but I just wanted to let you know that it's still worth doing a daily ping there ;-) |
and others
Basic information
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
Benchmark results
CPU
Power
stress-ng --matrix 0): 24.9 Wtop500HPL benchmark: 26.4 WDisk
Inland 256 GB PCIe Gen 3x4 NVMe SSD
Network
iperf3results:iperf3 -c $SERVER_IP: TODO Mbpsiperf3 -c $SERVER_IP --reverse: TODO Mbpsiperf3 -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-waylandresults:vkmark
vkmarkresults:GravityMark
GravityMark results:
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
ollamaLLM model inference results:See: geerlingguy/ollama-benchmark#13
Memory
tinymembenchresults:Click to expand memory benchmark result
sbc-benchresultshttps://0x0.st/8WAL.bin / ThomasKaiser/sbc-bench#115
Phoronix Test Suite
Results from pi-general-benchmark.sh:
The text was updated successfully, but these errors were encountered: