Same price (or more expensive!) as 10G SFP+ module 
Unless using this board as router or something like this - one two times faster port is just better. Also SFP+ is better than ethernet (uses less energy and easily can be transformed to ethernet).
This Orion board looks great. Absolutely loaded with features. And the case design is really nice too. Can’t wait to see some performance results and what kinds of AI workloads it will be used for.
@Peter.Wang some of the cixtech.com news articles mention verification on Linux 6.6 LTS and ACPI support. Can anyone from Radxa confirm those can boot without device tree with a generic arm64 image? This seems almost too good to be true even if that means some drivers don’t work yet
Yes, ACPI with 6.6 will land very soon. Currently we are booting with UEFI + DT.
This seems almost too good to be true even if that means some drivers don’t work yet
Finally we waited for this day.
BTW Thomas you guessed right, good catch 
Just noticed the availaibility of this new board very soon.
Compared to Rock 5 ITX, I am not sure it is a worth upgrade for NAS purposes, but NPU is far more powerful. At first, I would rather see this board in a compact mini PC or even in a laptop, as for NAS purposes, connectivity is a must and here, except if an extension board is used, there are no SATA ports. Sure, PCIe 4.0 is nice, even more if available on both M.2 M and E connectors.
OS support seems quite good, seamingly better than with other Radxa models, nice.
Many questions on myside:
Also, I am wondering…
Really, the cooling solution is always a weak point: from all the boards I purchased from Radxa, only the Rock 4B+ large heatsink was satisfying and I had to change the proposed stock cooling solution with a custom build, passive for 5 ITX and 3A, active for 5A/B. Otherwise, those SoCs/SBCs usually reach throttle conditions at full load under warm environments.
Well, to be honest, at time of purchase, Radxa did not propose any heatsink for Rock 5 ITX, so I choose a good heavy 1U tall heatsink and it is perfect, I just did not test the active one proposed a bit later.
Well, my questions and comments first aim at raising ideas and start the debate, not bury the model
A new product is always an exciting event, right?
Regarding performance gains vs RK3588, it’s not easy to find info, but based on various marketing documents found spread everywhere on ARM’s site, they seem to estimate that A720 is ~10% faster than A78, itself 7% faster than A77, which is 20% faster than A76 at the same frequency. That would mean roughly a 40% gain if all those numbers were true, and this doesn’t seem too unreasonable, or ~65% if we consider the frequency lift. And here we have twice the high-perf cores, so that leaves good hopes.
I agree that for a NAS, the Rock5 ITX truly shines. Here you have M.2 for the OS, and can use the PCIe slot for the storage (e.g. 4xNVME).
Regarding 5 vs 10Gbps onboard, it’s important to keep in mind that 10Gbps over RJ45 is extremely inefficient power-wise and heats a lot more than twice 5Gbps. It’s necessary to have a moderately large heat sink at 10G while here 5G is OK without. That’s probably the sweetest spot at the current state of technology. And if one wants 10G, some boards are available in various formats. I installed a 10G M.2 NIC in my Rock5 ITX and moved the storage to the second M.2 slot for example.
Regarding the choice of cooling, it’s always difficult for a vendor to provide one solution with a motherboard because users will install their board in different enclosures. Some will want a very flat one for a 1U enclosure and won’t care about noise nor lifetime. Others will want a large fan less device etc. IMHO it’s important to offer one reliable universal solution that serves as a starting point for users to qualify their needs and to later pick the right one for them. Ideally a vendor could suggest the smallest one they tested and a large fan less one, but even that would not always be fit for everyone.
And for the RAM upgrades, I think we would all like to have upgradable modules. But as was discussed before, SoCs made for LPDDR cannot support DIMM modules, and the newer LPCAMM modules are still hard to find and would likely inflate the price a lot. For the SoC, at this price tag, x86 boards also have soldered SoCs.
At least I do really appreciate that Radxa is taking risks to bring innovative products these days. The Rock5 ITX was a great and risky bet. That one probably even more, because once you get closer to the PC market, you suddenly start to be compared to entry-level x86 boards, and we know that it’s difficult to come close to x86 in terms of ease of use (starting from the boot). It seems to me that they’ve made the necessary efforts to succeed here, but it probably required a lot of investments again to get there.
Could anyone from Radxa share more information about the “AI PC kit” enclosure, specifically about the outer dimensions? I was thinking of mounting it in a 10" rack, but was unable to find any size specs. Will the enclosure be shipping the same time as the Orion O6?
Re: soldered RAM. Not sure if that’s actually going to be a SKU but 4GB seems to make no sense for this product right? Even 8 is pretty low but acceptable and Apple even understood this year if you market it as Ai it should start with 16GB
Take a look at this https://browser.geekbench.com/v6/cpu/5374715
It’s test for Exynos 2400 with Cortex A720 cores, it got about 1400 single core compared to max 900 for A78 on RK3588, so it’s about 1.5x faster compared to best results for rockchip (most are lower, so it may be 1.7x). Of course this is one and only GB test (mabe not optimised and done wright) but it’s at least something to compare.
With RK3588 there was not much pcie lines for disks and 10G. 4x pcie 3.0 + 1x pcie 2.0 is not that much.
Here we will have 14x pcie 4.0, that’s about 7x more bandwidth (as well as 2x 5Gbit already on board)
This is as easy as one sfp+ port that can be easily changed into 10G ethernet
are You able to saturate 10G port on RK3588?
I will be testing two 10G cards on Rock5, wondering now how much I can get out of them
Price for 32G model is really promising.
Luckily we got more and more professional BGA services if any reballing is needed. One of my Rock4 boards has changed SoC It’s not as easy process as with dimm modules, but still possible
RK3588 implements PCIe Gen3 x4 (or 2 x x2 or 4 x x1) and three times Gen2 x1.
P1/CP8180 has 16 Gen4 lanes.
In the actual Radxa implementations two PCIe lanes each are used for NICs (RTL8125B with Rock5 ITX, RTL8126 with the Orion here) and wrt ‘needed’ storage bandwidth in a NAS use case this is determined solely by the network connection. Orion O6 is limited to 2 x 5GbE using two PCIe lanes downgraded to Gen3 so all you ‘need’ storage-wise is another two Gen3 lanes to attach a bunch of spinning rust or whatever.
But ofc it’s great to have all those extra Gen4 lanes… even for the NAS use case since additional NICs might be attached to the E-key M.2 slot and then the requirement for storage bandwidth automagically increases as well.
Here some RK3588 experiments with multiple 10GbE NICs carried out by @willy back then: ROCK 5B Debug Party Invitation
No need to speculate about performance there’s an early geekbench result for the Cix P1: https://browser.geekbench.com/v6/cpu/8545345 (please keep in mind that it’s most likely been done on beta software)
and here compared to RK3588: https://browser.geekbench.com/v6/cpu/compare/8545345?baseline=4498568
@willy Thank You Willy for your interesting replies/comments. I agree with You, this product is exciting and can find its place, just maybe not in my case. For home NAS and light tasks, Rock5 ITX is the choice, this Orion 6 offers new options, but at a much higher price. Aside PCIe connectivity, the powerful NPU is probably a key point, for surveillance purposes in my case.
Finally, the only divergence is about cooling, and this is not really about Orion. Heat dissipation is really a critical problem with Rock 5A and is not satisfying with 5B. IMHO, Radxa can improve a lot on that side without difficulty. Again efficient passive cooling in a decently cooled case is always the right choice. A fan can fail, so active case cooling is the priority as it almost always offers redundant options.
Passive cooling of SoC with low TDP, for mini ITX, is mandatory! At the proposed price, this not a toy SBC anymore (…joke about the Rock 3A ‘toybrick’).
Right now, my interest in Orion might be about NPU and iGPU/GPU possibilities, not for (simple home) NAS. Well… I guess I will have to jump over the price trap and try it
Thanks to Radxa for this new board!
@dominik Yes, if You plan to setup a powerful workgroup/company NAS,10 Gpbs ethernet is probably nice. Willy and You have a better idea than me for sure, but this also increases the costs if You don’t need it. My home network relies on CAT6 (partially CAT 6A) but with 1 Gbit power efficient switches for the moment…
About RK3588, the performance cores are A76, not A78, but your 1.5x evaluation seems correct, so the 8 performance cores promise a potential 3x performance jump in multi-core benchmarks!
Let’s have a closer look at this new board… but arriving too late as a Christmas gift, too bad
I was about to try to order one, but I normally don’t put my credit card number on vendors sites, and while checking for other options I noticed that Jean-Luc of CNX software should receive one and review it. Thus I’ll first wait for his review to verify a few points. I’m particularly picky about mainline software support and the reality matching specs. Here I have good hopes but let’s see what it’s capable of. In my case it would become a dev board completing the LX2K and others that I’m regularly using to test my code. And I do think that this board would really make an excellent dev board! I’d probably buy 2 once convinced since I have an idea of offering one to someone I know
Not just ‘beta software’ but different clockings than the final product which affects both cpufreq and most probably also DRAM. Unfortunately Geekbench is such a poor benchmark that it does not even try to measure RAM latency/bandwidth to get an idea why the scores are as they are. And GB6 multi-core scores are pure nonsense anyway.
When referencing this Geekbench garbage you need to pay attention to details.
The single-score was measured on a Cortex-X4 (‘ARM implementer 65 architecture 8 variant 0 part 3458 revision 1’ -> X4 r0p1) at 2900 MHz -> check https://browser.geekbench.com/v6/cpu/5374715.gb6 for the measured cpu clockspeed. And obviously something went wrong since otherwise the score would be (much) higher. But Geekbench doesn’t check for e.g. background activity or other factors that ruin scores but simply uploads results whether they’re valid or not 
I was just heavily simplifying but you’re correct. Looking forward to an extensive review from you too when the board becomes available.
You can find more info about the AI PC Kits in this presentation:
https://dl.radxa.com/orion/o6/radxa_orion_o6_announcement_presentation_2024_12_18.pdf
Thank you so much for this PDF though it’s a bit weird that it contains no text (all text converted to paths) so no copy&paste possible. And while it answers a lot of questions I already wrote down to annoy you guys with one thing I really wonder is… why does Radxa call the SoC CD8180 when Cix Tech is calling it CP8180?
When zooming in to the presentation the board shots also show CD8180 etched onto the SoC’s surface. So what’s the real name?