A mini-ITX variant with slightly modified IO?


I was wondering: is there any chance of a Rock 5 ITX variant that has the following changes:

  1. four PCIe 3.0 lanes going to the M.2 M-key, like the Rock 5B, rather than only two lanes
  2. instead of a PCIe SATA controller feeding the onboard SATA ports could we have a SATA port multiplier, like a JMB575 connected to one of the RK3588’s SATA/PCIe 2.1 combo interfaces (thus freeing up two 3.0 lanes to go to the M-key above)
  3. instead of dual 2.5GigE controllers could we instead use the two GMACs on the RK3588
  4. add a traditional (desktop style) PCIe x1 slot connected to one of the RK3588’s SATA/PCIe 2.1 combo interfaces

This would fit my use case of a desktop and media station that has a PCIe tuner card installed alongside a fast SSD with some SATA hard disks for bulk storage and a SATA optical drive. I may be the only one interested in such a board though :slight_smile:

Something like this:

Thanks for the suggestion, I think the PCIe connector makes sense here. But others? I don’t think people want SATA PM, which has bad RAID performance. ROCK 5 ITX is finalized and will be available soon. Maybe we will cover all your requirements on the ROCK 6 ITX.

1 Like

Thanks for responding. Great! I think a real PCIe x1 slot would be excellent :+1:
You’re probably right about the SATA PM. Maybe a single native SATA port (or two if you use the existing E-key to SATA adapter) would be better. Then if you want a PM you can add it yourself.

In any case, I’ll keep my fingers crossed for the ROCK 6 ITX.

1 Like

Talking about ITX, seems like the CM5 IO board doesn’t use a standardized form factor, would you consider making the future CM6 IO Board in ITX form factor? Since the IO board is an application board, I think that the dimension of the IO board isn’t really a concern, and adding more IOs would probably be beneficial. The standardized ITX form factor may fit more use cases out of the box so instead of creating another carrier board, they could just get a CM6 with an ITX IO Board.

And for Rock 6 ITX, if the Rock 6 overall performance will be similar to 8CX Gen3 with PCIe 4.0, is there any chance it will support user-replaceable SODIMM or DIMM RAM just like other PC ITX boards? It looks like we can get up to 96GB RAM (48GB x2) with SODIMM DDR5.

Two Gmac‘s? I think most people would prefer the 2.5GBe port. Also 2x 3x2 makes sense as you can have more than enough ssd speed but also use cards like JMB585 in M.2 or 10GBe.

CM5 IO is half mini ITX, the purpose of CM5IO is to make a simple and low cost evaluation board for CM5. Well, CM6 will definitely somewhat compatible with CM5 but I am sure there will be new features that we want to add, so there will be new CM6 IO board for sure.

For ROCK 6 ITX, I think we are evaluating the LPCAMM2, which is higher frequency than DIMM/SODIMM so more dram bandwidth, which is important for AI.


I guess I’m in the minority that doesn’t own any 2.5GigE equipment :wink:

Given that the ROCK 5B has four PCIe 3.0 lanes going to the M.2 slot the ITX having only two lanes feels like a downgrade. But that could just be me again. For a use case like mine I’d rather lose the two 2.5GigE ports and turn those into two native SATA ports coming straight from the RK3588 to allow four gen3 lanes to the SSD. But it is all tradeoffs…


The eternal problem with PCIe lanes is that everyone has a different usage, and while 4 PCIe 3 lanes are sufficient for everyone in terms of aggregated bandwidth, they’re not easily distributed among various devices. A better approach could be to install a PCIe bridge to redistribute these lanes to multiple controllers, including a PCIe x4 one. That’s what has been done in servers for decades to please everyone, and here it would allow those who prefer a large PCIe connectors, those who want a SATA RAID controller, those who want M2 etc to be served. I just have no idea how much a Gen3 bridge exploding 4 lanes to 3x4 can cost.

1 Like

All features are now official, and I own a v1.11 sample to confirm.

Indeed, this ITX flavor divides the available 4 PCIe 3.0 lanes into 2 sets of 2 x PCIe 3.0, one for the onboard ASM1164 SATA controller, another for the M.2 M-key slot, each getting roughly 2 GB/s. This seems quite flexible to me, unless more than 4 SATA drives are desired, as it is the limit of the ASM1164 chip.

Using JMB585 or ASM1166 would have granted 5 to 6 SATA connections and the bandwidth offered by the two lanes are more than enough for RAID purposes, as the transfer is then limited by the CPU or the 2x2.5 Gbps Ethernet, so even if SSDs are used instead of HDDs.

Also, unless 10 GbE is desired, I think the 2 PCIe 3.0 lanes on the M.2 M-key are just satisfying for any other use, including a regular PCI/PCIe slot converter module. I will use a refurbished NVMe drive on this slot, for the main OS.

If WiFi is not required, the single PCIe 2 lane on the M.2 E-key is another extension option.

This really is a nice board for small NAS and surveillance purposes, quite well balanced and not too expensive.
To me, it is not a desktop replacement board, but a swiss knife for a compact server. I just regret it was not available one or two years ago, when I built my current NAS. Not too late however, as I decided to stop cloud storage for home and rely on two NAS (main + sync) and regular backups on cheap drives.

1 Like

One reminder, if you don’t use M.2 WiFi, you can use M.2 to SATA adapter to enable SATA for it.


I agree with your analysis, it’s also exactly the type of board I had been looking for about 2 years ago to replace my NAS. It’s well balanced, competitively priced and very flexible. I think Radxa managed to make it cheaper than N100 boards offering equivalent connectivity, which is not an easy challenge these days :wink:

BTW regarding the option of having a PCIe slot, I tend to think that an open slot PCIe x4 instead of the M.2 can be nice. I know this can sound shocking, but let me explain. M.2 cards are mostly SSDs, SATA adapters, rare network adapters with no fixation for the port, or PCIe slots that are not usable beyond testing due to the location the PCIe card would present itself outside.

Instead, a well-placed PCIe slot allows any PCIe card to be connected, and among such PCIe cards are some very cheap M.2, or dual M.2 or even quad-M.2 adapters that offer quite a lot of choice, often even in low-profile form factor. And regarding other card types (NICs, graphics, storage, acquisition etc) the choice is even wider. Of course a PCIe slot is not suitable for a small board like the 5B. But for a large ITX one, which is supposed to be placed in an enclosure often offering an open slot that also means that whatever external port will be offered good fixation, it would definitely make a lot of sense, particularly when there’s already eMMC storage for the bare OS on the board, that doesn’t mandate the use of an M.2 SSD.

The closer a board looks to a PC (station or server), the most likely users are to envision it for use cases where they’d have used a PC.