The Raspberry Pi CM4 Begets A Form Factor

It has become the norm for single-board computers to emerge bearing more than a passing resemblance to the Raspberry Pi, as the board from Cambridge sets the hardware standard for its many competitors. This trend has taken an interesting new turn, as a new board has emerged that doesn’t sport the familiar 40-pin connector of the Pi Model B, but the more compact from factor of the Compute Module 4. The Radxa CM3 sports a Rockchip RK3566 quad core Cortex-A55 running at 2.0 GHz, and is to be made available in a variety of memory specifications topping out at 8 GB. It is hardware compatible with the Pi CM4, and should be usable with carrier boards made for that module.

We’ve looked at the CM4 as the exciting face of the Raspberry Pi because the traditional boards have largely settled into the same-but-faster progression of models since the original B+ in 2014. The compute module offers an accessible way to spin your own take on Raspberry Pi hardware, and it seems that this new board will only serve to broaden those opportunities. Radxa are the company behind the Rock Pi series of more conventional Raspberry Pi clones, so there seems every chance that it will reach the market as promised.

Will it make sense to buy one of these as opposed to the Pi CM4? On paper it may have some hardware features to tempt developers, but like all Pi clones it will have to bridge the software gap to be a real contender. The Raspberry Pi has never been the fastest board on the market at any given time, but it has gained its position because it comes with a well-supported and properly updated operating system. For this board and others like it that will be a tough standard to match.

Curious as to what the first Raspberry Pi form factor clone was? We think it’s the SolidRun Carrier-one from 2013.

Via CNX Software.

34 thoughts on “The Raspberry Pi CM4 Begets A Form Factor

    1. From the Rockchip RK3566 Datasheet (Revision 1.1 March. 2021):

      Emdedded Memory
      SRAM 64KB
      ROM 32KB
      OTP 8K bit

      External Memory Interface
      eMMc 5.1
      Nor Flash/Async SRAM
      SD 3.0/MMC 4.51
      SDR/DDR/LBA Nand Flash
      32 Bit DDR Controller (DDR3/DDR3L/DDR4/LPDDR4/LPDDR4X)
      * Support 32bits data width, 2 ranks (chip selects), total addressing space is 8GB(max) for DDR3/DDR3L/DDR4
      * Support 32bits data width, 4 ranks (chip selects), total addressing space is 8GB(max) for LPDDR3/LPDDR4/LPDDR4X

      1. I’m looking at the above and thinking 2^32 is 4294967296 addresses and if they were using each address to read/write a 32 bit word that should work out as a maximum of 16GiB of RAM. But since the maximum for the RK4566 is stated as 8 GiB of DDR RAM that just appears a bit odd, to me, anyone any idea why ?

        Is it that the other 8 GiB of address space is being used for memory mapped hardware (e.g i/o) ?

        I’m curious because I have seen Intel’s marketing people tweak datasheets of chips to deliberately delineate product ranges (e.g. i7-970 has a maximum RAM configuration of 24 GiB, search for “site:ark.intel.com i7-970” in your search engine of choice, but I have seen people install 48 GiB of RAM using six 8GiB DDR3 modules with no issues at all, fully tested with memtest86/pcmemtest)

          1. Interesting, I did not think of that.
            But, if you look at the ranks (chip select), that should imply that
            2 ranks for DDR3/DDR3L/DDR4 chips, would be at most 4GiB each and the
            4 ranks for LPDDR3/LPDDR4/LPDDR4X chips, would be at most 2GiB each which is slightly strange.

            I definitely would not exclude that as the reason because I do not know enough about DDR3/3L//4 and LPDDR3/4/4X to say anything for sure.

    1. If you want that pitch then you just get a full sbc. These “som” (system on module) boards are designed for folks who want access to the various high speed and high pin count buses available on these chips. Fast enough that you might have signal timing constraints, etc, so nobody will break those out at 2.54mm.

      I agree that these fine pitch connectors can be a pain. I’m glad it seems that the cm4 has established a standard, because what’s really hard is finding the one obscure connector your vendor happened to choose (I’ve had this struggle with display panels). The cm3 tried using a common connector, a sodimm connector iirc, but it never really took off. I assume the tradeoffs weren’t worth it. I’ve seen way more cm4 projects despite needing the fancy connector, so they did something right with this generation I guess.

      1. Real thing that makes CM4 jump out and get so much use is its soo much more potent and has more interfaces available – like PCIe, I don’t think there was anything wrong with the SODIMM form factor really, but SODIMM only have so many pins, when you put a beefier chip with more IO on the module you want more pins, the mounting holes to attack to decent cooling solutions etc…

        1. Ironically enough it’s the lack of IO that has made me avoid the CM4 and continue a search for a useful SoM for my uses. I am keeping an eye on the Radxa board due to its “non-standard” 3rd connector to see what that provides.

      2. It wasn’t so much the connector as much as the fact that PCIe was broken out on the Pi4. That PCIe bus being available for sooooo many things changes the entire nature of what you can do with the Raspberry Pi so of course the connector standard was just a harder pill to swallow, but people did anyhow because of the major benefits they gained.

        1. I also suspect that the Pi Foundation worked with their vendors to make sure that availability of mating connectors wasn’t the headache it so often is.

          e.g. “We sell millions of units per year and have major worldwide brand recognition. If you want a design win with us, you WILL make the mating connector easy to obtain.”

          So many product manufacturers fail to do this. For example, I keep on getting emails from people pushing 100Base-T1 single pair Ethernet. I keep on ignoring them because the standard connector for this is TE MATEnet, which is unavailable in any quantity less than something like 30,000. If I need to desolder the connector from your eval board to attach a pigtail with a connector I can obtain, I’m simply going to put that eval board into storage and never touch your technology.

          1. I went looking for specs, but 802.3bw-2015 doesn’t seem to list a specific socket. Wiki says “no specific connector” and its just PAM3 encoding.

            Maybe the devices are all using some specialty part, but RJ-11 seems to be in spec.

    1. That is most likely the onboard ethernet to replace the functionality of the built-in broadcom ethernet PHY that is on the CM4.

      I would honestly just prefer RGMII at this point so I can use something from Microchip or TI that has an actual damn datasheet.

      The wifi looks to be some Azurewave module.

    2. This is true. But having it in mainline Linux doesn’t necessarily translate to a stable and well-supported distro. I hope that happens, because it will mean sharper competition in this space.

  1. Vaporware much?

    Their website is full of broken links – the “models” page for Rock Pi CM3 is completely empty. Click on Buy and go to Seeed and literally EVERY single Rock Pi product is “Out of stock”

  2. Competition on an embedded processor is always good news. It seems very similar performance wise.. through to the GPU. Pricing seems identical. Start with a pi and then switch to this for production depending on deal terms or supply issues?

    1. if you start with a Pi I think you are likely to stick with it, its just simpler – you pick this because it offers extra IO you need natively I would suggest – I’m actually wondering about redoing a design the chip shortage kept me from bothering with yet to try and use this – don’t need to hang a PCIe switch and PCIe-x IC’s off the SOM when it natively has most of what I wanted.

    2. I’m not sure, but usually the rock chip SOCs use a Mali GPU (they almost certainly don’t use a Broadcom GPU) which is one difference. I know the Pi recently got (slow) Vulkan support, but folks (ok, my coworkers) are working on a full open source GPU driver “panfrost” for newer Mali GPU, so depending on your needs or interests you might prefer this to the pi for that reason.

  3. This is really interesting and I do wonder how many Compute Module 4 lookalikes we will see going forward and whether they’ll have the same issues as most raspberry pi lookalikes. At the end of the day there are so many boards out there that are more powerful than the Raspberry pi but not one of them comes close to the support and stability and of course the community that the Raspberry Pi has. Anyway…its all very interesting thing and something im sure everyone will be keeping a watchful eye on! We talked about this article in this weeks episode of The Electromaker Show: https://youtu.be/2HkAvRWtTBc?t=585

Leave a Reply to Foldi-OneCancel reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.