Looking for cheaper and open-source friendly alternatives to RPi Compute Module

Hi, Linux experts!

We are a small software company and we have an idea for a product which will need a hardware platform to run on.

As we want our product to be open for maker/hacker community, I now would like to take the opportunity and talk to you out there.

Unfortunately, I cannot yet disclose much details about the project (who knows, how it will go) but I'll try to highlight some important points which might be affected by our choice.

At first we had an idea to use Raspberry Pi Compute module. Currently we are experimenting with Raspberry Pi B+ and it works fine, but it seems to be too powerful for our needs. We don't use its video core at all, and the CPU also is usually only loaded by 10-20. Also 4GB of eMMC flash means lots of unused space. Raspberry Pi Compute module is not the best for us - we'd be paying for unused resources.

But there is one thing we really like about Raspberry Pi Compute module - it is openness. A product based on it would be available to enthusiasts. Even flashing RPi Compute is pretty safe - just connect to USB and write the firmware to eMMC flash memory. It seems almost impossible to brick this device.

We'd like to find a cheaper, less powerful system-on-chip (SoC) or computer-on-module (COM) or just separate chips of CPU+RAM to use in our product, but we want to have the openness and almost fool-proof flashing of RPi. It would be fine to have some binary-only vendor specific drivers, if only the vendor would provide them to public or give us permission to distribute them at least in binary form (of course, open source would be the best, but let's be real - some SoC manufacturers want to keep some things secret).

Our product will require the following interfaces to the outside world:
- a WiFi b/g/n (with access point mode support) adapter
- a Bluetooth 4 adapter
- two Ethernet adapters in bridge configuration (ideally - two port hardware switch)

And internally we'll need and SPI port with support in Linux kernel. Also it would be great to have UART port for serial console.


We could just look for the chips with good support in Linux kernel for the embedded platform of our choice and then look for the prices and availability in the market. But there are some SoCs with some of this functionality built-in, thus it might turn out to be cheaper than to use separate components. Correct me if I'm wrong.

Price is really important to us. The product won't be high-end device, so we'd like to keep the price low, but maybe having just a little bit of spare processing power for future possibilities.

If there are cheaper production ready, stable and open alternatives to Raspberry Pi Compute module, I'd be really grateful for your experience and expertise.

There is an inherent fallacy in your question.

The world of Electronics has switched to Arm devices due to factors like familiarity, os compatibility, and that you get the IP core for your chip, not a finished piece of silicon. So they are cheap. Power has nothing to do with it. As a student I wanted to put together an digital FM discriminator: I had the College and the sponsor telling me what to do. I was thinking 250Mhz; the College said 'if there has to be a cpu, make it an msp430(16 bit, 8 mhz).' The sponsor said 'If there's any cpu, maker it an arm 32 bit.' Both were crazy, slow and overpowered options. The days of building around cpus that do not multitask are over.

For quick advice: Stick with your Pi, or consider SoCs from companies like Rockchip, Allwinner, Exynos, & Qualcomm and other competitors. Rockchip is in the bottom end of the market, but do an rk30 board with the complete electronics. They all have GPUs that you don't want and an Android OS(usually). Whatever is manufactured in zillions comes cheap. Price rules. Never mind the size.

Thank you, yes, you are right, these days there is no strict correlation between processing power and price.

I have seen a similar product to ours implemented on Freescale MPC8314E:
http://www.mouser.com/ds/2/161/MPC8314EEC-51265.pdf
but I'm not sure how open Freescale is and how do they treat small startup companies - is it possible to buy their SoCs in quantities less than 1000?

Also Atmel looks good. These guys have created modules for attractive price, it almost matches our needs, but we'll need eMMC flash instead of microSD slot. And then finally it might again cost the same as RPi compute...
http://www.acmesystems.it/arietta
Is Atmel ARM platform really that open as AcmeSystems claim?

Allwinner also might be a good choice, but I've heard that they are not that open in some areas. But how do they compare to Atmel and Freescale?

Arm is pretty open. You get an IP core which you can throw into your fpga along with your own dirty thoughts and get a finished SoC. The Fedora Electronic Spin gives you loads of ARM tools.

Try for an MOQ from Rockchip also. They seem to fit the bill. They offer chips to the bottom end of the phone/tablet market. There's a live thread on LQ atm about slackware linux on the RK3288. Their boards come with cpu, gpu, bluetooth, 2 types of ram, sdcard, hdmi, and wifi (usually a Realtek or broadcom chip).

The Pi is about the cheapest full solution example there is. There are Beagleboards, but more money. Maybe Arduino, or STMicro. I see where you can get an ARM dev board from STMicro on Mouser for about $10. As you've said, you either wish to use those existing boards and their designs or take the micro and design your own board. The intention of some of the hobbyist variations are that you would be capable of creating a custom design using their examples.

You say you want lesser and lesser of a chip, but yet you want a networking stack, Bluetooth LE capabilities, and WIFI b/g/n. This merits a full kernel. You either want to go no OS or you want to go Linux. You can kill yourself to strip down Linux as much as possible, but my thinking is that with these forms of networking and connectivity, you may be considering running some network capable services. Fine it will be headless and not be running an XServer, but I wouldn't fuss that the thing has 4G of flash. Segment that to have a secondary partitions for your product and boot and have another set of segments for default and saved system settings. This way you can manage upgrades properly with minimal risk.

Thanks, rtmistler, yes, that makes sense.

We definitely want Linux and we want to pick a platform on which it would be possible to get Linux running easy. A platform, which would not enforce too many vendor-specific restrictions, such as "we provide you this Linux driver for our component X, but you are not allowed to distribute the driver to anyone else, even not in binary form."

We initially picked Raspberry Pi because it is so easy for anyone to find all the docs and to assemble their own distribution (accounting for binary-only stuff from Broadcom), but now we are evaluating options to get our production costs down by getting rid of unused GPU and flash memory. Yes, we are planning to use staged partition approach to upgrade, but 4GB seems still too much. But I have read some options why 4GB is being chosen more and more for such platforms - it's because 4GB is becoming mainstream and lower capacity chips are being phased out and might become unavailable soon because it would be too inefficient to produce them if you can get 4GB chips for almost the same price.

But as it was suggested here by business_kid, it might turn out that there is no point to look for cheaper solutions because at the end they all turn to be the same, plus-minus a few dollars, and then only OS support matters.