Raspberry Pi Compute Module 3+ Launches At $25+

[gilboa]

They just joined RISC-V foundation so maybe they're moving into that direction? I agree the RPi3B is now pretty outdated.

Outdated for _what_ and compared to _what_?
(Keeping in mind, for a second, the mission target of R-PI organizational)

- Gilboa

[Tomin]

I've seen a lot of these kinds of modules over the years but I've never really looked into them. Is there an industry standard for them or the daughterboards? If there isn't, I don't really understand the point of making this thing, because the Pi Zero is smaller and easier for hobbyists (who want to do their own soldering) to work with.

There is no industry standard. The point is that if you want to make a custom board with RPi hardware all the really hard to design correctly high frequency stuff like DDR3 (or DDR2? I don't remember) are already there and you don't need to care about that. For (most) hobbyists the Zero is better indeed.

[tuxd3v]

They just joined RISC-V foundation so maybe they're moving into that direction? I agree the RPi3B is now pretty outdated.

Comparing it with recent launched boards , yes it is..
The Allwinner H6 trash's it completely
with no heating problems( small heatsink and small fan on it ), and it stays cool, @1.8Ghz

The major advantage of RPI, in my opinion is the OS..

[starshipeleven]

I've seen a lot of these kinds of modules over the years but I've never really looked into them. Is there an industry standard for them or the daughterboards? If there isn't, I don't really understand the point of making this thing, because the Pi Zero is smaller and easier for hobbyists (who want to do their own soldering) to work with.

There is no standard, nor a need for one, but SoMs (System-on-a-Module) are a thing in many embedded/industrial products that want to use decent hardware like for example Marvell Armada SoCs.

The main selling point is that the OEM now only has to design a simple daughterboard and not deal with VERY annoying high-complexity design like actually making the SoC+RAM+basic stuff run stable (which is not as easy as slapping them on a board), so you can be profitable even if you aren't a major OEM selling hundreds of thousands of these products every quarter.

For a practical example, SoMs allows things like Helios4 https://kobol.io/helios4/ where you have a NAS with a pretty decent ARM SoC with 2GB ECC ram with 4 SATA and 2 USB 3.0, that is still sold at a competitive price even if they are basically crowd-funding each small batch.
I can say with 100% degree of certainty that if they were not using a SoM there would have been exactly no way in hell that they could pull that off with a target price of 200$ (which is more or less on par with commercial designs).

And no, comparing a SoM to a Zero makes no sense as the Zero does not require you to design your own daughterboard, the Raspi Compute Module is aimed squarely at smallish companies, just like any other SoM.

Btw, also Purism's Librem phone is using a SoM, there is no way in hell they can pull off that kind of board design at their scale. https://puri.sm/posts/librem5-progress-report-21/
The SoM is probably going to look like this (if it isn't literally this) https://www.solid-run.com/nxp-family/imx8-som/

[schmidtbag]

There is no standard, nor a need for one, but SoMs (System-on-a-Module) are a thing in many embedded/industrial products that want to use decent hardware like for example Marvell Armada SoCs.

A standard maybe isn't needed, but it would be tremendously useful for certain applications. It'd be nice to be able to just swap out a module with another in the same daughterboard for a quick upgrade. Most ARM platforms have the same basic features and capabilities, so a standard isn't that unrealistic to achieve. Of course, some possess a few extra bonuses here and there, but those are usually in a class of their own and are better off not being in SoM form factors (like the Nvidia Jetsons, for example).
So in the case of SoMs, why not have a standard that includes the basics like a slew of GPIO pins (including stuff like serial, SPI, CAN, and I2C), pins for one USB 3.0 port and one 2.0 port, pins for one SATA port, pins for an ethernet port, and maybe enough pins for 4x PCIe lanes. That of course doesn't mean every module would include support for each of those, but rather those are the "guidelines" that a module would follow. If a module were to include any extra "non-standard" features, those could be accessed directly on the module itself, either through a cable or just a direct connector (like HDMI for example).

The main selling point is that the OEM now only has to design a simple daughterboard and not deal with VERY annoying high-complexity design like actually making the SoC+RAM+basic stuff run stable (which is not as easy as slapping them on a board), so you can be profitable even if you aren't a major OEM selling hundreds of thousands of these products every quarter.

I understand the point of modules, what I don't understand is the point of a RPi module. Pretty much none of the benefits you mentioned apply to the Pi (they are, however, very relevant to other platforms).

And no, comparing a SoM to a Zero makes no sense as the Zero does not require you to design your own daughterboard, the Raspi Compute Module is aimed squarely at smallish companies, just like any other SoM.

What you discussed there is an example of my previous point: if the Zero doesn't require you to design your own daughterboard but you need one anyway, that implies you have a specific workflow or form factor that you need to deal with. Since SoMs are generally used for developers, a standardized socket would make a major impact on saving time and reducing human error. If every socket is the same, you save a lot of time figuring out what gets mapped where, while being able to switch platforms seamlessly if the one you're on doesn't work out.

Also, I was saying how the Zero is a hobbyist device. I understand SoMs aren't, but I'm having a hard time imagining an application where a small business would prefer a Pi-based SoM over the Zero, when you consider how much more work you need to do to use the SoM (and, when you consider how much larger it is, even without the daughterboard).

[pal666]

By by the way that doesn’t mean baring legacy board support. It does mean a bigger board though

bigger board is not backwards compatible

[pal666]

With the introduction--for several years now--of SBCs with 2GHz 4-8 core processors, 2 GB of RAM, 64GB of eMMC...the Raspberry Pi has become irrelevant.

lol. nobody uses all those combined. while you post your cries, millions of people are using rpis

The only thing they do to "keep the dream alive" is to keep coming out with useless "innovations", such as the disaster known as "POE", which could have been designed by a 10-year-old.

that's actually a requirement for sbc from my pov. somehow nobody of chinese 8 core sbc vendors employs 10-year-old able to design it

[caligula]

I've seen a lot of these kinds of modules over the years but I've never really looked into them. Is there an industry standard for them or the daughterboards? If there isn't, I don't really understand the point of making this thing, because the Pi Zero is smaller and easier for hobbyists (who want to do their own soldering) to work with.

This is designed for large scale compute clusters (MPI, Beowulf).

With the introduction--for several years now--of SBCs with 2GHz 4-8 core processors, 2 GB of RAM, 64GB of eMMC...the Raspberry Pi has become irrelevant. The only thing they do to "keep the dream alive" is to keep coming out with useless "innovations", such as the disaster known as "POE", which could have been designed by a 10-year-old. Since the CM has eMMC, why have they not designed this into the original RPi, to replace the highly-corruptible (since the Very First Unit, and NEVER fixed) mass-data-storage SD card? This, in itself, is one of the biggest jokes about the RPi, and should be a major source of embarrassment. And don't talk to me about "price point". If the RPi group INSISTS on keeping their vaunted "$35 price point" in the face of the current competition, they are history.

You're just not the target audience. RPi is leaps and bounds faster than 8-bit Arduino. It has 1 GB of RAM (compared to tens of kilobytes). The POE hat works just fine now. RPi uses quality european engineers for building stuff. Chinese use child labor. No wonder they're cheap.

Given the fact that this non-profit *charity* (!) organization, an organization that was founded with a express purpose of building and spreading a couple of 1000's of cheap computing devices to school kids around the globe, manage to sell >20 million devices in 7 years and more-or-less became the de-facto standard for their class (a class that they more-or-less invented), I would imagine that your disapproval for their hardware design and pricing will not keep them awake at night.

Exactly. And there's excellent software support. Many clones aren't even supported so that you could boot Linux (except some ancient 2.x kernel).

[pal666]

The Allwinner H6 trash's it completely

unless you need video driver, then it's the other way around

[LaeMing]

There is no industry standard. The point is that if you want to make a custom board with RPi hardware all the really hard to design correctly high frequency stuff like DDR3 (or DDR2? I don't remember) are already there and you don't need to care about that. For (most) hobbyists the Zero is better indeed.

Having said that, there are cross-product compatibilities within some producers' product ranges - all the Pi Compute modules are interchangeable with other Pi Compute modules with little-to-no motherboard changes, and several other companies offer a range of boards of different capabilities that are generally interchangeable (depending on SOC-specific features used) within that range, allowing quick design upgrades provided you stay with that brand/range.

But in the end, these types of board are not intended for one-off projects, but short-run products in the tens to thousands of units where the costs of a completely-custom board (especially the /very difficult/* high-speed bus design) can't be justified.

* I come from the age of DIP-packaged CPUs and hand-drawn (usually with special tape, rather than an actual pen) circuit board masks, and these sorts of boards are my savior as there is no way in the world (even with an assist from modern eCAD, which helps immensely but isn't a panacea) I could implement a DDR memory bus with anywhere near the reliability needed to feel comfortable releasing for other people to use! That level of design has gone from 'competent technician' back in the 80s to 'masters degree with the maths skills to back it up' today, and I prefer to leave it to those qualified!