The forgotten child of the Raspberry Pi family finally has an update. The Raspberry Pi Compute Module 3 has been launched.
The Pi 3 Compute Module was teased all the way back in July, and what we knew then is just about what we know now. The new Compute Module is based on the BCM2837 processor – the same as found in the Raspberry Pi 3 – running at 1.2 GHz with 1 gigabyte of RAM. The basic form factor SODIMM form factor remains the same between the old and new Compute Modules, although the new version is 1 mm taller.
The Compute Module 3 comes with four gigabytes of eMMC Flash and sells for $30 on element14 and RS Components. There’s also a cost-reduced version called the Compute Module 3 Light that forgoes the eMMC Flash and instead breaks out those pins to the connector, allowing platform integrators to put an SD card or Flash chip on a daughter (mother?) board. The CM3 Lite version sells for $25.
The Compute Module was always the black sheep of the Raspberry Pi family, although it did find a few applications in its desired use case. The Raspberry Pi Foundation heralded NEC’s announcement of a line of large-format displays using the Compute Module recently. The OTTO, from Next Thing Co., makers of the C.H.I.P. single board computer, also had a Pi Compute Module shoved in its brain. Whether or not companies will choose the Compute Module 3 as a platform remains up in the air, but the value proposition is there; the Pi 3 is a vastly superior computational platform compared to the Pi 1. Putting this power on an easy-to-use module will make for some very interesting products.
If you’re looking for a really cool project for the Compute Module 3, I would suggest a cluster of Pis. The problem with a cluster of Compute Modules is that nearly all SODIMM sockets are horizontal, and for maximum efficiency, you’ll want a vertical header. The good news is vertical SODIMM headers do exist, and you can buy 20% of the world’s supply of these headers for about $500. I know because I did.
109 thoughts on “Raspberry Pi Launches Compute Module 3”
Can somebody tell if the pinout is the same as with CM1?
Great job! Had its doubts about whether it’d be released, and this is wonderful. Guess I will have to try this with the Raspberry Pi phone, with all the success it got on Reddit recenly ( https://www.reddit.com/r/raspberry_pi/comments/5nwmfx/im_making_a_pipowered_opensource_mobile_phone/ ).
Another Raspberry Pi you can’t program in assembly language?
Nothing is stopping you from doing that.
You can compile and run assembly programs on Linux, and there is sufficient documentation/source in a few projects for freestanding C/assembly programs as well.
Get back under your bridge.
I get the feeling you don’t really know what assembly is.
At least my customers don’t like the idea of a module with a connector, and since it’s impossible to buy the processors used on the Raspberry PI boards on medium quantity, I successfully propose to integrate a $5 Allwinner H3 into there projects. The is cheaper and more powerful.
The only problem with Allwinner processors is their manufacturer complete disrespect of OSS licenses and general hostility against the Open Source community. Do not expect anytime them to even pretend to be OSS friendly like the RPi guys do. But even then, the H3 not only is available for purchase, but also much much much better documented than the Broadcom one used in the Raspberries.
Data sheet at: https://linux-sunxi.org/images/4/4b/Allwinner_H3_Datasheet_V1.2.pdf
Actually, I found the sunxi situation very similar to most ARM SoC out there. The sunxi OSS community add new feature almost on every kernel relase. I expect that the sunxi mainline kernel will reach a fairly good usability this year, at least on the H3. http://linux-sunxi.org/Linux_mainlining_effort
I don’t see the Raspberry PI as so OSS friendly. It’s one of the SoC that require many binary blobs.
>>> many binary blobs
Just how *many* are there?
I don’t know exactly how many features requires blobs, but without them the Raspberry PI seem to only have a very minimal set of features: http://hackaday.com/2017/01/14/blob-less-raspberry-pi-linux-is-a-step-closer/
I don’t know actually, but I had the impression that virtually everything, graphics, ethernet, wifi etc needed a blob…
Allwinner is for some reason refusing to release any Android version newer than 4.4 for the H3. It is possible to play 1080p HEVC on an H3 with Android 4.4 – as it’s used in the Orange Pi Plus 2E.
How do you source the H3 at medium volumes ?
Directly from Allwinner or from some Taiwan contacts. H3 MOQ is usually 1200.
I want a Zero 3…
reported this comment by mistake, sorry
What about that the Zero just was available in quantities?
There is no demand for vaporware version 3.
Go visit whereismypizero.com and get your facts up-to-date.
Didn’t you understand the phrase “in quantities”?
Oh, sorry, didn’t notice =) Yes, that’s a problem I’ve encountered myself.
AFAIK it was never meant for that since that means industrial or commercial application, where they’d like you to buy the Compute Modules or full-size Pi’s instead.
Unless you’re hellbent on that particular form factor and are willing to buy a substantial quality, in that case I guess money up front and legally binding contract does the talking.
A Zero 3 would (should) have built in WiFi and bluetooth. Sure you can do the same with a dongle etc., but built-in reduces the system cost.
More importantly, size.
Also more robust, since there nothing *dongleing* around :P
I’d actually like to ditch that because they’re using the SDIO/extra SD-card bus.
So you want to lower the available pins for I/O and increase the cost because you need them? There’s the Raspberry Pi 3 if you need those.
I’ve never seen a PiZero in real life, and even when they are in-stock at Adafruit I still get an out-of-stock error on checkout. I went to Maker Faire Orlando, and there were at least 3 companies selling things for the PiZero, but none of them had an actual PiZero with them. My friend in Texas lives near a Microcenter and says they’ve never had any available either. He’s never seen them at the store, and the employees say they never got any in.
I’m still standing by the theory that the PiZero is a hoax.
I got two hoaxes laying around on my desk then…
They’re manufactured with a VERY TIGHT profit margin, so to avoid losses they’re not making more than they know there is guaranteed demand for.
Also try lesser known yet non-scummy sellers, that’s how I got mine.
>I got two hoaxes laying around on my desk then…
Obviously a paid shill, like those people who claim to have moon rocks.
I’m joking, of course, but pretending it’s a conspiracy makes it easier to deal with while I wait for market saturation.
“They’re manufactured with a VERY TIGHT profit margin, so to avoid losses they’re not making more than they know there is guaranteed demand for.”
Ok, then I have a suggestion – double the price (but no fake pricing, meaning that it shall be the global RS/Farnell price, ex. VAT and tax), and then produce a huge quantity.
My guess is that they would be sold as ice cream on a hot summer day.
you and me both.
Worked for me!
FWIW, they are in stock at amazon.com.
I’ve bought a handful of zeros from the microcenter in dallas. Sometimes for only $0.99 as a loss leader.
You’d figure they’d only have that price as a deal to buy with something else.
my pi zero is a couple feet away from me at the moment.
It isn’t :-) I have one. I think it was bought at PiHut.
“Vaporware” that is sitting in physical form on my desk, hooked up to a spare monitor and keyboard and chugging along nicely.
What’s with all the Pi trolls suddenly? Did [RazPl] learn how to clone himself?
Are you me?
I don’t, and I wouldn’t want one even if it was free.
After over one year after it went to the market the Zero is still impossible to find in single quantities or not bundled with other unneeded stuff. This is ridiculous, and any attempt to explain this with high demand is utter rubbish by paid shills (so please don’t even try): any company that doesn’t mass produce its most wanted product after so much time should have all its businessmen fired, which won’t happen because it’s an intentional move to drive away customers from the competition.
This is why the one and only application I use the RasPI for is where accelerated video output is important, mostly
media players. For everything else, and I mean really everything, The OrangePI, NanoPI, Odroids, Beagles etc. are a much better and often cheaper solution. This is also the reason I’m mad at Allwinner. If they played nice to the OSS community we wouldn’t even consider the RPI3 just for media players.
Of course “in NON single quantities”
be careful with your wording… it’s not impossible to find in non-single-quantities… If you look under a rock, you’re most-certainly going to find it in the non-single quantity of zero.
I had no idea they were hard to find. They are just sitting on the shelves at the local Micro Center. They even had them on sale for $1 briefly.
During the course of 2016 I must’ve gone to flushing microcemter 5 times and on every visit I would get one. Find somone in queens and ask them to get you few zeros, they aren’t selling and I paid a dollar at least twice during my visits there
It’s look like an opos6ul from armadeus systems (http://www.opossom.com/english/products-processor_boards-opos6ul.htm)
wrong link : http://www.opossom.com/english/products-processor_boards-opos6ul.html
Or… Any generic SO-DIMM form-factor compute module? There are certainly more than a few modules occupying this form-factor!
Some very, very powerful options are available outside of the rPi ecosystem, but they all have their use cases!
I’ve been slowly campaigning for Pi integration in projects where I work, designing boards with 40-pin Pi-compatible headers and 16-bit daughterboards that I can swap out for a Pi3 and demonstrate the benefits of an embedded Linux system in an industrial controller.
Hopefully we’ll be able to transition to a compute module. What are the chances this thing is rated for -55-85C?
…Slim. More like none.
From the datasheet:
10.1 Temperature Range
The operating temperature range of the module is set by the lowest maximum and highest minimum of
any of the components used.
The eMMC and LPDDR2 have the narrowest range, these are rated for -25 to +80 degrees Celsius.
Therefore the nominal range for the CM3 and CM3L is -25C to +80C.
However, this range is the maximum for the silicon die; therefore, users would have to take into account
the heat generated when in use and make sure this does not cause the temperature to exceed 80 degrees
I hadn’t gotten to research it myself yet, so thank you for saving me the time.
Still, for lab testing equipment it should be fine. I’ll never understand why my boss wants to put a laboratory test setup inside a thermovacuum chamber.
Hey, where are you doing Industrial controller design? Did a.lot.of the work on the old Moore Products APACS controllers, back when they were the new, new thing .. it was one of the first commercial uses of C++ for embedded systems (the old ACMs still out-gun a lot of modern hybrid controllrs even today)
I work for a little American company that makes waveguide switches, maybe 80 people tops. We have systems for testing switches dating back to the 1960’s, and when I was asked to design a family of replacements I went immediately for Pi support. I have a working prototype that does everything the 1960’s unit does, but is controlled from a mobile-friendly web interface, logs failures to a SQL table, and grabs oscilloscope image of current draw.
It’s primitive, or at least the software is because I’m more a hardware designer than programmer, but I’m quite proud of it.
-55 is always going to be impossible: save TI, you can’t even get microcontrollers down that low. But I’m pissed they’re limited down to only -25. You can get -40 rated eMMC and LP-DDR. Bit less common, but they exist. And unfortunately NAND flash is so touchy I’d never run it outside operating.
>save TI, you can’t even get microcontrollers down that low
And that is the sole reason we use MSP-430, instead of AVR or PIC.
Code Composer is an abomination.
Have you tried the PIC tools?? Ive found CCS and the MSP-FET better than MPLAB X. Our ICD3 bugs out constanty, is always wanting to reflash its firmware and is slow even when it works. I’ve also had numerous issues with breakpoints and compiler bugs.
I can’t judge PIC because the last time I used it I was still using Basic and couldn’t afford the full PICBasic compiler, so I worked with Assembly, which wasn’t that bad.
The MSP-FET is awesome, but Code Composer is built on Eclipse, which is also garbage IMO. Atmel Studio is built from Visual Studio, my favorite IDE of all time.
You should definitely try out the MSPGCC toolchain, as it allows you to compile outside of the IDE, thus allowing Vim and friends. As of debugging, I’m pretty sure OpenOCD handles MSP-FET as a GDB server.
There are chip vendors that make industrial temp range i.e. -40C to 85C for embedded/industrial applications. Embedded ARM chips such as STM32F, Kinetis etc are spec for -40C.
Mil stuff and some other harsh environments might see -55C spec. Usually you can get parts from specialty vendors that do parts screening and verify that the parts will work to the specs you ask for.
Yeah, for exorbitant prices ($100+ each in small quantity), and high lead time. The nice thing about TI is you can get mil-specced stuff for a minimal extra cost, and they’re stocked. Plus it’s the original vendor qualifying them.
It’s insane to think an extra 15 degrees is really an issue, but it is. The -55 rating on the MSP430s is a real help: a design I worked with had problems with the microcontroller rebooting at -45 (and not higher temps). If we hadn’t known the microcontroller was okay to -55, we probably would’ve blamed it. Instead, we looked elsewhere and it turned out to be external power switches turning on too fast (probably due to cap temp variation).
One of the defense programs (where I used to work) did their own system level qualification for a Mil specs with commercial temp chips. The critical part was not available outside of the usual 0C – 70C. We had our own assembly house, temperature chambers, vibration tables etc for the qualification tests.
We had a talk with a memory vendor that did screening and they do quite a bit of understanding what’s going on inside the chip. $100+ isn’t too much for small quantities for the amount of work.
The problem is that tacking on $100+ makes it pointless to buy something like the Pi3 Compute Module, and I’d be damned surprised if a company would requalify the Pi3 compute module for ~$100-ish in small quantities. There are already plenty of -40 rated CPU assemblies out there for that cost, and then you don’t need to bother.
It’s just stupidly annoying because it’s like, a dollar savings on the BOM. Even if they had made an industrial-temp version available for double the cost, and without the eMMC, it still would’ve still been competitive with plenty of other offerings on the market.
Is it not cheaper to add a heating element to bring the temperature up?!
As long as you turn off said heating element when it’s warm enough not to need it… otherwise you’ve lowered the max temp ;)
Extended temp range calls for nor flash instead of nand.
No. There are plenty of -40 to +85 rated eMMC flash chips. Even MLC flash is now readily available in extended temp.
Keep in mind the lifespan of MLC drops off much faster with high temperatures than SLC flash.
MLC has a significantly shorter lifespan than SLC in any case.
See our proposed CM3 cluster from the spring of 2016.
Maybe we should revive it?
Why is this $20 more expensive than a Pi zero? AFAICT, the zero actually has more stuff on it. (I know the zero’s not using the latest SOC, but then the CM1 was at the same price point as this).
Do you want another module so unprofitable that it is never actually sold?
*that*’s what I was missing… thanks.
I wish the Raspberry people would get over their thing with Broadcom. I second the vote for the Allwinner H3, in spite of the objections already raised. There may be even better alternatives that are yet more open.
Thats impossible. The Raspberry people are from Broadcom.
Raspberry Pi is Broadcom
I wish the Android people would get over their thing with Google.
I’m not all for Allwinner at all. They’re really closed source minded. All their older chipsets for tablets for example never got any android SDK updates after all. Almost all cheap chipset tablets are still leaving from the factories with 4.4.4 since it’s as good as it gets.
It would be interesting to design a PCIe board to house a few of these together inside a server. can think of many applications at work where this will be handy.
I’ve been kicking around similar ideas for a while now. Unfortunately PCIe interfaces are neither well-documented nor available as off-the-shelf modules for prototyping. I was thinking of cannibalizing a PCI RS-232 card and using some level shifters to connect it to a Pi’s UART.
Short of a $200 FPGA, do you have any suggestions for getting an PCIe interface working?
There are PCIe to PCI bridges without having to use a FPGA. PCI side is pretty well documented on the web.
google for “PCIe bridge”.
RPi only expose the GPIO lines and various I/O, but not a bus. So not really sure what you are trying to do.
wasnt thinking of connecting the bus up, more have a board with an internal switch and something like an atmega connected to the uart or gpio to get information from the boards
Use either a PCIe to USB or PCIe to serial bridge. Several vendors offer them.
To be fair, at that point I’d just make a PCI card.
What I’d like to do is have a small cluster of Linux boxes inside my PC, running independently. The PCI/PCIe interface would just be a convenient way to make a serial or emulated network connection. Given the excessive complexity of PCIe, I’ve spent more time thinking out how many Pis I can cram in the 5″ drive bays of my PC. With full-size Pi3s, I can fit 2 per bay, and with compute modules I can do more.
Why? Same reasons I keep a Pi1 and Pi3 on my desk now. They’re useful when I want a freestanding Linux box but don’t want to muck around with my main server. All the convenience and horsepower of VMs or Docker, but independent of my actual PC. It’s not really a practical project, at least for me.
PCI is on the way to becoming the new ISA. There are plenty of motherboards now that are all PCIe, though most only have one fully connected x16 slot. Why don’t we have a board yet with at least four full x16 slots???
Because in order to make efficient use of motherboard real estate and not have 32 bit or 64 bit shared buses sprawling across it, they went to serial lanes of PCIe, which are not shared….. so there’s not enough room for more than a couple of dozen and x16 slots want 16 each.
Well they fall back, the old dual slot desktop boards would only give you 8+8 if you filled both slots, you only got 16 in a single slot… though some would have a 4x slot too, that may or may not have forced fallback on the 16s… Anyway, nicer newer boards let you have 16+8 and even one or two spare for your NIC which is pointless being a gigabit, etc….
However, newer than new, bleeding edge desktop stuff coming out the end of this quarter has more.
Not entirely sure how server boards handle it for multi multi lanes, more controllers I guess, but then some of them have multi CPUs on anyway.
This is what i was thinking of, that and having a load balanced across them for things like web hosting inside a server will be handy
This is great for making a GameBoyZero (GameBoyCM3). Have a look at http://www.sudomod.com
They posted the spec sheets a while back so I’ve been working on it around the specs. I’ve got most of it designed but it’s still a work in progress. It’s on my hackaday.io if you want to leave any feedback.
Maybe future builds will move to the GBA SP case.
I still want to make a tablet that uses this form factor processor/ram setup. That way when a new iteration hits just pop the processor thing out drop new in. Then again I like tablets but hate their user hostile serviceability.
Lacking in interest in actually looking up something Pi-related, nor even read the article in any level of depth… am I to guess that these have *way* more GPIOs than the normal Pi’s? Maybe even access to the data/address bus, so they could be useful?
AFAIK, no. It’s largely just the 40-pin header replaced with a card-edge connector, and the board layout made as flat and compact as possible.
If you’re doing an embedded application where you don’t need those bulk ports and are mounting the Pi on a PCB anyway (like most of my Pi projects), the compute module is for you.
It seems like you do. You get up to 47 GPIO and looks like an extra camera input.
On the dev board I see disp0 and disp1, what are the chances it can handle MIPI DSI to a high res (wqhd) smart phone screen?
Next to none I think, there haven’t been any promises from RPF to release the documentation for DSI, if I’m not mistaken.
I could actually think of a use case for a cluster of these. A low power low cost ci-testing rig. With 60 devs constantly pushing code into 8 codebases, having 10-20 of these just dedicated to running phpunit/go test/selenium would kick butt
I think these are great .. a little cheaper, a lot smaller. I can find plenty of uses.for them in my work.
Gotta hear the vertical SODIMM story?
I second this…
What is Benchoff planning with that many vertical sodimm sockets…
I think he told it on here before, as far as I remember it went something like this…
I just really really love this form factor of computer. I can imagine a near future where this is how all computing is done, with modular cores that can serve at any form factor, smart phone, laptop, drivable, wearable.
$66 USD here in South Africa when buying it from RS :(
Ive given up on getting anything Raspberry Pi from RS SA , I stopped buying Pis completely and have switched over to OrangePi from Aliexpress or if I absolutely need a Pi I would get one from Ali aswell.
The irony is one of the big blerbs when the Pie Launched was that it could be used as a computer replacement in 3rd world countries ALA South Africa.
if only my tv had a sodimm to hook it up
Despite the fact that the Armbian OS has not been mainlined yet, the fact is that the Orange Pi Zero, $7/9 with 256/512 MB, full Ethernet port with passive PoE, USB 2+ uUSB OTG power, Wifi , TF slot and H2+ quadcore ( same as H3 on higher models sans 4k display or GigE) blows out even RPi3 on CPU performance ! And it is available in all quantities.
The $2 expansion board gives you TVout, mic/ speaker, extra 2x USB2 and IR ports.
Armbian runs fast and auto regulates CPU frequency, thus no heat sink needed. It is available with both legacy and vanilla kernels, for both media player and headless modes.
RPi0 is still a mirage. It served its purpose to force the industry to produce cheap SBCs as a response to the $9 CHIP, now mainlined and working well. So yes, RPi0 has done good for the industry as a whole that can now move from hobbyist mode to the commercial media- IoT market to benefit all of us :)
While armbian has a great OS that has really made the orangePi boards and others accessible. I still prefer RPI3 boards. You ultimately still get better support and documentation a multitude of libraries in C/C++/Python/Java e.t.c for io access. The Raspberry Pi boards are also manufactured in the UK and this translates into decent jobs in the UK…something that at least other fellow Englishmen should support and encourage. Cheapness is not everything….something that unfortunately many readers on hackaday seem to forget.
Innovation is the best job creation platform, which is what open source is about. And nothing drives innovation like price/performance since it creates the sales volume that drives more jobs and further competition to serve the end user.
Raspberry Pi has great vendor ( Broadcom) support, and is not open source. Not sure how many UK jobs it can create given that US Broadcom likely benefits more from Pi Foundation than Chinese Allwinner does from its SBC vendors. Whatever UK jobs are there is a short term phenomenon given global pressures.
There is this myth about Chinese clones. Ironically, it is RPi types that cloned, at a very high price, Chinese vendors who have been catering to their local market for many years with Chinese documentation.
Ultimately consumers need price/performance forcing firms to move up the value chain. Both CHIP and OrangePi have revved up in the market.
Innovation is indeed the best job creation platform and I won’t deny that cheap Chinese clones have nudged the RPi foundation to offer more value i.e. add onboard WiFi+Bluetooth & use more powerful SOCs. But let’s not forget that it was the Raspberry Pi Foundation (Along with the beaglebone guys) that created an affordable (under $35-50) and accessible SBC market to begin with without compromising on quality.
Open source hardware is great and I’m a big advocate of it…But it isn’t everything. Many of the Chinese boards are not open source hardware nor software. AllWinner is one of the biggest GPL violators software wise.
Also building clones of something isn’t exactly very innovative….even if it costs less and the clones aren’t truly clones…they can’t use the same GPIO python C/C++ libraries as the RPI due to fundamental differences in the hardware. so they’re not ‘true’ replacements….they usually need more time, headaches, work to make them do certain tasks compared with the Raspberry Pi boards.
And I’ve purchased many SBC in my time…RPi’s, AllWinner’s, Samsung…..I now recommend to everyone to stick with RPi boards….they’re more convenient and easier to use. The $20 extra that you pay for a Raspberry Pi are worth it if just only for avoiding the headaches that one gets when using the other boards.
The CHIP is an interesting experiment…Chinese hardware but with RPi foundation style support. It is pretty neat….But I’ll stick to recommending what I know and what is manufactured in my part of the world. ‘Globalization’ and ‘Free trade’ concentrate wealth and in the end only benefit the rich. They are dirty words and people from all over the world need to stop embracing them.
BTW RPI zero is not a mirage…I own four of them and I live in Canada.
Will it be performant with stero video? there were some attempts for depth perception with stero imaging with the CM1, how far this vesion will be performant?
Please be kind and respectful to help make the comments section excellent. (Comment Policy)