Interesting "Competitors" for the Raspberry Pi

The parallela sounds and looks cool but we are talking about a major project that includes sillicon rebake and that probably is 6-12months away from mass production.

Meanwhile while there are other developments that are not exactly comparable to the R-Pi, they are slowly but steady making progress, like the OLinuXino boards from Olimex. I'm testing both the micro and maxi using the Freescale iMX233 application processor SoC (fully documented by the way,) waiting for the A13 one and they also have under development one using the A10. The iMX233 Maxi uses also the LAN9512 and I didn't find any issues with it yet (and BTW the connectors are properly aligned and it has a nice swtiching power supply, the board is slightly bigger than the R-pi, no HDMI/VGA tough.)

I also recently got a TI AM335x Starter Kit, you will start to see more stuff based on the AM335x and TI will get more agressive to dettach the OMAP from some applications and put it on boards similar to the R-Pi.

Still the R-pi continues to be a good idea, unfortunatelly with an associated plan badly executed and with more focus on hype and promotion than really make it a strong and solid "product" with the participation of a community that is eager and able to contribute.

No schematics, no Gerbers, major concern is "cloning" .... I'd not clone something that is not working properly ...

-J

At least it looks like Olimex is taking the problems from people using their boards seriously.

Their Forum is an example of what a forum should be like.

The supply dip when you connect an usb device is a known fact. It might be hard to notice it on a scope, unless you have a good memory scope and are able to trigger it properly. The issue happens more frequently on the newer boards with the usb polyfuses removed. If you look at the Pi schematics, there aren't much buffer capacitors on the 5V supply rail. (just 1 small capacitor). I guess we get what we payed for.

It looks like a lot of the Pi usb issues are caused by timing.

As linux isn't a real time OS, and as stated before that the cpu can become heavy loaded, I doubt they will ever be able to fully fix the issue.

Besides the usb, there are other dissapointing things. One mentionned already is the fact that X11 isn't hardware accelerated.

Another one is the poor quality of the analog audio output. A third one is the fact that you need to properly shutdown the system to avoid sd card corruption. This makes it difficult to use the Pi as an embedded solution. A fourth one is the critical power requirement of the board.

Some of those can be resolved, but all solutions make your cheap solution (not so cheap) anymore.

I know it's a little off topic, but these points are worth checking when seeking for "Competitors" and might rectify a higher price.

George Ioakimedes wrote:

 

So, from all of our discussion I haven't really seen a viable contender to the RPi considering it's cost and video capabilities. I really like the OMAP boards but they aren't even close to the $35 mark. The TI AM335x looks pretty good and there are some boards starting to come out which could get you close to a $50 mark but I haven't touched one yet so I can't say 1st hand how they perform. TI does do a pretty good job supporting the community though.

 

Is there anyway we could collectively rally together and try to fix the RPi or make a strong collective recommendation to The Foundation or is the closed nature of Broadcom too much of a deterent for anyone to jump on board?

It's going to be hard for anyone to compete with RasPi, especially when they start shipping Model A at US$25.  Most companies want to make a good profit, which already puts them at a disadvantage against a not-for-profit.  The main thing is volume: if BeagleBone had the same volume as RasPi, they could probably match the price.

Broadcom has a bad reputation for openness.  Anyone who's tried to design with their parts has found out how difficult it is to get data unless it's a ridiculously common part like an Ethernet PHY.  I really don't think they're going to open up BCM2835 documentation until it becomes blatently obvious that it's costing them business.  The best way for this to happen is for people to choose products with open documentation.  However, this doesn't happen with mobile technology: most people just want a consumption device and couldn't care less whether you can get a tech reference manual.  I don't expect Broadcom to open anything up any time soon.

Different companies have different traditions regarding open documentation.  This mostly has to do with corporate culture and corporate history.  I have a theory about one company that's particularly good about open documentation.  According to rumor, this company used to publish data sheets before they did the layout of the part.  If enough orders came in, then they'd assign engineers to do the layout, prototype fab, and testing on an exhaustingly short schedule.  This meant (a) you probably didn't want to be a chip designer at this company, and (b) you'd better get samples before designing that company's parts into your product.  In fairness, the hardest part of designing a part is determining its function and target specs, and once you have those doing the layout is straightforward as long as the timing is reasonable.  Also, I don't think that company does this anymore because modern parts take way too long to get from start of layout to samples.

The point of this story?  I think one big reason that company has such open documentation today is because their data sheets literally sold their products in the past.

While it may not be the most important, one of the main reasons that nowadays some companies are so picky about releasing information is becasue they may probably don't own all the IP associated with a product, and some of the SoC type chips are just a collection of IP modules (like the Synoptics USB module) from different vendors that are glued together with some extra stuff and under non disclosure type of relationships.

Then knowing from where each piece comes from it makes more easy to clone some chips.

Companies like TI with very long history and tradition in electronics development (besides its acquisitions) has been always a company generating IP, so even if you have a detailed datasheet and reference manual, plus other support documentation, you can develop your products openly but not clone their chip.

And by cloning I don't mean reverse engineering or copying masks, etc.

Besides the particular relationship with Broadcom, one of the major issues I see with RPF is the total lack of transparency and accountability for a non-profit organization.

How much will compromise their confidentiality if Broadcom at least releases the list of pins and what each does, the electrical and timing characteristics of their chip, how come there are not yet schematics for Rev 2, etc, etc.

And about the subject of the R-pi being intended for educational use, nowadays we live in a connected world, for schools it will be much more easy to implement a wireless network than a wired network, so the model A with a good certified WiFi adapter will work, but alas, if you want to have a keyboard and monitor you are out of USB ports !! then you need an external hub.

Again, I still believe the R-Pi is a good idea, but with many design issues and shortfalls and with very bad execution compensated by a lot of hype and promotion.

My .02

-J

@ George: Regarding community fixes for the Pi: This is a tough one. My gut says that this is a commercial product and it's the vendor's job to provide fixes. Unfortunately my (rather Darwinian) view is subject to these caveats:

1) It's not fair to abandon Pi customers to the miserable level of support that the Foundation provides.

2) The Pi doesn't have any direct competitors (i.e clones or functionally identical units), so there is no opportunity for customers to shop elsewhere and hit the Foundation in the wallet..

3) Regardless of 2) there seems to be no shortage of potential customers for sub-optimal, poorly supported chaff. This is one of the problems I have with the whole GNU / FOSS "community" thing - if something is clearly broken and fixes are not forthcoming then don't keep it on life support - just toss it out and make room for something better. And make a note not to get burned again.

Regarding the long-running discussion on Broadcom's closed tech - I don't buy into this at all. Any serious customer will (in return for signing a few non- disclosures) have access to extended technical specs and will hopefully have the brains to do something useful with it. Remember also that a lot of this closed information isn't Broadcom's to give away, but is third party intellectual property used under licence. Regardless of this I feel that handing out source code just gives unscrupulous manufacturers another excuse to kick back and let the community pick up the slack.

I do feel that the market is crying out for a low cost general purpose hardware / software package, but there are currently two barriers - the hardware and the software.

1) The hardware: ARM - not so much a platform as a sprawl. It's overwhelmingly used to power consumer devices, so maturity, backward compatibility, future security and all the other stuff that a community needs to get a piece of hardware to do something interesting just isn't there. Your chosen software may not even have an ARM port. Obviously this is no barrier for the dedicated geek with access to source code, but it's a faff. The fact that ARM is such a mess certainly doesn't encourage software developers to embrace it.There are a lot of excellent embedded boards around though for more specialised applications and while Pi users might balk at the price and the learning curve at least they won't have to try three power supplies and five SD cards to get it working.

2) The software: Android anyone? Fine if you want to poke about on the internet and spend your pocket money on apps. Desktop Linux? It's not the Penguin's fault, but the limited power, proprietary hardware and all really mean that a lot of customisation is in order and with it a great deal of compliling and terminal-fu. Not a great recipe for universal take up, but as ARM is so splintered I really don't see the situation changing any time soon for consumers who want an ARM P.C. for general duties and want it to work out of the box without having to constantly *** around in terminal.

Maybe one day it will be as easy as x86 and I'll wonder what all the fuss was about.

Edit: I was so slow to type this nonsense that a couple of interesting posts have popped up in the interim.

@ John: The fact that ARM is overwhelmingly a platform for consumer devices is key IMO. Intellectual property will be closely guarded. Unfortunate, but if it wasn't for the consumer thing then these devices probably wouldn't get built in the first place, so these arguments would be moot! Of course some companies are a little more open than others - in my experience Texas Instruments have always been pretty generous with research papers that they could have kept to themselves, as well as extended specs, hardware samples and the like. A particular corporate culture, as you say.

The Foundation is a registered charity - not a non-profit per se. The initial run of 10k was to fund the educational project, so there is markup factored into the Pi's price. Now multiply that markup by two orders of magnitude... It will still be hard to compete with the Foundation in the short term, but I think that's more due to the ARM thing not being quite ready for general purpose computing yet.

@ jamodio:The whole argument for the Pi seems to revolve around the fact that school PC networks are too locked down and vulnerable to allow kids to do anything interesting with them. As soon as a classroom full of Pi are networked then the same vulnerability argument applies (Linux might be more robust, but kids are smart...) TBH it would have been more practical to utilise all those PCs that we've all paid handsomely for by sandboxing. Obviously the Pi's secret weapon is GPIO, but I'm beginning to get the feeling that buying in Arduino / Gadgeteer would cover interfacing and it's coding.

OK, so maybe trying to bolt onto the RPi isn't going to be a viable solution but in theory it really shouldn't be. I guess since Eben was/is(?) an employee of Broadcom explains why the BCM2835 was chosen to begin with.

So where do we go from here? The Pandaboard is nice but at $160 it's in a different market (although I do have one as well as an ES and BeagleBoard and a TrimSlice, and ...). That kickstarter project in the OP does look interesting but I'm on the fence with it. I like the OMAP (in fact I'm working on a board which uses the OMAP4460) but you really can't buy it.

So "we" can't fix the RPi and there doesn't seem to be a viable alternative so I guess it's time to hibernate?

George: We can hope that "the market" fixes it! Maybe a smart manufacturer will come up with a hardware / software package (a bit of a long shot, but at least potential manufacturers now have a blueprint of how not to bring a low cost board to market!). We can also try to apply continual gentle pressure to the Foundation to provide proper fixes (rather than handing them workarounds that just buy them time) and let people out in the wider world know what's really going on. It's not much, but it's all I've got just now! *

I don't think that the BCM2835 is a terrible chip per se (notwithstanding the ongoing USB fun) - if the GPU were better utilised then it would free up some CPU time. It is a shame that it's older architecture, but an ounce of tight software integration is worth a pound of overclocking, as my mum might have said.

* Edit: If at some point in the future it appears that the Foundation have pulled up the drawbridge on development in a particular direction then it will doubtless spur the community into an independent fix. That's great IMO - I'm not suggesting that people abandon Pi development work, or to stop writing applications for it, or to stop providing help for other users. I just feel that providing basic functionality  is the responsibility of the Foundation. They are best equipped to do it and it's what they are there for. Let them show what level of support they can provide.

There, I didn't want anyone thinking that I was being even more grumpy than usual.:)

Assume the goal is really education.

To get a computer class up and running, you will indeed need a lot more than just a bunch of cheap Pi computers.

The biggest cost seems to be the screens.

I would expect a low budget computer class setup to use second hand screens, but those don't have a hdmi or dvi connection.

So the problem is they just learn how to use word and how to create websites...

The closed nature of the Pi isn't very helpfull either to learn what is going on under the hood.

The non hardware accelerated x makes it feel sluggish.

In my opinion, if price is what matters, a good classroom setup could be some thin clients or old pc's used as thin client, connected to a decent computer used as terminal server.

In times of windows nt4, M$ already had a version for an arm desktop platform. They are creating a version of windows 8 for arm as well.

So, it looks like arm will find it's way into desktop computing.

The reason we find it in so many consumer products is due to it's low power specification that makes it the best choice for battery operated equipment.

The fact you need to properly shutdown the Pi to avoid sd card corruption is indeed something you have with all os.

My point is that nothing is provided in case of the Pi to do this properly, specially not when you intend to use your Pi as an embedded board. The competitors and most commercial products using linux have an on board flash chip used to store the os "read only".

Some of the Olimex boards have a connector for a battery that can gracefully shutdown the system in case of a power failure.

And don't we all connect our servers to an UPS so that they can proper shutdown in case of a power failure?

Besides that, the Pi seems to corrupt it's sd card more frequently than a desktop linux corrupts it's harddisk.

Some say it's due to the nature of sd cards making them a bad choiche for such applications.

All I can say is that it is an existing issue you should consider when you start using the Pi.

Jamodio:  We are not worthy!!!  /me bows in admiration.

John, perhaps the answer is to go non-commercial.  A modular open hardware board could be layed out to enthusiast-level specifications, avoiding the problems of soldering BGA packages by accepting a SoC on a header instead of the BGA device directly.  Separately, we could commission a tiny BGA SoC on a corresponding breakout header to be manufactured for us after a suitable SoC is chosen by community consensus.

It's certainly doable if the open hardware community wanted to do it.  The cost of the unpopulated boards would be minimal, far undercutting Raspberry Pi.  Admittedly it's not a fair comparison, but unpopulated boards are not a disincentive to most electronics enthusiasts, and I'm sure that companies would spring up selling kits of parts as well as fully assembled boards for profit.

PS.  We could even ask competing SoC manufacturers to develop their own SoC-on-a-breakout-header for us.  Very little effort and cost for them, and potentially good PR since it constitutes a form of marketing targetted at engineers and enthusiasts, just like normal prototyping boards but even cheaper.

Morgaine.

The competitors and most commercial products using linux have an on board flash chip used to store the os "read only".

Well you can also treat the sdcard as "read only" and use something like jffs2 instead of ext4, but it does take somewhat more effort to build a useable system with a read-only root filesystem. The obvious problem here is that the current OS images for the Pi want to think it's a 'normal' desktop PC, fill it with essentially irrelevant crap that doesn't need to be running etc.  Frankly I'm surprised it works as well as it does given the choices you make for spinning rust can be so very wrong for sdcard.

@Morgaine Well I'm in a unique position to do start such a project, on the funding, design, and manufacturing. I've waited for years now to find a good solution in the market and I had hoped that the RPi would hit most of the requirements but it just didn't happen. I don't think it's reasonable to find a solution that isn't BGA, besides, how many people really are doing any soldering on parts with pin counts more than ~50, and why would they need to?

Personally I like the OMAP line of parts but all the boards are higher cost than what I want and the BeagleBone doesn't do 1080P and if I was to do anything it doesn't make sense not to support 1080P. Is TI/distribution pricing their boards like standard retail markets so the real cost is sub $50? The last time I spoke with TI the OMAP was still only sold to major OEMs but with recent announcements that might not be the case.

@Morgaine Well I'm in a unique position to do start such a project, on the funding, design, and manufacturing. I've waited for years now to find a good solution in the market and I had hoped that the RPi would hit most of the requirements but it just didn't happen. I don't think it's reasonable to find a solution that isn't BGA, besides, how many people really are doing any soldering on parts with pin counts more than ~50, and why would they need to?

Personally I like the OMAP line of parts but all the boards are higher cost than what I want and the BeagleBone doesn't do 1080P and if I was to do anything it doesn't make sense not to support 1080P. Is TI/distribution pricing their boards like standard retail markets so the real cost is sub $50? The last time I spoke with TI the OMAP was still only sold to major OEMs but with recent announcements that might not be the case.

how many people really are doing any soldering on parts with pin counts more than ~50, and why would they need to?

Well I certainly do.  It's becoming less of a problem these days as lots of stuff is moving to being connected serially meaning that pin counts are coming down.

You're right though, until something can challenge the Pi in terms of cost with a roughly equivalent set of features it's kind of pointless. The $25/$35 is some sort of magic number that boards costing many times more really can't compete with, people aren't looking beyong the headline price - it doesn't matter if the more expensive one is somehow 'better'.

My comment about the sodlering of parts was aimed more at the comments made earlier about "needing/wanting" a board that didn't have a BGA. I just don't see that as a requirement even for a hobbist board. Are any of the Arduino users even replacing the processor on those boards? Morgaine's comments about utilizing an add on module for the main CPU is what I'm seeing more and more of with a 200-pin SO-DIMM being rather common. The problem with what I've seen though is that those aren't cheap. Even the Variscite AM335x SO-DIMM board is priced at $39 which seems nice but then you need a baseboard and all the connectors which brings you right back the BeagleBone cost.

George, it's no different to hobbiest suppliers often providing a semi-populated board containing only the SMT components for the design already placed, in order to support those enthusiasts who like to construct but for whom soldering SMT is difficult.  There are many such people, even today with the wide presence of SMT.

Likewise, soldering BGA would be problematic for the majority of hardware enthusiasts, so providing the BGA part on a plugin module would play a similar role.  The rest of the board would be constructable by the majority of enthusiasts using their more common building skills, which does not include handling BGA.

Morgaine Dinova wrote:

 

George, it's no different to hobbiest suppliers often providing a semi-populated board containing only the SMT components for the design already placed, in order to support those enthusiasts who like to construct but for whom soldering SMT is difficult.  There are many such people, even today with the wide presence of SMT.

 

Likewise, soldering BGA would be problematic for the majority of hardware enthusiasts, so providing the BGA part on a plugin module would play a similar role.  The rest of the board would be constructable by the majority of enthusiasts using their more common building skills, which does not include handling BGA.

I don't mind SMT, as long as lead pitch isn't below 1 mm.  I really like the EOMA-68 form factor, which uses a PCMCIA 68-pin connector.  Then you design and build a base board with your custom thingummies, and plug in the EOMA module to provide computing.  Next year, you plug in the next generation module with twice the computing, keeping your base board.

I like SODIMM connectors except for the fine pitch, because they have reliable high-pressure contacts.  I might be convinced to use an SODIMM-144 connector (0.8 mm pitch), but fuggeddabout an SODIMM-200 with 0.6 mm pitch.

John Beetem wrote:

 

I like SODIMM connectors except for the fine pitch, because they have reliable high-pressure contacts.  I might be convinced to use an SODIMM-144 connector (0.8 mm pitch), but fuggeddabout an SODIMM-200 with 0.6 mm pitch.

There are already a few companies using the SO-DIMM-200 and it's mainly used because you almost need all of those pins to get all of the signals out off of the processor. They aren't that bad to solder because most of them have mounting feet which lock there posistion down on the board and then you only need to add solder. But if "we" went this route, wouldn't you want to provide a low cost baseboard that brings all the signals out anyways? If you're trying to reach economy of scale, what percentage of people are really going to care that they might not be able to solder their own connector? Did millions of people not buy the RPi because they couldn't solder the parts themselves? Or for that matter, how many people are even building their own Arduino boards even though you could? If you price a board at ~$50 I don't think too many people are going to go, oh, that's too much, I could solder those parts down and save $5. Look at what TI has done with their launchpad boards, how can you beat a $5 development board? Let people buy the base and fiddle with their own stuff.

George Ioakimedes wrote:

 

John Beetem wrote:

 

I like SODIMM connectors except for the fine pitch, because they have reliable high-pressure contacts.  I might be convinced to use an SODIMM-144 connector (0.8 mm pitch), but fuggeddabout an SODIMM-200 with 0.6 mm pitch.

 

There are already a few companies using the SO-DIMM-200 and it's mainly used because you almost need all of those pins to get all of the signals out off of the processor. They aren't that bad to solder because most of them have mounting feet which lock there posistion down on the board and then you only need to add solder. But if "we" went this route, wouldn't you want to provide a low cost baseboard that brings all the signals out anyways? If you're trying to reach economy of scale, what percentage of people are really going to care that they might not be able to solder their own connector? Did millions of people not buy the RPi because they couldn't solder the parts themselves? Or for that matter, how many people are even building their own Arduino boards even though you could? If you price a board at ~$50 I don't think too many people are going to go, oh, that's too much, I could solder those parts down and save $5. Look at what TI has done with their launchpad boards, how can you beat a $5 development board? Let people buy the base and fiddle with their own stuff.

I'm not interested in soldering to save money.  That concept went away a few years before Heathkit went out of business.  The reason I'm interested in 1mm+ pitch is so I can create custom baseboards with specialized interfaces and assemble them myself, and then plug in a module that has all the fine pitch and BGA components, such as Variscite SODIMM-200 or Gumstix.  However, both of those require fine-pitch connectors.  Now, it may be that in addition to low-cost PCBs you can find places that will also solder down a few fine-pitch components as well.  Unless such exist, I either need 1mm+ pins or I'll just buy a BeagleBone or Cubieboard or ST Discovery or whatever and make the case tall enough to fit it.

John Beetem wrote:

 

I really like the EOMA-68 form factor, which uses a PCMCIA 68-pin connector.  Then you design and build a base board with your custom thingummies, and plug in the EOMA module to provide computing.  Next year, you plug in the next generation module with twice the computing, keeping your base board.

Me too.  What's more, it's very appealing for making clusters, since such a large number of vertically-oriented EOMA-68 modules can be placed side by aside in a horizontal rack.

This is the age of multicore, allegedly.  We should not be building general purpose computers of any ilk with single CPUs anymore, it will count against our karma at the pearly gates.

Morgaine Dinova wrote:

 

Me too.  What's more, it's very appealing for making clusters, since such a large number of vertically-oriented EOMA-68 modules can be placed side by aside in a horizontal rack.

 

This is the age of multicore, allegedly.  We should not be building general purpose computers of any ilk with single CPUs anymore, it will count against our karma at the pearly gates.

Yes, a EOMA-68 cluster sounds great!  It would look something like the inside of the HAL-9000 in 2001: A Space Odyssey

The lesson of multicore: if RasPi had a multicore SoC, one core could handle USB properly and one could run X server full time, eliminating two serious performance issues.

I totally agree with a multicore processor but I fail to see how you think 68 pins is near enough. If we're considering the embedded space then you can't ignore direct connection to an LCD which means 24 pins just for RGB, then there's Ethernet, USB, SPI, PWM, and on and on. You guys were complaining about wanting an easy method to solder devices but it's specialized equipment to weld those PCMCIA clam shells together.

George Ioakimedes wrote:

 

I totally agree with a multicore processor but I fail to see how you think 68 pins is near enough. If we're considering the embedded space then you can't ignore direct connection to an LCD which means 24 pins just for RGB, then there's Ethernet, USB, SPI, PWM, and on and on. You guys were complaining about wanting an easy method to solder devices but it's specialized equipment to weld those PCMCIA clam shells together.

The pincount on a processor module within a multiprocessing cluster isn't really related to all those things you mention, for a variety of interrelated reasons:

• Memory should be local to each processor module because global memory doesn't scale beyond very small processor numbers.
• I/O isn't done on processor modules because all processors need to access it symmetrically without bothering other processors.
• Realtime I/O should be handled by dedicated microcontrollers, not by a  *nix kernel which is very bad at it.
• Microcontrollers provide large numbers of GPIOs and are dirt cheap, so microcontroller-based I/O modules should do all the I/O.
• Communication between processor modules and I/O modules is best done as high-speed serial messaging.
• Serial message communication doesn't need a lot of pins and is topologically very flexible.  Remember the transputer.

I suspect we have completely different designs in mind and hence we are talking about different things.

Morgaine.