Possible Hardware fixes

selsinork wrote:

 

... As everyone keeps pointing out there's plenty of other boards out there, but none seem to be gaining the necessary momentum.

 

How much of the success of the R-Pi is down to aiming low to begin with and praying that the interest will be there ?

I think the success of RasPi is the US$25 price.  "A computer for $25?  You've got to be kidding!"  And yes, they are kidding because you can't get the $25 unit, only the warmer $35 version.  Oh, and that doesn't include a case or compatible power supply or mass storage or monitor or cables or keyboard or mouse or any of the other gewgaws you expect to have with a computer -- but you already had those, right?  So there's a big red asterisk beside that "An ARM GNU/Linux box for $25" slogan.  Must be a new meaning of the word "box" I'm not familiar with.

[I'm not complaining.  I knew exactly what I was getting when I ordered a RasPi and it's a great board for the price.  But having grown up reading MAD Magazine I dislike misleading marketing.  "An ad that's bad gets spoofed in MAD."]

Anyway, the US$25 price point gets you interested.  When BeagleBoard came out in 2008, the US$150 price point was terrific for a tiny GNU/Linux capable development board, especially compared to the US$800 (IIRC) full-size board you could get from TI.  But as Morgaine has pointed out, US$35 buys you a good dinner with shipping paying for the tip.  The risk in buying a RasPi and discovering that it's useless for your needs is pretty low.  And US$35 is usually below the level requiring spousal approval.

I imagine that the RPF is flabbergasted by the level of interest -- in a world where desktops are dominated by Windows and most consumer computers are closed consumption devices, who could imagine that you could sell 1 million uncased boards for which you have to download the operating system yourself and transfer it to an SD card.  I'm flabbergasted by the level of interest as well -- the idea that millions are becoming GNU/Lunix users is terrific and helps the future of computing.

at least you find out that the advertized $25 price isn't real before you

commit to the purchase.   Only later do you find out that your keyboard

doesn't work, or your SD card, or your power supply, or your usb hub,

or your wifi dongle, or your mpeg-2 TV tuner, etc.   And only after getting

everything to work do you find out the performance issues, that the hyped

GPU isn't used for X acceleration, that the SD cards run slow, that the

cpu overhead for USB interrupts is 20% at idle, the L2 cache isn't attached

to the cpu, etc.  And then you find out about the intermittent problems,

such as usb packet loss, SD card corruption, audio pops, failure to boot

after working fine for a week, etc.

All the front-page blog stories are happy stories, the sad stories are

in the discussion forums, and the media hasn't picked up on them yet.

I think it's interesting that half (10 of 19 by my count) of next month's

magpi interview questions have to do with hardware fixes.  So that's

clearly on people's minds.  We'll see if they actually answer those,

or just dodge them like they've been doing so far.

http://issuu.com/themagpi/docs/the_magpi_issue4_draft

John Beetem wrote:

 

I think the success of RasPi is the US$25 price.  "A computer for $25?  You've got to be kidding!"  And yes, they are kidding because you can't get the $25 unit, only the warmer $35 version. 

That's the sum total of it, it's all about price perception, and perception management which is a polite way of saying "lying".  RPF are good at it, and this seems to be why they don't like people messing them around with reasoned factual discussions.  It gets in the way of spin.

Any competitor will have to live with the same magic price point though, $25, that's the key to mass adoption.

Oh, and that doesn't include a case or compatible power supply or mass storage or monitor or cables or keyboard or mouse or any of the other gewgaws you expect to have with a computer -- but you already had those, right?

Oh it wouldn't do to point out that after buying an SD card and several USB chargers to find one that works with Pi, your total investment could be higher than for BeagleBone which comes with a 4GB micro-SD card in the box and takes +5V from a generic barrel connector with proper power management so that everything "just works" without needing to play compatibility games.

BeagleBone and Pi aren't directly comparable because each has benefits that the other lacks, but comparing price and nothing else, I spent more to get a basic 4GB system running on Pi than on BeagleBone, so my bank manager could validly claim that the BeagleBone was cheaper.

It depends on what gear you happen to have around of course, but the "$25" tag is creative marketing that verges on deceitful because that expenditure alone doesn't give you a bootable system.

Morgaine.

selsinork wrote:

 

My area of head-scratching:  how to turn the SD card into a network boot.

Are you prepared to keep an sd card installed with some minimal code on it ?  If so then perhaps u-boot. There's been some efforts to port it to the Pi and as there appears to be somewhat functional usb and network drivers it's possible to have u-boot load from the sd card then pull kernel etc from something on the network.

Aye, some people are already doing that, and it's definitely the easiest approach.  But in the "no limits" spirit of this thread, what I had in mind was bypassing the Pi's built-in requirement for the VideoCore to boot from SD card by making a networked subsystem emulate the SD card at the hardware level.

It's certainly doable in principle.  What's less certain is whether it makes sense to invest so much time though in a device with restricted documentation and many closed-source blobs.  Probably not.  There are better targets out there.

Morgaine.

At least they removed that silly picture on the front cover.

I'll not deny that the R-pi was a good market test that if there is demand for this type of boards, but the success (not sure how to define that) in the rapid volume of units ordered was essentially driven by a misleading amount of hype and confusing messages, most of them emanating from RPF and its minions. Eben still today keeps saying that you can do whatever other computers do with the R-pi, which as we know is not quite true.

When you get the beaglebone you know exactly what you are getting and all the parts used are documented and it is a true open source project, you get full Gebers, etc.

Everything around the R-pi is a mystery cloud, and if you can't ask or comment freely.

I'm almost sure that we'll see a lot of R-pis on eBay in the near future from people that got it because it was cool or some friends told them, etc, etc.

-J

selsinork wrote:

 

I could certainly be interested, but I see two stumbling blocks. First is that anything with a GPU is unlikely to have good enough docs, second is that anything without a GPU seems unlikely to have the sort of mass market appeal to get the cost down to something reasonable.

The closed nature of GPUs will change one day, and that change may not even be so far away.  Although this is just a gut feeling, it's based on observing the rapid march of progress and the evolution of devices towards ever-greater complexity.

Before long, basic graphic functionality will be taken for granted instead of being regarded as a crown jewel, and given the huge number of players in the ARM scene, I bet that someone will release a device with an openly documented graphics core.  It's almost inevitable as devices evolve and their subsystems turn into commodity functions.

Also, there is steadily growing pressure towards making graphics drivers open-source, and that pressure isn't coming only from inside the FOSS community, but also from device manufacturers themselves because having to maintain closed drivers gets in the way of making profit.  Broadcom and nVidia are in denial about it, but Intel and AMD/ATI certainly aren't.

Things are gonna change.

Things are gonna change.

When ? 

You're right that it only takes one to do it and they'll all have to..  To work though it really wants to be a shiny new competition trashing GPU that gets open drivers on the day of release. If it's an older or slower design (like the intel ones) it's not really making the case to the vendors with 10x the performance.

selsinork wrote:

 

To work though it really wants to be a shiny new competition trashing GPU that gets open drivers on the day of release. If it's an older or slower design (like the intel ones) it's not really making the case to the vendors with 10x the performance.

No no, my point was just the opposite --- that with the march of progress, what is crown-jewel functionality today is implemented as a commodity item tomorrow..  It's then no longer considered worthy of protection because it's completely lacklustre compared to the crown jewels of tomorrow.

FOSS reinforces this still further.  When  beancounters sell product which includes a commodity function that is too basic for a marketing bullet point, they are less likely to want to fund a development team to create and support its software.  Open source gives them a cheap way out, so when selecting that commodity function, they'll pick an implementation that already has software available which costs them nothing.  Their accounting will drive it.

Put both of those factors together and I expect tomorrow's "basic" functionality to include 2D and 3D acceleration and to have become open, despite the new crown jewels of the day still being myopically restricted.

Morgaine.

I think we probably agree. We'll get some basic 3D functionality in the FOSS side, and probably sooner than later.

We also seem to agree on the one constant being that progress happens and the world moves on.

So my point is simply that by the time we get that basic 2D/3D FOSS driver will anyone care ?  What's been clear in both the linux world with the likes of KDE4 and in the windows world is that they're not developing for todays GPUs, they're developing for what they hope to have in a year or two.  Performance tends to be crappy on current hardware. So does getting that basic functionality help, or is it just hindering the eventual goal ?  Oh, and I don't pretend to know the answer to that question

selsinork wrote:

 

So my point is simply that by the time we get that basic 2D/3D FOSS driver will anyone care ?  What's been clear in both the linux world with the likes of KDE4 and in the windows world is that they're not developing for todays GPUs, they're developing for what they hope to have in a year or two.  Performance tends to be crappy on current hardware. So does getting that basic functionality help, or is it just hindering the eventual goal ?  Oh, and I don't pretend to know the answer to that question

There's a directly relevant universal constant that applies here, commonly called "YMMV" and with a value of 'Always True'.

I know the answer to your question in my specific case:  there is no eventual goal, there is only the current moving target.  What's more, because I try to implement systems that make engineering sense rather than follow the latest fads and trends and eye candy, my machinery typically has almost no bloat, for example no KDE nor Gnome desktops which just get in the way of getting stuff done.

Consequently, I'm totally happy with non-bleeding edge technology, because my lean software systems run like blue light even on older hardware.  For me then, the answer to your question is an unqualified 'Yes', the basic functionality is all important.

I'm sure that others will object violently and demand 32 cores just to make their pointer crawl across their desktop, but that's their problem and I'm not a missionary out to convert them.