Pi vs BeagleBone-Black

Morgaine Dinova wrote:

 

selsinork wrote:

 

Personally I'm reasonably happy with the line being at the hardware as long as the hardware is documented well enough to allow the software to be written to use the hardware. The flaw in my idea comes when the 'hardware' really isn't hardware but is something implemented in software pretending to be hardware. You now have a blurry grey line instead of a nice crisp one.

 

It's not a blurry line as long as the following test can be applied to it:  "Is it field-programmable?"

 

Hardware/software/firmware/donutware are all immaterial in this regard.  The acid test is whether it's possible for the device to be reprogrammed in the field by anyone at all.  If a binary blob can reprogram it, then so can a user if the programming information is available.  This rule is simple, easy to understand, and clear to apply.

Indeed it is a simple rule, and comes right back to what I've been saying, if the foundation made it such that the GPU was controlled via an EPPROM and could only be updated by flashing it, then by the definition you supplied the Pi would become more open source by becoming less controllable by the end user.  If removing access to features increases how open sourced it is then there is a fatal flaw with the standard.

>No we don't need a "weighted average".  It's either fully open or it's not.  You can't be partially pregnant.

True, but you can be partially through a pregnancy, or to follow through on the concept of openness, a door or a window can be partially open.  Which is why I'm saying we should recognize how open a product is with a weighted average.  Such that a board that had 5 items that were closed could be considered more open than a board that only had one thing that closed provided the 5 things were of lesser importance.  Or to be more direct, if both boards only have the GPU closed, then the board which provides a higher amount of access to the GPU would be the more open one

Moargaine wrote:

It's ARM's loss really.  ARM took a step backwards in their excellent "profit by licensing" model when they didn't make MALI completely open and reverted to the utterly dumb "profit by secrecy" wishful thinking.  Message to ARM:  1) You failed totally. 2) Get rid of the person who suggested that  stupidity.  All it did was put a throttle on ARM's fantastic profit stream.

Actually the ARM MALI is doing very well in the embedded world, second (and climbing) only to the also fully prorietary PowerVR (used in the TI SoCs among many others).

Morgaine wrote:

Engineers tend not to be concerned with GPUs on SoCs not being open (probably because they don't use them) and so SoCs in which everything but the GPU is fully documented are regarded as "open hardware".  It's easy to understand that point of view --- why care about something in which you have no interest?

The ones that work on tablets, pads, kiosks, intelligent cell phones and any other embedded devices that use a display certainly do care about the GPUs.

Morgaine wrote:

In an ideal world everything will be open, but only the PC world has examples of that --- most Intel and AMD GPUs are fully open.

The AMD drivers are still proprietary, nVidia and Via hardware and drivers are proprietary so the PC world has examples of both.

Morgaine wrote:

In the absence of interest in openness from any ARM licensee, those who use MALI seem destined to win just because that's where the open source community is active.

There are also active efforts to reverse-engineer the PowerVR drivers so picking a winner at this time is premature at best.

I'm pretty sure ARM are ecstatic with their profit stream. Whther the mali stuff is open or not is irrelevent. Their business models work and works well. trhey dont need to change it. I don;t reckon they have made any mistakes in their business pla. You dont get chips in the majority of embedded devices with a bad business model. To say arm got it wrong is wrong. They got it right and good on em.

SOmething people seem to be misunderstand here. You don't need an open gpu.You need a gpu that supports the standard interfcaes, which means directx, opengl, and whatever flavour of acceleration google keep changing for android. ANd that's it. Because by supoprting open interfaces (or actually, just android) you win. the need for open source gpus is limited to hobbiest people liek the people here. noone who matters wants to reinvent the wheel when they are working their butts off byon guis and ***, working on gpu code. They want opengl or whatever. Let the experts work on gpus, and provide interfaces that make all gpus look the same even though they are all completely different underneath. The argument about being able to fix bugs if you have the code is bs. if their are bugs in the code, the customer get the supplier to fix them - that why the suppliers have huge teams of engineers working on this stuff. by the time the gpu get to you lot, there are so few bugs in the api's its irrelevent.

OPen interfaces is where its at. Not buried in obscure gpu code that only works on one model of gpu. You want education - teach people to use the open interfaces, there are a hell of alot more jobs there than there are in writing the gpu code itself. Thera re probably less that 10000 people in the world at any one time working on gpu code and getting paid for it. Thats not a big job market. And people who know the interfaces can move in to gpu code as and when they get experience.

John Beetem wrote:

 

selsinork wrote:

 

I think I've said it before, but I believe that the GPU on any board will always be the sticky point. None of them are any better than the rest. There's simply too many people/companies intent on protecting their 'IP' and they don't care if we believe in the concept or not. Given this stuff always tends to be a minefield of patents and cross-licensing, I can't see there being improvements anytime soon.

I think Parallella has a good shot at improving the situation.  I believe you can use the Epiphany chip as a parallel processor for GPU functions, and then use the FPGA to blast the result out HDMI.  You can also use the FPGA for GPU tasks that Epiphany doesn't do well.  If it's indeed true that the data paths exist to do this, it's just SAMP (simply a matter of programming).

 

Anyone know how the open-source Lima driver for ARM's Mali GPU is going?  (Wrote this before seeing Morgaine's comment above!)

parallella looks good, but its not got any dedicated gpu hw so wont be fast enough to replace a general purpose gpu. fast 3d? Probably? camera support, no. h264? probably need dedicated hw to keep the power requirements down. Next gen soc will need 4k encodeing decodeingstereo cameras, 40MP camera support, xbox360 graphics. Can't see a general purpose chip being about to do that and still have battery left to woek more than a few hours. There's a reason soc use so much custom hw.

mynameisJim wrote:

 

Indeed it is a simple rule, and comes right back to what I've been saying, if the foundation made it such that the GPU was controlled via an EPPROM and could only be updated by flashing it, then by the definition you supplied the Pi would become more open source by becoming less controllable by the end user.

WRONG.  My acid test "Is it field-programmable?" stands solid.

If the device can be updated by flashing it, then it's field-programmable and hence by not providing the programming information it is less open source.

Billy Thornton wrote:

 

Something people seem to be misunderstand here. You don't need an open GPU...

You do if you want to use the GPU for anything other than standard graphics.

My understanding is that most GPUs are general-purpose RISC machines with some additional instructions for graphics and image processing.  They are parallel processors with supercomputer performance.  Yes, you can use them for OpenGL and other standard graphics operations, but you could also use them for non-graphical high-performance parallel processing -- if you could get at the architecture.  For example, RasPi's BCM2835 is a high-performance media engine -- plus a cheap, low performance ARM11 I/O processor.  You can use this I/O processor for anything you want, but you can only use the supercomputer for the graphics and video functions allowed by RasPi's "binary blob".  This is like buying a Cray supercomputer -- but only being allowed to use the I/O processor.

This is why Parallella is such an exciting technology.  You can use the supercomputer for anything -- not just the applications the GPU supplier has decided to let you do.

John Beetem wrote:

 

This is why Parallella is such an exciting technology.  You can use the supercomputer for anything -- not just the applications the GPU supplier has decided to let you do.

That's an excellent comment ... +1

Morgaine Dinova wrote:

 

mynameisJim wrote:

 

Indeed it is a simple rule, and comes right back to what I've been saying, if the foundation made it such that the GPU was controlled via an EPPROM and could only be updated by flashing it, then by the definition you supplied the Pi would become more open source by becoming less controllable by the end user.

 

WRONG.  My acid test "Is it field-programmable?" stands solid.

 

If the device can be updated by flashing it, then it's field-programmable and hence by not providing the programming information it is less open source.

Okay, then get rid of my sentence about the EPPROM.  This comes right back to what I've been saying, you're literally saying that if they completely closed it down such that we couldn't do anything to it then it becomes more open source.  Let's say that they accomplish this by having a bit of code built directly into the chip that handles it booting right off an SD card.  No more configuration of the GPU, no more camera module add-on without purchasing a new board with the feature, but since it's behind the scenes it's considered more open source?  By it's very nature the more control we have of something the more open it is.  The current definitions we're working with are flawed as they encourage manufactures to put more info where we can't touch it in order to get the stamp of open source approval instead of actually rewarding those who make the effort to give us more control over the equipment.

mynameisJim wrote:

 

The current definitions we're working with are flawed as they encourage manufactures to put more info where we can't touch it

You're confusing two things:  being more open, and having more functionality.  They're completely and utterly different concepts.

Sure, those who want to use advanced but closed functions will praise their availability, but this has absolutely nothing to do with the hardware being more open or more closed.  You're looking at two entirely different types of stakeholder.

The open metric is not flawed, it's just not valued highly by those who don't rate openness highly.  Doh!

John Beetem wrote:

 

My understanding is that most GPUs are general-purpose RISC machines with some additional instructions for graphics and image processing.  They are parallel processors with supercomputer performance. 

The place I work builds (amongst other things) lots of HPC clusters, you'd be surprised how many of them get a little 2U nvidia box for every node these days.