SoC: Back and forth!

[EvilDragon]

He has a very legitimate fear that if he tries to go x86 demand will exceed what his company can currently support, both in terms of gross numbers and an increased percentage of returns for reasons of "it doesn't do what I thought it would". I don't think the number will be nearly as high as he seems to be afraid it might be but it will still not be zero, I've seen people complain about multi-thousand dollar Apple laptops that couldn't handle desktop like graphics in Windows games, people that just didn't do their research before making a purchase, and I have no reason to believe that those same people won't jump on a $600 palmtop that advertises the ability to run some Windows games natively.

 Your full post pretty much sums up my current fears.

Right now, I'm a company that consists of me and Askarus (with minimum wages).

To handle a large amount of shipping and production, I would suddenly need a LOT more workers.

Those cost a lot of money.

When initial peak of sales goes down, demand will go down as well and then I have a lot of space I need to pay rent for as well as workers.

Additionally, in Germany you can test a product and send it back within 2 weeks of testing - without any reason.

I have to pay the FULL amount back, including shipping back and forth AND have an already opened box which needs to be completely repackaged as well.

This could lead to tremendous costs - a LOT to handle. A lot more than I can.

I rather have a steady growing company, i.e. if Pyras sell well, increase the size with more space and workers and work on the next product.

Companies that grow too fast can easily explode...

Intel would not stop their production line, but they can refuse to sell us any socs some time down the line. Pyra is too small a project for them to care.

Do you even read, what people write? There is no need to buy anything from Intel directly. You can buy those SoCs from resellers (even in single quantities). Intel would have to stop selling their chips to all resellers, which is extremely unlikely to happen. Why would they do such a thing?When choosing the right SoC, there are a lot of things to consider. That's for sure, but this is not one of them.

If that's true then why can't Ed get anything from samsung? It's my fear that Intel could someday just say screw those pandora hipsters, and we would be getting bugger all.

Samsung (as well as all other companies in the mobile area) decided that only selected people can buy their SoCs and get the NDA'd documents.

Intel didn't. Intel offers it to anyone, you don't need to apply to them.

That's a huge difference.

As I understand it, the OMAP comes with support from TI.   Does Intel offer similar support to companies who but their chips through a reseller?  does the reseller?  how does that work?

Yes and no.

TI doesn't officially give us support, but I know quite a few people of the OMAP5 dev team, and they would try to help us in their spare time as well.

 

[Hồng Thất Công]

One thing I haven't thought about till now, the Shield is subjectively the Pandoras biggest competitor. (I know lots dont see it as a competition for various reasons) but could agree at least in form factor, architecture, and several use cases if not most. It doesn't seem to go away anytime soon and will keep pushing out revisions as new tegras come out. While it may not totally or even partially eat into your desired sales it is for some a point of fairly similar comparison. X86 would re-solidify the pandoras niche and compare it to things like the razor switchblade / blade which would be so loosely comparable its almost not worth mentioning.

You can't stop people from comparing, but just as with the Pandora, Pyra is not going to try to compete against or try to be like the best and latest gadgets out there.  Talking about re-solidify the  pandora niche, since when the pandora loses its niche so that it needs to be re-solidified?   

 

[Hồng Thất Công]

Of course Intel offers support for their products. It does not matter where you buy the chips eventually. They even offer things like analyzing your schematics.

Many current devs will quit as well

Maybe so, maybe not. You probably all remember that ED has recently been presenting the Pandora (and the Pyra) at FOSDEM, which is a pretty large open source event with lots of skilled people around, with many of them not (yet) being part of this community. With the Pyra being an interesting device for Open Source developers, it should certainly be possible to get some of them interested in developing for the Pyra (including low-level stuff like kernel development), especially if its hardware is open source friendly.

Yes intel will provide support, but who's the technical guy to interface with them?

See? You are HOPING that a bunch of smart people will come and automatically provide support for the PYRA.  What if they don't?  Or if they come 2 years later who's gonna deal with the initial issues?  RISK again. 

 

[wej]

automatically provide support for the PYRA

I did not say they would come automatically, and I certainly did not mean that. You have to at least encourage them somehow or simply ask, maybe offering Pyra development boards to skilled people, like it was the case with the Pandora. What I was trying to say is, that you should probably not only look for developers inside this community. I'm not saying there are no skilled people in this community, quite the opposite is true, but there are lots of other people out there too.

 

[Askarus]

Many current devs will quit as well

Maybe so, maybe not. You probably all remember that ED has recently been presenting the Pandora (and the Pyra) at FOSDEM, which is a pretty large open source event with lots of skilled people around, with many of them not (yet) being part of this community. With the Pyra being an interesting device for Open Source developers, it should certainly be possible to get some of them interested in developing for the Pyra (including low-level stuff like kernel development), especially if its hardware is open source friendly.

Lots of the FOSDEM visitors did like the OMAP 5.

No one did mention that an other chip might be better (except the Power VR issue).

Didn't hade a single conversation where one asked for x86.

 

[Hồng Thất Công]

 @wej:  Ed does not know who they are and does not have any prior working relationship with them to have enough trust that they will commit themselves to help ED.  On the other hand, Ed has a long relationship and trust with notaz and Exophase and the Pandora team.  Big RISK.

Also, how many PYRA can he give out for free in order to attract the PROPER and TRUSTWORTHY people to help him?  What if they just took the PYRA and run away?  DragonBox is not Sony, Nintendo and or Microsoft.

 

[rohezal]

I want to mention 3 things.

First: If we use a x86 cpu all pnds so far will stop working (maybe not the java ones).  How do we use the pnds with an omap when we use 64 bit hard floats? Can we use a compatiblity layer (if possible)? Or will we use "softhardfloats" again? In the future, the PND should contain its source code, so the repo can build packages for the different devices.

Second: With x86 linux you can a lot more libs and tools which are closed source (like unity). You get untity and games like Baldurs Gate 3  (Project Eternity). And the thing is fast enough for PS2 emulation, looks like it is to slow for full speed emulation.

Third: Exophase prefered x86 as far as I remember. Notaz wanted to stay at arm as far as I know. If Notaz would leave if ED chooses a x86, I would prefer to stay at arm.

 

[Natsu]

OMAP 5 to OMAP 6 might work.

That's also a reason why I would like the OMAP.

We can use further TI SoC.

OMAP 6 isn't even a guarantee.

I would not go the way to offer a Version with SoCs of different companys.

OMAP 5 to OMAP 6 might work.

But no OMAP to Snapdragon or vice versa upgrade.

Why wouldn't it work?

 

[Hồng Thất Công]

...

Third: Exophase prefered x86 as far as I remember. 

Didn't know this.  No wonder why he's MIA

 

[wej]

Lots of the FOSDEM visitors did like the OMAP 5. No one did mention that an other chip might be better (except the Power VR issue). Didn't hade a single conversation where one asked for x86.

That was to be expected. ED presented the Pyra with an OMAP5 development board, so the SoC choice seemed to have been made, so there was no reason for people to ask about other possibilites. The PowerVR issue is one of the biggest concerns with the OMAP for me as well. For me it does not matter that much if the Pyra uses ARM or x86, but it does matter if the hardware used is well supported by the Linux kernel and quite frankly the state of the Pandora kernel is pretty sad (still stuck with the ancient 3.2 kernel, lots of important stuff not mainlined).

 

[Natsu]

Additionally, in Germany you can test a product and send it back within 2 weeks of testing - without any reason.

I have to pay the FULL amount back, including shipping back and forth AND have an already opened box which needs to be completely repackaged as well.

There has to be a limit to this, right? Otherwise they can break the product and then send it back.

 

[fusion_power]

To handle a large amount of shipping and production, I would suddenly need a LOT more workers.

Those cost a lot of money.

When initial peak of sales goes down, demand will go down as well and then I have a lot of space I need to pay rent for as well as workers.

No matter which chip is choosen, this could happen anyways if the Pyra gets popular. Did you have prepared for the "worst case" scenario where the Pyra gets so popular that hundredthousands of people suddenly want to buy one? It would be actualy a good thing because the device may deserves it and it could make you rich, but of course as a small company this also could be a dangerous thing like you described. It seems everything has two sides. ^^

First: If we use a x86 cpu all pnds so far will stop working (maybe no the java ones).  How do we use the pnds with an omap when we use 64 bit hard floats? Can we use a compatiblity layer (if possible)? Or will we use "softhardfloats" again? In the future, the PND should contain its source code, so the repo can build packages for the different devices.

Wasn't this so or so the case? Pyra get's a new file format and all PND's have to be recompiled anyways to make them work onto the new Handheld? I mean, if the old PND's could run "out of the box" onto the Pyra if an OMAP and not an X86 is choosen as chip, then we don't need to discuss anymore because then the OMAP would be the ONLY choice for the Pyra of course. Nothing is more important for a new system than an existing software base. Starting basicly from zero is dangerous (hello Wii U! ) but with the solid PND base of the Pandora, the Pyra would have one big advantage.

 

[Linux-SWAT]

Lots of the FOSDEM visitors did like the OMAP 5. No one did mention that an other chip might be better (except the Power VR issue). Didn't hade a single conversation where one asked for x86.

That was to be expected. ED presented the Pyra with an OMAP5 development board, so the SoC choice seemed to have been made, so there was no reason for people to ask about other possibilites. The PowerVR issue is one of the biggest concerns with the OMAP for me as well. For me it does not matter that much if the Pyra uses ARM or x86, but it does matter if the hardware used is well supported by the Linux kernel and quite frankly the state of the Pandora kernel is pretty sad (still stuck with the ancient 3.2 kernel, lots of important stuff not mainlined).

3.2 is not old as it is LTS.

Of course being mainline would be better. I encourage you to make patches.

 

[EvilDragon]

Additionally, in Germany you can test a product and send it back within 2 weeks of testing - without any reason.

I have to pay the FULL amount back, including shipping back and forth AND have an already opened box which needs to be completely repackaged as well.

There has to be a limit to this, right? Otherwise they can break the product and then send it back.

Yes, of course, it has to be fully working.

But I still have to pay for the shipping costs both to and from them as well and they can't be sold as "new" anymore. 

Lots of the FOSDEM visitors did like the OMAP 5. No one did mention that an other chip might be better (except the Power VR issue). Didn't hade a single conversation where one asked for x86.

That was to be expected. ED presented the Pyra with an OMAP5 development board, so the SoC choice seemed to have been made, so there was no reason for people to ask about other possibilites. The PowerVR issue is one of the biggest concerns with the OMAP for me as well. For me it does not matter that much if the Pyra uses ARM or x86, but it does matter if the hardware used is well supported by the Linux kernel and quite frankly the state of the Pandora kernel is pretty sad (still stuck with the ancient 3.2 kernel, lots of important stuff not mainlined).

The PowerVR is not the reason the Pandora can't be upgraded to a new kernel.

Repeating: The kernel part of the PowerVR IS OpenSource.

There's just no one who really worked on the kernel except for notaz.

He usually checks the more recent kernels for changes that are useful and what has been changed / removed and then decides whether it's worth to switch to the kernel or not.

A new kernel can be ported with all the hardware to the Pandora (why don't you do it, if you want that kernel?)

What feature are you missing on the Pandora that a new kernel would give us?

AFAIR, the kernel developers have thrown out a lot of OMAP3 stuff for some reason (not sure about that, but I remember notaz mentioned something).

No idea why, and it doesn't make sense to me, but then again, a lot of decisions of the kernel developers don't make sense to quite a few devs at all...

 

[Michoko]

He has a very legitimate fear that if he tries to go x86 demand will exceed what his company can currently support, both in terms of gross numbers and an increased percentage of returns for reasons of "it doesn't do what I thought it would". I don't think the number will be nearly as high as he seems to be afraid it might be but it will still not be zero, I've seen people complain about multi-thousand dollar Apple laptops that couldn't handle desktop like graphics in Windows games, people that just didn't do their research before making a purchase, and I have no reason to believe that those same people won't jump on a $600 palmtop that advertises the ability to run some Windows games natively.

 
Your full post pretty much sums up my current fears.

I can easily understand your concern, ED. And I find this topic interesting because it exposes many critical choices. However, if this fear is related to the fact people may make a purchase based on a misunderstanding (and then ask for a refund), isn't it where your communication and marketing campaigns will play an important role? Of course, if you sell the Pyra as a "Powerful gamer PC in your pocket", it might be misleading. But if you communicate more about a versatile device, based on open-source and Linux, you are immediately targeting another audience, with a different mindset than the regular PC gamer IMO. If you orient your communication wisely, you may still be able to target the "power users" without going too mainstream.

 

[canseco]

It's best to say Personal Computer than PC this days, because publishers and media made the word synonymous to MS Windows...

Anyway, i'm a gamer and i use Linux, almost same mindset as when i used Windows,

 

[EvilDragon]

I can easily understand your concern, ED. And I find this topic interesting because it exposes many critical choices. However, if this fear is related to the fact people may make a purchase based on a misunderstanding (and then ask for a refund), isn't it where your communication and marketing campaigns will play an important role? Of course, if you sell the Pyra as a "Powerful gamer PC in your pocket", it might be misleading. But if you communicate more about a versatile device, based on open-source and Linux, you are immediately targeting another audience, with a different mindset than the regular PC gamer IMO. If you orient your communication wisely, you may still be able to target the "power users" without going too mainstream.

Well, the problem is, there won't be a huge marketing campaign (as we would never have the money for that).

It would be similar to the Pandora - mouth-to-mouth propaganda, social networks, press releases.

And that's where the issue is - if one has the idea it is basically a full x86 PC, these "false" (not really false, but you know what I mean) marketing slogans could go around the world.

 

[Granitehead]

I can easily understand your concern, ED. And I find this topic interesting because it exposes many critical choices. However, if this fear is related to the fact people may make a purchase based on a misunderstanding (and then ask for a refund), isn't it where your communication and marketing campaigns will play an important role? Of course, if you sell the Pyra as a "Powerful gamer PC in your pocket", it might be misleading. But if you communicate more about a versatile device, based on open-source and Linux, you are immediately targeting another audience, with a different mindset than the regular PC gamer IMO. If you orient your communication wisely, you may still be able to target the "power users" without going too mainstream.

Well, the problem is, there won't be a huge marketing campaign (as we would never have the money for that).

It would be similar to the Pandora - mouth-to-mouth propaganda, social networks, press releases.

And that's where the issue is - if one has the idea it is basically a full x86 PC, these "false" (not really false, but you know what I mean) marketing slogans could go around the world.

Well, but you would still be the one in control of selling them. So you could make it very clear on your website what the Pyra is capable of and what not. At the moment you aren't doing this on Pyra-Handheld.com. But if you fear this kind of thing it would be trivial to add another few sections which don't promote features, but talk about shortcomings, stuff the Pyra can't do, etc.(edit) You really shouldn't limit the Pyra's success with technical decisions. You don't need to sell to everyone, if your problem is too much success you can do lots of unusual (because usually company don't do them) things to keep success down. Hell, you could have mandatory quizzes before allowing anyone to buy one, if you really fear uninformed buys. You don't know the difference between ARM and x86? You aren't able to go look up whether the Pyra has enough RAM for game xxx? Go fuck yourself, you can't buy one. You could do that, ED, and no one could do anything about it.

But unlike a technical decision which will be fixed for years and years, you could change that at any time, to slowly grow, exactly like you said you want to.

(edit)

So that's why I think the decision should really be based on technical aspects and availability of good software and willing coders, etc., but not because you fear demand could be too high.

 

[Hồng Thất Công]

Merry-go-round

 

[wej]

3.2 is not old as it is LTS.

LTS does not make it less old. It does not get any new features. Also LTS support ends eventually.

The PowerVR is not the reason the Pandora can't be upgraded to a new kernel. Repeating: The kernel part of the PowerVR IS OpenSource. There's just no one who really worked on the kernel except for notaz. He usually checks the more recent kernels for changes that are useful and what has been changed / removed and then decides whether it's worth to switch to the kernel or not. A new kernel can be ported with all the hardware to the Pandora (why don't you do it, if you want that kernel?) What feature are you missing on the Pandora that a new kernel would give us?

Even though the kernel part of the graphics driver is open source, the PowerVR driver still enforced the use of soft float in the userland. So even when the kernel part of a driver is open source, it still can do harm. Sure, when you just use the Pandora OS, you probably don't have to care much about the kernel unless something does not work properly or you want to attach a USB device which is supported by a newer kernel but not the 3.2 kernel.

As soon as you try to use other operating systems, especially bleeding-edge ones, you can easily get into trouble with older kernels because they simply do not support certain features required by the userland/init system/whatever.

Using a SoC with good mainline support to begin with, could probably make the kernel development part easier. Additionally when carefully designing the hardware, one could avoid the need to develop additional kernel drivers at least for some parts entirely, e.g. using USB HID for the input devices instead of attaching them via GPIO, SPI or I²C. If I remember correctly, especially the way the Pandora's keyboard was implemented was problematic in such a way that the hack needed to properly support it (I think it was the way the FN button functions were read) in the kernel was unacceptable for the kernel developers so there was no way of getting it mainlined (correct me if I'm wrong here).