My Thoughts on the Pyra

[Granitehead]

That's more expensive than the OMAP5, which is one of the most expensive SoCs out there.

MediaTek and Allwinner are usually lower than $10...

 Maybe Allwiner could offer Linux drivers if you pay those SoCs at OMAP5 price or higher. Or maybe you could even buy the drivers.

They would probably supply the drivers for free if he bought larger quantities, but at the quantities he is at, he can do nothing with money.

 

[Kiga]

We dont know... If there isn't linux drivers yet, someone from Allwinner could make them if he is paid, for exemple 720p screens are twice cheaper, so there is enough money I assume.

 

[Granitehead]

Money which could be spent making the rest of the Pyra better or recovering from potential disasters.

And it would probably need more money to pay them to release Linux drivers than ED spent on the whole development of the Pyra.

 

[Kiga]

Money which could be spent making the rest of the Pyra better or recovering from potential disasters.

What worth "the rest of the Pyra" without power? The rest of the Pyra can be improved following years, but as all the software optimisation will be done for the SoC, the best SoC have to be chosen because it will last for years.

And it would probably need more money to pay them to release Linux drivers than ED spent on the whole development of the Pyra.

What do you know about that? If they don't want to develop Linux drivers because they don't need them yet and that someone pay them for that and buy them units, why would they ask for such an amount of money?

 

[comradekingu]

Why pay a company that doesn't care enough to have linux drivers already?

I understand the lower end stuff from allwinner with Mali, the A31 was just mainlined, with its PowerVR stuff, it cant (or probably wont) ever be good, the A80 has PowerVR G6230 hardware, which is even worse.

They are *powerless* to do anything about it.

Spending money on allwinner getting things better, when that would put it back to square one as far as TI is concerned _is_ the disaster. The stuff they have that fixes the powerVR problem, is slower again, so there is no all good preposition here.

A80 is powerful at benchmarks, in Android... Why go blind for power? To rule the latest synthbench?

What does that extra power do? Make for a less weak but still not powerful desktop machine? We _all_ have those, already. Exceed the power envelope a plastic handheld can have?

TI is a reputable company, with support everywhere, allwinner isnt, you will be hard pressed to get ahold of them, same with finding info.

4GB ram i understand, atleast that has some use that i can think of.

 

[thatgui]

I want to run my Pyra with desktop, web and game applications and -two- Android 'virtual machines' (one for business/work apps and one for games) and an emulated X86 command line desktop - maybe running Desqview to multitask DOS applications there too.

I have a hard time believing that RAM will be the bottleneck in that scenario.

And it would probably need more money to pay them to release Linux drivers than ED spent on the whole development of the Pyra.

What do you know about that? If they don't want to develop Linux drivers because they don't need them yet and that someone pay them for that and buy them units, why would they ask for such an amount of money?

Do you have any idea how much a dev wants for his work, especially a very capable one that is needed for driver development?

 

[ekianjo]

Do you have any idea how much a dev wants for his work, especially a very capable one that is needed for driver development?

This. That's why nobody spent much time on drivers/hardware support on Linux, until Valve started paying people full time for it. Linux specialists cost A LOT.  

 

[Kiga]

What do you know about that? If they don't want to develop Linux drivers because they don't need them yet and that someone pay them for that and buy them units, why would they ask for such an amount of money?

Do you have any idea how much a dev wants for his work, especially a very capable one that is needed for driver development?

If I knew it I wouldn't ask "what do you know about that?", anyway I brought that topic to know if it is possible, never assert anything. However as you guys  seems to know, you better tell us how much.

 

[thatgui]

If I knew it I wouldn't ask "what do you know about that?", anyway I brought that topic to know if it is possible, never assert anything. However as you guys  seems to know, you better tell us how much.

Sorry, thought this was a rethorical question.Keep in mind that this isn't a product they expect to have high demand for, so they probably will try to cover their development costs with this one sale. And developing drivers isn't something average Jo programmer will come up in two months, you need somone which is much more specialized (and probably more creative), which costs extra. And there is the possibility that they need to pull in external help due to not having enough specific knowledge available inhouse or simply not having the manpower to lift that project without slowing inhouse project down, which again costs extra. And they sure want to gain some profit out of it as ED has no big purchase numbers that support a prospect of a very fruitfull business relationship on their side.

Hard to tell exact numbers, but I expect development costs at least in the range of beeing almost in the six digit number range, if ED is lucky

 

[Kiga]

The fact is nobody really know.

I knew that it cost a lot that's why I gave an exemple with 720p screens that allowed to save at least $250.000 which was for buying useless pixels.

I think it would be much less hard to develop Linux drivers as they already developed for Android and iOs. And maybe Allwinner already plan to get Linux driver later, otherwise why did we mind about A80? So they wouldn't ask too much.

 

[_wb_]

The TI OMAP5 EVM has 2GB of RAM, which is implemented using 4x Micron 4Gb DDR3L devices (MT41K256M16HA-125:E).

DDR3L operates at 1.35V and each of those 4Gb (512MB) chips draws 20mA if it is doing nothing more than self-refresh. So when idle, 2GB of this type of RAM should consume about 1.35V * 20mA * 4 = 108mW.

For comparison, with the lid closed and cpu idle, a Pandora uses slightly less than 100mW. That number includes not just its 512MB of RAM but also all the other stuff.

The Pyra battery will be bigger than the Pandora one, and the Pandora has a standby time of about 7 days (4200mAh * 3.7V / 95mW).

Let's say a 2GB Pyra uses 150mW when standby (108mW for RAM and the rest for the other stuff). With a 6000mAh battery that results in 6 days of standby time.

If the same type of RAM is used, 4GB will consume twice as much, so let's say the total is 250mW. In that case, standby time is reduced to less than 4 days.

So the effect on standby time is quite big.

One potential solution would be to use suspend-to-disk instead of suspend-to-ram to achieve better standby times. But the eMMC will probably not be large enough to do that, so you would have to use free space on an SD card. The theoretical maximum transfer speed we can achieve is about 100MB/s. That means that suspending and resuming takes about 20 seconds each if 2GB needs to be transferred between RAM and SD card (it could be less because only used memory needs to be stored/restored), and of course twice as long if 4GB has to be transferred. In practice it would probably be slower than that because SD cards that max out the theoretical speed limit are very expensive. So that does not look like an acceptable solution.

I would rather spend money on more expensive lower-power 2GB RAM, than on 4GB. In general, I would like to make sure that the Pyra improves upon the Pandora in all respects, instead of having it improve some aspects (amount of RAM) by sacrificing other aspects (standby time).

 

[EvilDragon]

Those 250K USD are not saved - they are used elsewhere (i.e. keymat is a lot more expensive) and they keep the price in a proper range (makes them 50 EUR cheaper).


The price will be calculated from production costs, higher costs leads to a higher price leads to less sales.


The current calculation covers the production costs with at least 3000 sold units - that's what we at least need to sell.


The lower we can keep the price, the better our chances are.

 

[thatgui]

The fact is nobody really know.

I knew that it cost a lot that's why I gave an exemple with 720p screens that allowed to save at least $250.000 which was for buying useless pixels.

I think it would be much less hard to develop Linux drivers as they already developed for Android and iOs. And maybe Allwinner already plan to get Linux driver later, otherwise why did we mind about A80? So they wouldn't ask too much.

What the point of this ? Of course no one can really know how much it actually will costs, the only possibility would be to ask them directly, which alone will be hard thing to do for someone that wants no business with them. But money alone isn't the problem, no one knows wether they have any interest in doing this at all, they are not offering software development service for all I know. So they may not even by inclined to put any thought into it as they probably have their own development schedule they need to follow.All I can offer is my personal eight year experience in the software development industry, which I spend four years as a embedded software engineer in the industrial automation sector. I know what my company took from their customers for me spending a day at their site fixing problems, and developing drivers for gfx chips is a whole other level.

 

[Kiga]

Those 250K USD are not saved - they are used elsewhere (i.e. keymat is a lot more expensive) and they keep the price in a proper range (makes them 50 EUR cheaper).


The price will be calculated from production costs, higher costs leads to a higher price leads to less sales.


The current calculation covers the production costs with at least 3000 sold units - that's what we at least need to sell.


The lower we can keep the price, the better our chances are.

Of course they are not, and fortunately there isn't any rotator chip for the 1080p screen.

What the point of this ? Of course no one can really know how much it actually will costs, the only possibility would be to ask them directly, which alone will be hard thing to do for someone that wants no business with them. But money alone isn't the problem, no one knows wether they have any interest in doing this at all, they are not offering software development service for all I know. So they may not even by inclined to put any thought into it as they probably have their own development schedule they need to follow.

I thought about all these hypothesis and I asked about the idea mostly to have details, it's still as vague and I tend to think that it's feasible.

All I can offer is my personal eight year experience in the software development industry, which I spend four years as a embedded software engineer in the industrial automation sector. I know what my company took from their customers for me spending a day at their site fixing problems, and developing drivers for gfx chips is a whole other level.

And I thank you for that.

 

[rohezal]

A friend of mine works as a sysadmin and linux developer. He takes 80 Euros per hour. If he needs a half year for linux drivers for the chip thats:

8 hours a day

5 days a week

4 weeks a month

6 months

5*4*6*8*80 = 76,800 Euros which are over 100,000$.

Craig paid some money for a better wifi chip driver for the Pandora if I remember me right. Not sure what it was but it was far over 5000$.

 

[Granitehead]

And I would expect making Linux drivers for an ARM SoC is way above the paygrade of a sys admin.

Craig paid some money, yes - but if I remember right he got nothing for it, that was just to look at the problem and maybe make a quote for the actual work.

 

[fusion_power]

The current calculation covers the production costs with at least 3000 sold units - that's what we at least need to sell.


The lower we can keep the price, the better our chances are.

I'm sure the Pyra could sell much more than 3000 Units. Let's say 10k Units, this could drop the price noticable I guess. ^^

A friend of mine works as a sysadmin and linux developer. He takes 80 Euros per hour.

Wow, and people pay him that much? Not bad, he must be rich. Almost funny when I remember, here in germany they discuss about embrassing 8.50€/h minimum Wage and Companies try to even underrun that. in Practise, it's sometimes not even 5€/h Wage.

 

[Granitehead]

A friend of mine works as a sysadmin and linux developer. He takes 80 Euros per hour.

Wow, and people pay him that much? Not bad, he must be rich. Almost funny when I remember, here in germany they discuss about embrassing 8.50€/h minimum Wage and Companies try to even underrun that. in Practise, it's sometimes not even 5€/h Wage.

Someone capable of coding drivers for an ARM SoC probably gets 1.5 times or twice that, I would guess.Those people need to be /good/, and you pay for that.

 

[rohezal]

He lives in Germany, like I do. But he is good . The average SysAdmin with his own busisness (Selbstständiger) earns around 50 Euros. The cheapest one I know got 35 Euros / hour and he was a machine builder engineer (Maschinenbauer der ein bißchen in Linux rumhacken konnte).

 

[Grench]

x86 is -not- more expensive on an SoC to SoC basis.  The flaship Z3795 is $40 per SoC in tray prices.  

http://ark.intel.com/products/80267/Intel-Atom-Processor-Z3795-2M-Cache-up-to-2_39-GHz

That's more expensive than the OMAP5, which is one of the most expensive SoCs out there.

MediaTek and Allwinner are usually lower than $10...

That was for the current flagship though.  The list prices for the Z3000 series start at $17 per SoC.

http://ark.intel.com/products/series/76761/Intel-Atom-Processor-Z3700-Series

It's 100% irrelevant to this project at this stage though - so onward.