Comprehensive Status update!


Wow, what trip!

Honestly, when work had started on the Pyra, I never, ever thought it would take so long and we'd run into so many issues again! Seriously!

While I was part of the Pandora team, I wasn't involved very deeply into the whole design / production process (until I moved the production to Germany).

I was mainly the one speaking to the community (with as many details as I was allowed to), but this time, I didn't only do that but was responsible for all the decisions and handling all the various companies and involved persons... as well as the funding.

This surely wasn't easy, it was a lot more complex than I feared.
It was exhausting, and partly discouraging (especially when the first prototypes we produced didn't really work at all!)

I am more than happy now that things FINALLY seem to come together, and I want to give out my respect to EVERYONE who was involved with the design process of the Pyra - and with the Pandora!

It was an achievement, a HUGE achievement! Probably much bigger than most would ever think. But having experienced the design and production of such a complex device with all the issues myself, I can only show respect.

A huge thanks to Nikolaus, Fatih, my contacts in Greece and Austria, Global Components, everyone involved with the software development for the Pyra (OS, Kernel, anything), all investors, the prototype-preorders and of course: You.
Even though it took much longer than planned, I only had a couple of refund requests - and most of them were people who had some misfortune in life and urgently needed the money.

I am REALLY grateful for your support and your patience, I am happy you've been with me here during the development, kept us going and the spirit alive.

I've got huge respect and thanks for everyone involved and for you for enduring this tiresome trip.
Nobody can take away the experience and knowledge you gained from this and you have the makings of becoming one of the big players some day if you choose to.

1187.gif
 
Great work ED, I've been following this (and the Pandora) for a long time, really great to see it come to fruition - your attitude and work ethic are really something you should be immensely proud of.

Looks like the GPD Win is about to start shipping, it will be interesting to see how reviews go.
 
What might still need some tweaks is the DPad - as I'm very, VERY picky with that
really? when did this start : O )

seriously though - what an immense achievement, what a staggering piece of kit, who on earth has heard of a replaceable CPU in something the size of a phone! and the almost bewildering array of output and input options, not to mention storage... wow... and in what a few years, many large organisations would throw serious cash at a project like this to get it done in that time...

... all that and thankfully ED hasn't managed to drive himself into an early grave....
 
Thank you for the update. I've liked all of them, but generally get the impression I learn from some details and still don't have the whole picture (because it's a complex project).
This update is even better in that it gives me a whole picture impression.

[...]
1. The status of the PCBs
[...]
About half of the CPU PCBs will be populated with the 4GB memory RAM so we can send those to developers so they can enable them in the kernel and they'll hopefully work without any issues.
There's not any reason why the memory shouldn't work, so I'm positive about that.

I don't know anything about electronics but I've heard that memory is one of the most difficult components because
it uses wide, very fast buses, and any slight discrepancy in track length or width may cause problems (maybe to one
chip and not another). Also, at least for PCs I've heard getting RAM timings right is also very tricky (but it may be
2x or 3x the Hz than the Pyra). And you can't really be sure they work until you get the timing right in u-boot (or the kernel?).
The other risk is if the 4GB chips draw more power than 2GB chips, and that somehow requires some change, but I'm not even
sure, since it's 2 chips for 4GB and 4 chips for 2GB. They might even draw less power ? No, I guess it's still double the bits
to refresh periodically but what do I know?

So I wonder how long this will take. Having it working with 2GB is a hint PCB changes may not be needed but I'm not
sure it can be completely ruled out. In fact having the board designed by Nikolaus is an even surer hint no changes
will be needed. So I'm not really worried, just not so confident it'll be sure enough in a short time. Let's hope so.

I somehow thought there was one prototype already built with 4GB from this june posts.

We ordered the samples a while ago while working on other things.

Nikolaus picked up the sample RAM at GC yesterday and already populated the CPU boards with them today, so tomorrow he can test if that all worked fine.



So if all goes well, we will know this week whether the Samsung Chips work fine.

We have also put the new 4GB RAM chips on one of our existing CPU PCBs. Once the new timings have been implemented into the bootloader, we can test these as well.

But now it seems like the 15 4GB prototype CPU boards to be populated soon will be the first ?
Maybe I'm not understanding, or you just mean the first for this revision of the PCB (with the 1.5 GHz fix),
but 4GB is already tested in another revision.

Once the CPU boards have been thoroughly tested, we can order the needed CPU boards for the mass production run!
We can even do that while developers are working on the 4GB RAM, as the layout is exactly the same.

As the current CPU boards work well already, this could happen pretty fast - but of course, we won't rush it (we don't need 2000 non-working CPU boards :))
clap, clap.

That's what's more lovely in this project, hard work, no rushing and transparency.
I trust you won't order the 2000 prototypes until you're sure they'll work with the 4GB RAM.

Once the mainboards have been tested, we can order them for the mass production as well. They're not as complicated as the CPU boards, so the production only takes half of the time - but we still would need to be very very lucky to get them produced in time for christmas (we'll try our best, but time flies, damn!)
And especially as production ALWAYS takes longer before christmas (as ALL companies want to produce their stuff for christmas), it doesn't make things easier.

Maybe it's not worth it rushing for Christmas. If it's for Christmas fine, if it's for New Year's Eve fine, if it's for FOSDEM, fine...
This is great hardware, let's not risk a flaw for rushing.

a) Source and order heat dissipating material:
I've already found a few companies who are offering such material, just didn't have the time to contact them yet.
The production shouldn't take long though, as we simply need pre-cut rectangles from a self-adhesive heat-dissipating material which will be put into the case.
The thickness can be up to 0,5mm, most of the materials are 0,3 - 0,4mm thick.
What I also want to find out is what kind of material would be best for our use case - but hopefully, an experienced company can help us out here.

I didn't know it was needed. Sounds even better with it, but I thought it was said the Pyra was already cooler than other similar size devices ?
Will it be for the battery or the CPU board ?

c) Calibrate the Wifi antenna
The Wifi antenna needs to be calibrated using a proper resistor value. So basically, what needs to be done is to have a standard setup (Wifi router at a certain distance) and change the resistor value until you have the best possible reception.
It's nothing too complicated, just takes a bit of time - and Nikolaus can do that once he finished the mainboard layout and while we're waiting for the prototype PCBs to arrive.

I think I'm going to be silly as hell since I know nothing of electricity, electronics, electromagnetism or radio.
But if the resistor value depends on the distance to access point or environment where the Pyra is put, would it make sense to use a potentiometer instead of a resistor
and let software control it so that people can maybe experiment for best setting in different places ? Even if you need to find a sane default.
I guess that would need changes to the mainboard PCB, so maybe not worth it. But then if it's on the mainboard PCB that needs that change for the battery connector...


d) CE-Testing
Can't be done before we have the prototypes finished. Usually takes one day.

Sorry for my ignorance again. Is this something sure to pass or might it fail it and require more changes ?
Maybe it's just something you have already tested in house but need a third party testing for certification, so it's just kind of paperwork.
Or maybe it's something that may require some shielding or any changes if the finished prototype does not pass the test ?

e) Produce the box and packaging and manual
We already decided for the box (remember the samples I showed to you months ago), so we only need some box design and then produce it.
Production takes 3 weeks, so I plan to have that finished early November.
the boxart should be VERY simple, nothing fancy. Nice, clean and beautiful :)

I can't help in aesthetics and I don't care, so I'll leave it for others...



Reason why I think the box should be wine red is not only because of the first images of what the Pyra could look like, but also because....just as wine is something that gets better with age, so does the Pyra.

like this. Good idea.

Honestly, when work had started on the Pyra, I never, ever thought it would take so long and we'd run into so many issues again! Seriously!
[...]
I am more than happy now that things FINALLY seem to come together, and I want to give out my respect to EVERYONE who was involved with the design process of the Pyra - and with the Pandora!

It was an achievement, a HUGE achievement! Probably much bigger than most would ever think. But having experienced the design and production of such a complex device with all the issues myself, I can only show respect.

A huge thanks to Nikolaus, Fatih, my contacts in Greece and Austria, Global Components, everyone involved with the software development for the Pyra (OS, Kernel, anything), all investors, the prototype-preorders and of course: You.
Even though it took much longer than planned, I only had a couple of refund requests - and most of them were people who had some misfortune in life and urgently needed the money.

Yes, it's a lot of work, well done, and well communicated. Good work. There're some things still to do and we know you'll do it so well as you've shown us, so no wonder people can wait for it.
Well I haven't really joined until later, so I'm not waiting for so long.

And when you start shipping it won't be over. You can still market it, or tour to show it, maybe consider new products (not only a CPU upgrade, but maybe a RYF version,
or Electronics Watch certification or other endorsement that may imply some changes in software or sourcing for next batches or maybe just learning and explaining...).
There's one page in the EOMA wiki with some ideas you might be interested in for the future, and it does not look as the Pyra is very far to comply with...

In any case, there's hope that this will be an example of a project working with their customers to help them get what they want instead of consumism sellers
trying to get away with giving as little as they can and tricking people to pay the as much and as often as they can. Showing there's another way to work
is important on its own.



We're in this til the end ED :)


Are you going to announce on crowd supply soon? once the prototypes are ready?

Why on crowdsupply ? I thought ED had already managed the financing.
I mean crowd supply is fine with me, and they offer cool products (did you see the TALOS workstation
with both computing power and blobless?) specially for the USA, but ED has already a shop and seems to manage things, doesn't him ?
Why would we need crowdsupply for the Pyra ?
 
Last edited:
We still need LINK for the United States market. Or somebody. Americans just buy more stuff, and Pyra is stuff....burp.
 
You know, I think the level of transparency and documentation seen on this project is easily as valuable as the end product.
It's probably tempting fate to count one's figurative chickens, but even the progress to date has been an outstanding feat of project management.

Very well done to all involved!
 
I don't know anything about electronics but I've heard that memory is one of the most difficult components because
it uses wide, very fast buses, and any slight discrepancy in track length or width may cause problems (maybe to one
chip and not another). Also, at least for PCs I've heard getting RAM timings right is also very tricky (but it may be
2x or 3x the Hz than the Pyra). And you can't really be sure they work until you get the timing right in u-boot (or the kernel?).
The other risk is if the 4GB chips draw more power than 2GB chips, and that somehow requires some change, but I'm not even
sure, since it's 2 chips for 4GB and 4 chips for 2GB. They might even draw less power ? No, I guess it's still double the bits
to refresh periodically but what do I know?

Well, it was a misunderstanding by me that the 4GB chips would only be 2. They will also be four, and they are pin compatible.

RAM is not so difficult, actually. Yes, you need to make sure your bus is compatible with the RAM, but that can be checked upfront.
So we know the RAM should be compatible. Power supply is not a problem here as well.

The Pandora actually had three or four different RAM / NAND-Chips during it's lifetime, and every chip needed an update of U-Boot (to include the timings and set it up properly).
It never took too long.

With the 4GB chips, it's a bit more complicated as we need an LPAE kernel. But that's not impossible.

So I wonder how long this will take. Having it working with 2GB is a hint PCB changes may not be needed but I'm not
sure it can be completely ruled out. In fact having the board designed by Nikolaus is an even surer hint no changes
will be needed. So I'm not really worried, just not so confident it'll be sure enough in a short time. Let's hope so.

Yes, it's 100% sure no PCB changes are needed.

I somehow thought there was one prototype already built with 4GB from this june posts.

Yes, one CPU board had been reworked.... but only with 2 of the new chips.
So it's a frankenstein board, with two old and two new chips. AND it's reworked, so we don't even know the hardware is 100% intact.
So not the best board to try to get the software to work.

It's better to wait for real produced prototypes.

That's what's more lovely in this project, hard work, no rushing and transparency.
I trust you won't order the 2000 prototypes until you're sure they'll work with the 4GB RAM.

At least not the RAM chips, as that would be deadly ;)
As said: there's nothing in the board design that needs to be changed that make the RAM chips compatible.
All traces are within specs, the RAM chips are pin compatible.

Maybe it's not worth it rushing for Christmas. If it's for Christmas fine, if it's for New Year's Eve fine, if it's for FOSDEM, fine...
This is great hardware, let's not risk a flaw for rushing.

Of course there's on need to rush for christmas. But it would've been nice!

I didn't know it was needed. Sounds even better with it, but I thought it was said the Pyra was already cooler than other similar size devices ?
Will it be for the battery or the CPU board ?

Well, cooler than even faster devices: Yes. But that doesn't mean it can't heat up.
Also, the OMAP5 is an older SoC. So even though it might not heat up as much as the Intel, it heats up more than it compared to the processing power.

So having something to keep the SoC a bit cooler for a while is a good idea, and it will only cost a couple of cents.

I think I'm going to be silly as hell since I know nothing of electricity, electronics, electromagnetism or radio.
But if the resistor value depends on the distance to access point or environment where the Pyra is put, would it make sense to use a potentiometer instead of a resistor
and let software control it so that people can maybe experiment for best setting in different places ?

Not quite. An antenna needs to be setup properly in order to receive the correct wavelength (i.e. 2,4GHz or 5GHz).
An antenna is basically just a wire from the Wifi chip to the battery, but that "wire" is laid out in a way that it receives the wavelength as good as possible.
However, between the antenna and the wifi chip, there's a trace on the PCB. That trace changes the resistance of the antenna, so the reception is not as good as it could be.
To compensate that, you add additional resistors between the antenna and the Wifi chip.

What we need to do is find out the optimal resistor value for the compensation. This can be done by measuring the reception quality of a Wifi signal under the exact same conditions each time while changing the resistor value.

Simple example (resistor values are just made up here):
With a 500 Ohm resistor, reception is -20dB.
With a 600 Ohm resistor, reception is -30dB (so worse).
With a 450 Ohm resistor, reception is -25dB (so still worse than with 500 Ohm).
With a 550 Ohm resistor, reception is -15dB (so better).
With a 525 Ohm resistor, reception is -18dB (so worse than the 550 Ohm one)
With a 535 Ohm resistor, reception is -12dB (best value so far)
With a 540 Ohm resistor, reception is -14dB (getting worse again)
With a 538 Ohm resistor, reception is -11dB (best value).

So you change the resistor and narrow down until you find the best reception.
Then you know the resistor for best possible reception. That value doesn't change, it depends on the board layout.

This is a very simplified example, but I think it should be easy to understand.

Of course, reception also gets worse when you move away from the access point, but that is not being compensated with resistor values but with increasing / decreasing the Wifi power (which is what the Wifi chip and the router does automatically).

Sorry for my ignorance again. Is this something sure to pass or might it fail it and require more changes ?
Maybe it's just something you have already tested in house but need a third party testing for certification, so it's just kind of paperwork.
Or maybe it's something that may require some shielding or any changes if the finished prototype does not pass the test ?

There are rules you have to follow during development to improve the chance it will pass on first try.
Of course, there can ALWAYS be something that radiates too much, and that needs to be fixed.
But if you are carefuly during development, you can really minimize the risk.

Why on crowdsupply ? I thought ED had already managed the financing.
I mean crowd supply is fine with me, and they offer cool products (did you see the TALOS workstation
with both computing power and blobless?) specially for the USA, but ED has already a shop and seems to manage things, doesn't him ?
Why would we need crowdsupply for the Pyra ?

The first 1000 units are funded, yes.
Once all have been shipped, we will have around 50k EUR left on our bank account - which is not enough to start the next production run on our own.

To produce 1000 more units, at least 380k EUR is needed. So yes, we either need a business angel who would like to provide us with that payment so we can produce and sell the next batch, or we need another round of preorders.
It should be easier once we delivered the first 1000 units, as then the device will already exist, but some popularity by using some crowdfunding platform should be helpful as well :)
 
Thanks a lot for the update as always. Nice to see everything seem to finally be coming together.
Also I fully agree with the others regarding the level of transparency of the project: Simply awesome !

While it would be nice to have the Pyra for Christmas, I think it's more important to make sure everything works. I am fine with my Pyra arriving in 2017.

It will be quite exiting to have the prototype go out the developers.
 
Well, that's it for todays post. I hope you enjoyed it, I hope I haven't forgotten anything - and of course: if you have any questions, feel free to ask :)

I hate to ask that, but are you thinking that 2016 is a likely mass production year?

I think there's no harm in being transparent if you already feel it's going to be very tight (which is the impression your post gives me)
 
Thanks for the update ED, as many have said it's a genuine pleasure to have so much behind the scenes info.

Well, it was a misunderstanding by me that the 4GB chips would only be 2. They will also be four, and they are pin compatible.
...

Good to hear that, if the 4GB version works as expected do you still plan to mass produce the 2GB version?

If you do then it would be much appreciated if you could run some benchmarks showing actual impacts on battery life & performance (vs using Nand swap file) with the 2GB & 4GB units.
 
I hate to ask that, but are you thinking that 2016 is a likely mass production year?

I think there's no harm in being transparent if you already feel it's going to be very tight (which is the impression your post gives me)

I think the post made very clear that it's unlikely, but we'll try ;)

However, that also depends on what you define as "mass production year".
It's very probably we START with the mass production already (for example, the CPU boards are candidates to be mass produced in December, maybe even the mainboards), but it might be we're missing LCD Cables or the keymats so we can't assemble them in December.

Mass production doesn't happen in one day, but if nothing bad does happen, it's likely to start in December.

But it's too early to tell - it depends whether the prototypes and remaining samples will work out of the box or not.
 
To summ it up: no Pyra for the masses until 2017. ;)
Well, better late than never. I hopethings speed up then when the pyra becomes more popular so more units can be made.

Honestly, when work had started on the Pyra, I never, ever thought it would take so long and we'd run into so many issues again! Seriously!
If you ever do a "Pyra 2" from scratch, what would you change to speed the entire dev process up? Or is there no way to do this for small Companies?
 
Why would we need crowdsupply for the Pyra ?
CrowdSupply is a location that attracts lots of eyes balls from highly technical people whom have lots of money to spare ;)

It's also a source of legitimacy for high quality products and a place to venture the Pyra into people whom might have different use cases... professionals or security researchers.

It also works as a store more than a crowdfund... for the most part

bonus: as far as I'm aware - CrowdSupply has never failed to reach a target ;)
 
If you ever do a "Pyra 2" from scratch, what would you change to speed the entire dev process up? Or is there no way to do this for small Companies?

Easy: Have billions of money available so multiple people can work on multiple things at the same time and you can produce three different case molds for testing even though you would throw most of them away.

Being fast is easy, if you have enough money to throw.

FormAction, for example, could only work on one case part mold at once, as they don't have so many machines.
If one case part needs 2 weeks and you have 10 molds, that's 20 weeks just for the basic molds.

If you have 10 machines, it is only 2 weeks.
 
I'm slightly confused having thought about the heat dissipation material a bit more. As I understand it, it needs to connect the SoC and the SD card cages. Those two components face each other, so you can't stick a single sticky sheet over both, but even if you could that would make the CPU cards less removable. You say the sheets you're looking at are ~0.3mm slimmer than the gap you're putting them in, which would leave a 0.3mm gap between the sheet and the SoC (assuming you're attaching it to the mainboard). Are you thinking of filling that gap with thermal grease, perhaps? As far as my understanding of thermal dynamics goes, conducted heat cannot pass even such a small gap without some medium to pass through. The SoC will set up convection currents, but those will heat up a lot more than just the conductive strip potentially, and won't evacuate enough heat when the CPU is running flat out to maintain a low stable temperature, and if you wanted to SoC package to radiate a lot of heat you should at least consider painting it white. A blob of thermal grease sounds the simplest solution to me, and a new tube of thermal grease for those of us who haven't ended up with a half-used one after messing about with x86 PCs isn't a big expense for hardware hackers.
 
omg so close! :confused: while it'd be cool to mess around with the pyra on christmas (especially during the family's "downtime" after eating massive amounts of food) and trying to visualize a countdown for new year's eve with the logo on the lid, i'd rather have a fully developer approved and tested device than one assembled in a hurry with a no-ears lcd cable :D also, "new year, new hardware, new games" makes for a nice motto/new year's resolution :p
 
  • Like
Reactions: rSl
@levi I know the mainbord PCB has some bare copper under the SoC area to dump the heat to. I assume thermal paste is just a mess waiting to happen so I think something like this may be a viable solution.

http://solutions.3m.com/wps/portal/...nterface-Pad-5519?N=8704982+3294001841&rt=rud

Yep, those pads will be put on the SoC.
[doublepost=1476886795,1476886583][/doublepost]
I'm slightly confused having thought about the heat dissipation material a bit more. As I understand it, it needs to connect the SoC and the SD card cages. Those two components face each other, so you can't stick a single sticky sheet over both, but even if you could that would make the CPU cards less removable. You say the sheets you're looking at are ~0.3mm slimmer than the gap you're putting them in, which would leave a 0.3mm gap between the sheet and the SoC (assuming you're attaching it to the mainboard). Are you thinking of filling that gap with thermal grease, perhaps? As far as my understanding of thermal dynamics goes, conducted heat cannot pass even such a small gap without some medium to pass through. The SoC will set up convection currents, but those will heat up a lot more than just the conductive strip potentially, and won't evacuate enough heat when the CPU is running flat out to maintain a low stable temperature, and if you wanted to SoC package to radiate a lot of heat you should at least consider painting it white. A blob of thermal grease sounds the simplest solution to me, and a new tube of thermal grease for those of us who haven't ended up with a half-used one after messing about with x86 PCs isn't a big expense for hardware hackers.

We've got heat dissipation on TWO sides.
The SoC is connected with the mainboard with a thermal pad as TrashyMG has shown.

The backside of the CPU board (between the CPU board and the battery) will have some heat dissipation material put on. It's soft, so it is squeezed a bit inbetween the case and the CPU board.
It will shield the heat from the battery and also dissipate it a bit, so the heat won't be too spotted in one area.
The material also covers a bit of the SD Card slots, so this is covered as well.
 
Don't they say:
  • Speed
  • Cost
  • Quality
Pick any 2 of the above, i.e.:
  • Want it fast and cheap? Sure, it'll be pretty crap though.
  • Want it fast and good quality? Sure, it is going to have a serious cost though.
  • Want it cheap and good quality? Sure, it is going to take a long time though.
Sums it up: https://goo.gl/ARhLfx
 
Back
Top