wraggster
April 7th, 2010, 00:08
News via http://www.open-pandora.org/index.php?option=com_content&view=category&layout=blog&id=2&Itemid=2&lang=en
Finally - a new blog post. Did take a real long time, since there was so much we needed to figure out first.
Okay, here is what happened:
1. The WiFi issue - finally resolved
A lot of guys have probably already followed this on the boards: notaz was working on the WiFi driver and still couldn't get it to work reliable. As he didn't know what else to try, he thought it could also be a hardware issue - of course, that got us alarmed. Michael immediately stopped the board production (700 had been produced so far back then) to first do more tests with WiFi. We were totally in the dark. If you ever wanted to know how sudden death feels... well, we do know for sure yet.
Michael did begin investigating the hardware while notaz did more tests and tweaks with the software. A kernel which disabled everything except WiFi did get a bit more speed out of the WiFi, but it still was very slow. Was it the driver causing problems when other stuff was running? Or was it the hardware? We still didn't know for sure.
As we didn't know, we decided we will get some WiFi sticks for the beginning and ask every customer whether he wanted to get the Pandora with a stick and internal WiFi that might be fixable by software or maybe needs a hardware fix later.
Then, at one time, we found out that WiFi was faster on Rev2/3 boards then on the recent Rev4/5 boards - pretty weird, as nothing had been changed in these revisions that could affect WiFi. Except the board manufacturer. Using that clue, we figured out that the impedance of one trace had been changed. The impedance is important here, as a wrong impedance will distort the signal and then the module doesn't get a clean clock. Measuring the clock signal did prove that: Yep, the impedance had been changed.
Who's fault was it? Hard to tell. The company had changed that trace without asking, but we also didn't explicitely say it needs to have a correct back. So it's a hard thing, the trace is working, it was just not good enough for the signal.
Knowing that, Michael did work on it for a couple of days, doing some trial and error (not much you can't do unless you do know the exact new impedance).
He got it working better - but still not good enough to really be usable - was this due to software? Or still hardware?
So he started to work on a new testing board (Rev6) which could also be used as new board for the rest of the 3300 Pandoras if it fixes WiFi.
While doing that, he had another idea - modded another Rev5 board and got a HUGE speed increase! On minimal kernel, he can get 350KB/s, on normal image he gets 66KB/s stable speed (tested downloading from the internet. That fix is easy: Simply piggyback a resistor on top of an existing one. Takes about 1 - 2 minutes. And the rest of the 3300 boards can be populated with the new resistor immediately, so that fix won'tl be needed on them.
Speed can probably still be improved - Michael just found out about that yesterday, he will try to tweak more (or even combine the two mods) during the next days, until he leaves for UK to start assembling (he will do the modification of each board while assembling).
Finally! That news did make us more alive again, we were really suffering during the last days...
2. The cases
Since board production had been stopped for a while, we allowed the Chinese guys to tweak the case moulds a bit more. As you might have seen on the latest image, the bottom part (with the keyboard) looks spot-on right now - however, the top part still has some weird discoloration. No problem with stability here, just looks a bit weird if you look closely.
3. The OS - ready for first release
The latest image had been tested a lot by all of us already. It is pretty stable, has some minor issues (like you need to restart after calibrating the touchscreen at the moment... no biggie, I only needed to calibrate once in the last four months).
It has a working browser, eMail client, IRC, ICQ, MSN, etc. everything you need for normal life.
There is room for speed improvement in the future, SDL is not using hardware surfaces yet, but that will follow.
Also, boot process will be further optimized and will use Upstart to be a lot faster.
If you don't like a fancy desktop system, you can also use some minimal GUIs. We also designed that system to be configurable, so as soon as anyone codes a new GUI, it can easily be included by simply changing a config file.
The libpnd system is also working nicely: download a PND onto your SD Card and it will automatically appear in the menu (or on the desktop). Games will be sorted into categories, every dev can include a documentation in his PND so you can easily check the manual or installation settings, etc. Really convenient.
Of course, the categories, icons, etc. can be overriden, in case you want to have your own ones.
One thing you should not expect from the beginning is (as said before) accellerated movie playing (and therefore, also youtube movie playing).
This does need a new kernel and is on our TO-Do list for the next firmware release, which will be worked on as soon as the first one is ready to go on the NAND.
You can also help us testing then, as you can install the new beta image on an SD Card and try it out.
As the image is working well already, devs already concentrate to port more things. At the moment, we got fullspeed Amiga 500 emulation (without frameskip) at 500MHz, fullspeed Genesis and MegaCD, fullspeed C64 (Vice!) on 500 MHz without any frameskip, VCS2600, Colecovision, Atari ST, ... tons of stuff to play. Skeezix is also porting PocketSNES, so fullspeed SNES will also work when you get your unit.
PandoraPanic (minigames created by members of the GP32x.com community) is working nice, FreeDink, some PyGame-Games, etc. have also been ported already.
Overall, you will get a good working first release OS which still leaves a lot of room for improvements.
4. The current situation
We lost two weeks due to that WiFi issue, but now that this has been resolved being a fixable hardware issue, Michael told the company yesterday to ship out all the produced boards we have as soon as possible to UK. Additionally, Fatih told the Chinese to do the same: Produce the cases as they are now and ship them!
DJWillis is working on making the image NAND-installable, so yes, you might see us assembling them on webcam soon!
Finally - a new blog post. Did take a real long time, since there was so much we needed to figure out first.
Okay, here is what happened:
1. The WiFi issue - finally resolved
A lot of guys have probably already followed this on the boards: notaz was working on the WiFi driver and still couldn't get it to work reliable. As he didn't know what else to try, he thought it could also be a hardware issue - of course, that got us alarmed. Michael immediately stopped the board production (700 had been produced so far back then) to first do more tests with WiFi. We were totally in the dark. If you ever wanted to know how sudden death feels... well, we do know for sure yet.
Michael did begin investigating the hardware while notaz did more tests and tweaks with the software. A kernel which disabled everything except WiFi did get a bit more speed out of the WiFi, but it still was very slow. Was it the driver causing problems when other stuff was running? Or was it the hardware? We still didn't know for sure.
As we didn't know, we decided we will get some WiFi sticks for the beginning and ask every customer whether he wanted to get the Pandora with a stick and internal WiFi that might be fixable by software or maybe needs a hardware fix later.
Then, at one time, we found out that WiFi was faster on Rev2/3 boards then on the recent Rev4/5 boards - pretty weird, as nothing had been changed in these revisions that could affect WiFi. Except the board manufacturer. Using that clue, we figured out that the impedance of one trace had been changed. The impedance is important here, as a wrong impedance will distort the signal and then the module doesn't get a clean clock. Measuring the clock signal did prove that: Yep, the impedance had been changed.
Who's fault was it? Hard to tell. The company had changed that trace without asking, but we also didn't explicitely say it needs to have a correct back. So it's a hard thing, the trace is working, it was just not good enough for the signal.
Knowing that, Michael did work on it for a couple of days, doing some trial and error (not much you can't do unless you do know the exact new impedance).
He got it working better - but still not good enough to really be usable - was this due to software? Or still hardware?
So he started to work on a new testing board (Rev6) which could also be used as new board for the rest of the 3300 Pandoras if it fixes WiFi.
While doing that, he had another idea - modded another Rev5 board and got a HUGE speed increase! On minimal kernel, he can get 350KB/s, on normal image he gets 66KB/s stable speed (tested downloading from the internet. That fix is easy: Simply piggyback a resistor on top of an existing one. Takes about 1 - 2 minutes. And the rest of the 3300 boards can be populated with the new resistor immediately, so that fix won'tl be needed on them.
Speed can probably still be improved - Michael just found out about that yesterday, he will try to tweak more (or even combine the two mods) during the next days, until he leaves for UK to start assembling (he will do the modification of each board while assembling).
Finally! That news did make us more alive again, we were really suffering during the last days...
2. The cases
Since board production had been stopped for a while, we allowed the Chinese guys to tweak the case moulds a bit more. As you might have seen on the latest image, the bottom part (with the keyboard) looks spot-on right now - however, the top part still has some weird discoloration. No problem with stability here, just looks a bit weird if you look closely.
3. The OS - ready for first release
The latest image had been tested a lot by all of us already. It is pretty stable, has some minor issues (like you need to restart after calibrating the touchscreen at the moment... no biggie, I only needed to calibrate once in the last four months).
It has a working browser, eMail client, IRC, ICQ, MSN, etc. everything you need for normal life.
There is room for speed improvement in the future, SDL is not using hardware surfaces yet, but that will follow.
Also, boot process will be further optimized and will use Upstart to be a lot faster.
If you don't like a fancy desktop system, you can also use some minimal GUIs. We also designed that system to be configurable, so as soon as anyone codes a new GUI, it can easily be included by simply changing a config file.
The libpnd system is also working nicely: download a PND onto your SD Card and it will automatically appear in the menu (or on the desktop). Games will be sorted into categories, every dev can include a documentation in his PND so you can easily check the manual or installation settings, etc. Really convenient.
Of course, the categories, icons, etc. can be overriden, in case you want to have your own ones.
One thing you should not expect from the beginning is (as said before) accellerated movie playing (and therefore, also youtube movie playing).
This does need a new kernel and is on our TO-Do list for the next firmware release, which will be worked on as soon as the first one is ready to go on the NAND.
You can also help us testing then, as you can install the new beta image on an SD Card and try it out.
As the image is working well already, devs already concentrate to port more things. At the moment, we got fullspeed Amiga 500 emulation (without frameskip) at 500MHz, fullspeed Genesis and MegaCD, fullspeed C64 (Vice!) on 500 MHz without any frameskip, VCS2600, Colecovision, Atari ST, ... tons of stuff to play. Skeezix is also porting PocketSNES, so fullspeed SNES will also work when you get your unit.
PandoraPanic (minigames created by members of the GP32x.com community) is working nice, FreeDink, some PyGame-Games, etc. have also been ported already.
Overall, you will get a good working first release OS which still leaves a lot of room for improvements.
4. The current situation
We lost two weeks due to that WiFi issue, but now that this has been resolved being a fixable hardware issue, Michael told the company yesterday to ship out all the produced boards we have as soon as possible to UK. Additionally, Fatih told the Chinese to do the same: Produce the cases as they are now and ship them!
DJWillis is working on making the image NAND-installable, so yes, you might see us assembling them on webcam soon!