Hi, after someone sent me the links to the data sheets, I must say I am impressed by the Atom Z3770. It indeed looks as if it is really comparable - concerning interfaces.
One issue is that it is a 1300+ BGA in 0.4mm pitch while the OMAP5432 is 0.5mm (the 5430 would also be 0.4). So the system design of the 5432 is a little relaxed (being 20% more spacy).
With the Snapdragon discussions we now (again) have indeed three options to consider.
Maybe the best one to strive for would be replaceable CPU boards for all of them :)
This of course needs: 1. well defined interface connectors (positions, signals) 2. makes planning more difficult (we need to know in advance which versions in which quantities) 3. reduced number of units per variant - increasing the total price a little 4. we have to maintain 2 or 3 kernels + drivers
One solution could be that we only do step 1, publish some interface specifications and leave it to others to really design, build and maintain the alternate SoC boards.
So my proposal would be to continue with OMAP5 (because we know it best) and leave the Snapdragon and Atom designs to the community (but help them to succeed if possible).
BR, Nikolaus
PS: after studying the data sheets a little - the 1V Core power line may draw up to 6 Amperes (!) This is a special challenge for the power controller design
Am 07.02.2014 um 07:56 schrieb Dr. H. Nikolaus Schaller:
Hi, after someone sent me the links to the data sheets, I must say I am impressed by the Atom Z3770. It indeed looks as if it is really comparable - concerning interfaces.
One issue is that it is a 1300+ BGA in 0.4mm pitch while the OMAP5432 is 0.5mm (the 5430 would also be 0.4). So the system design of the 5432 is a little relaxed (being 20% more spacy).
With the Snapdragon discussions we now (again) have indeed three options to consider.
Maybe the best one to strive for would be replaceable CPU boards for all of them :)
This of course needs:
- well defined interface connectors (positions, signals)
- makes planning more difficult (we need to know in advance which versions in which quantities)
- reduced number of units per variant - increasing the total price a little
- we have to maintain 2 or 3 kernels + drivers
One solution could be that we only do step 1, publish some interface specifications and leave it to others to really design, build and maintain the alternate SoC boards.
So my proposal would be to continue with OMAP5 (because we know it best) and leave the Snapdragon and Atom designs to the community (but help them to succeed if possible).
BR, Nikolaus
PS: after studying the data sheets a little - the 1V Core power line may draw up to 6 Amperes (!) This is a special challenge for the power controller design _______________________________________________ Kernel mailing list Kernel@pyra-handheld.com http://pyra-handheld.com/cgi-bin/mailman/listinfo/kernel
To help discussions/documentation I have sketched a Score Sheet:
If you think this is useful to document and communitcate a decision (well, there is always an arbitrary and discussable component inside), we could fill in the missing pieces and ED could define the Importance (because one person must do that).
BR, Nikolaus
Hi,
okay, I thought a lot about Intel again, and I still prefer ARM.
Here is my main reason:
In power consumption, the Intel might be similar to the ARM SoCs now.
However, not in optimization of software (mostly games).
x86 games typically come from a desktop computer - which never really had issues with CPU power. Game developers never cared much about optimization. They simply suggested to upgrade your PC - and as these were running off AC Power, that wasn't really a problem.
ARM had always been a lot slower than desktop PCs. But game developers often tried to get the same or similar games out of them (at least for Android games). In order to do that, they had to optimize a lot.
We've got so many ARM-optimized emulators. I'm sure our current PCSXReARMed with NEON needs less CPU power than the normal PCSX on x86.
So in the end, running PCSX on an x86 SoC would lower the battery life, simply by using more CPU than the ARM pendant.
Intel claims 95% of all Android stuff runs on their system as well - but that's either everything except games that use ARM compiled code (which would probably be interesting games) or they do some ARM emulation - but that would also need more CPU power once again.
It doesn't help if perf/Watt is the same if most games / emulators are coded so unoptimized that they need more performance and therefore more power.
I'm no developer, and not that much into the materia, but that's my personal experience.
What do you thiink about that?
Am 07.02.2014 um 16:30 schrieb Michael Mrozek:
Hi,
okay, I thought a lot about Intel again, and I still prefer ARM.
Here is my main reason:
In power consumption, the Intel might be similar to the ARM SoCs now.
However, not in optimization of software (mostly games).
x86 games typically come from a desktop computer - which never really had issues with CPU power. Game developers never cared much about optimization. They simply suggested to upgrade your PC - and as these were running off AC Power, that wasn't really a problem.
ARM had always been a lot slower than desktop PCs. But game developers often tried to get the same or similar games out of them (at least for Android games). In order to do that, they had to optimize a lot.
We've got so many ARM-optimized emulators. I'm sure our current PCSXReARMed with NEON needs less CPU power than the normal PCSX on x86.
So in the end, running PCSX on an x86 SoC would lower the battery life, simply by using more CPU than the ARM pendant.
Intel claims 95% of all Android stuff runs on their system as well - but that's either everything except games that use ARM compiled code (which would probably be interesting games) or they do some ARM emulation - but that would also need more CPU power once again.
It doesn't help if perf/Watt is the same if most games / emulators are coded so unoptimized that they need more performance and therefore more power.
I'm no developer, and not that much into the materia, but that's my personal experience.
What do you thiink about that?
Yes, that is an important argument.
And I am not really sure about the power hungryness of the Atom. This max 6A@1V means 6W for the core alone. We have to add non-core and display. Of course this translates to 2A power drain from the battery @3.7V.
We have to get that heat out of the device.
BR, Nikolaus
Am Fri, 7 Feb 2014 07:56:13 +0100 hat "Dr. H. Nikolaus Schaller" hns@goldelico.com geschrieben:
Certainly not an easy task... if we could use CPU modules, we could easily update future revisions, so that would be good.
An interesting article about that: http://www.androidauthority.com/intel-vs-arm-future-mobile-technology-338340...
I've not yet seen a 3rd party Intel benchmark, and I don't trust Intel benchmarks (it's known they faked it at some time), so we don't know if performance is better.
Of course, being able to run all the Humble Bundle Linux games directly on the native Pyra system would be nice as well...
Not easy, definitely not. Nikolaus: While I wait for the QUalcomm informations, could you do a quick check on Intel, whether it's feasible and wouldn't cause us issues in the hardware design?
Hi, after someone sent me the links to the data sheets, I must say I am impressed by the Atom Z3770. It indeed looks as if it is really comparable - concerning interfaces.
One issue is that it is a 1300+ BGA in 0.4mm pitch while the OMAP5432 is 0.5mm (the 5430 would also be 0.4). So the system design of the 5432 is a little relaxed (being 20% more spacy).
With the Snapdragon discussions we now (again) have indeed three options to consider.
Maybe the best one to strive for would be replaceable CPU boards for all of them :)
This of course needs:
- well defined interface connectors (positions, signals)
- makes planning more difficult (we need to know in advance which
versions in which quantities) 3. reduced number of units per variant
- increasing the total price a little 4. we have to maintain 2 or 3
kernels + drivers
One solution could be that we only do step 1, publish some interface specifications and leave it to others to really design, build and maintain the alternate SoC boards.
So my proposal would be to continue with OMAP5 (because we know it best) and leave the Snapdragon and Atom designs to the community (but help them to succeed if possible).
BR, Nikolaus
PS: after studying the data sheets a little - the 1V Core power line may draw up to 6 Amperes (!) This is a special challenge for the power controller design _______________________________________________ Kernel mailing list Kernel@pyra-handheld.com http://pyra-handheld.com/cgi-bin/mailman/listinfo/kernel
Am 07.02.2014 um 11:34 schrieb Michael Mrozek:
Am Fri, 7 Feb 2014 07:56:13 +0100 hat "Dr. H. Nikolaus Schaller" hns@goldelico.com geschrieben:
Certainly not an easy task... if we could use CPU modules, we could easily update future revisions, so that would be good.
An interesting article about that: http://www.androidauthority.com/intel-vs-arm-future-mobile-technology-338340...
I've not yet seen a 3rd party Intel benchmark, and I don't trust Intel benchmarks (it's known they faked it at some time), so we don't know if performance is better.
Of course, being able to run all the Humble Bundle Linux games directly on the native Pyra system would be nice as well...
Not easy, definitely not. Nikolaus: While I wait for the QUalcomm informations, could you do a quick check on Intel, whether it's feasible and wouldn't cause us issues in the hardware design?
Well, from a first glimpse it looks ok. We can hook it up to RAM, USB3, 2 USB2 hosts and I2C, GPIO and MIPI-Display.
One minor limitation may be that it has only 3 SD ports while the OMAP5432 has 4.
We get either a problem with the second SD slot or WiFi. But we can connect either one through USB (because the Atom has one more USB port).
But this is IMHO not a blocking point.
The SoC has the same size as the OMAP. Only the BGA pattern is more dense and has more balls to connect so that we might need more layers of PCB. But I did not look into the pattern - if half of them is GND it would be ok.
Hi, after someone sent me the links to the data sheets, I must say I am impressed by the Atom Z3770. It indeed looks as if it is really comparable - concerning interfaces.
One issue is that it is a 1300+ BGA in 0.4mm pitch while the OMAP5432 is 0.5mm (the 5430 would also be 0.4). So the system design of the 5432 is a little relaxed (being 20% more spacy).
With the Snapdragon discussions we now (again) have indeed three options to consider.
Maybe the best one to strive for would be replaceable CPU boards for all of them :)
This of course needs:
- well defined interface connectors (positions, signals)
- makes planning more difficult (we need to know in advance which
versions in which quantities) 3. reduced number of units per variant
- increasing the total price a little 4. we have to maintain 2 or 3
kernels + drivers
One solution could be that we only do step 1, publish some interface specifications and leave it to others to really design, build and maintain the alternate SoC boards.
So my proposal would be to continue with OMAP5 (because we know it best) and leave the Snapdragon and Atom designs to the community (but help them to succeed if possible).
BR, Nikolaus
PS: after studying the data sheets a little - the 1V Core power line may draw up to 6 Amperes (!) This is a special challenge for the power controller design _______________________________________________ Kernel mailing list Kernel@pyra-handheld.com http://pyra-handheld.com/cgi-bin/mailman/listinfo/kernel
-- Mit freundlichen Grüßen,
Michael Mrozek
OpenPandora GmbH Geschäftsführer: Michael Mrozek
Schäffbräustr. 11 85049 Ingolstadt Deutschland Tel.: 0841 / 13 177 33 http://www.openpandora.de/ HRB 4879, Amtsgericht Ingolstadt
ICQ: 15691903 MSN: EvilDragon1717@hotmail.com eMail: info@openpandora.de _______________________________________________ Kernel mailing list Kernel@pyra-handheld.com http://pyra-handheld.com/cgi-bin/mailman/listinfo/kernel