ED P2 successor notes - and a suggestion.

[Nagelfar]

I don't think there is going to be another "Pandora" (whatever the name is going to be) with preorders. Too much pain down that road.

 

[Klumpen]

Ed mentioned already, that there will not be any preordering.

They will be sold on stock or not at all.

Guess whose idea the whole preorder stuff was...

 

[Mr_Loon]

...


Would it make sense to approach Odroid and see if you could use a slightly modified version of the Odroid XU as a 'motherboard' for P2?


...

Apologies folks, missed this thread, still think there is some worthy discussion to be had on this subject but will start a new topic so as not to pull this one any further off topic

Edit : or maybe not (see below) glad to see it did make sense, sad to see they said no.

 

[EvilDragon]

Would it make sense to approach Odroid and see if you could use a slightly modified version of the Odroid XU as a 'motherboard' for P2?

They had a module for the Exynos4, but won't have one for the Exynos5 (already asked), so, no.

If worst comes to worst, and ED cant get a SOC, (I'm sure he'll get one though), he could always try getting PC's sticks PCBs(which will fit) and working out a way to hook up a touch screen to them, perhaps one of the chinese manufacturers would make a custom one. Actually, now that I mention it, I know they do that, I have seen somewhere where they talked about customisation options. Might be the way to go IF ED cant get a SOC.

Would be WAY too huge. You'd have to make the device a lot thicker.


The maximum thickness a module (or stick) could have so we could fit it into the current size would be about 2mm, soldered directly to the PCB.

Basing P2 on a SoC that is outdated by at least one generation upon its release would make it a very hard sell as a premium priced device IMHO.

Yes and no.


The Pandora has multiple things that can improve:


One part of it is the overall hardware (processing power, WiFi, screen size and resolution, backlit keyboard, etc.), another part the case in terms of quality and usability (better keyboard, shoulder buttons, etc.)


Of course, improving EVERYTHING would give us a lot of press coverage and would be an awesome device, with probably quite a few sales.


This is what I'm aiming for.


However, even if we can improve everything but can only get a Quad Core A9, it would still be a huge improvement over the current Pandora and would certainly sell a lot better than the Pandora will in the future.


With that, you have stable and growing customerbase, and with product that LOOKS like a premium product thanks to an improved case, it would also be easier approaching SoC manufacturers in the future and get the next generation of SoCs (i.e. the A57).


So regardless if we could get a big.LITTLE SoC or only an A9 Quad Core, it would be a path forward opening more possibilities.

I don't think there is going to be another "Pandora" (whatever the name is going to be) with preorders. Too much pain down that road.

Yes and no. I don't want paid preorders, but preorders to find out how big the first batch has to be will certainly be helpful.


And we might use Kickstarter as well (but only AFTER our prototype run, so we can REALLY be sure the device is ready for production), which would basically be something like a paid preorder.


But it will NOT happen like with the Pandora, where paid preorders are being taken long before the device is finished, and false delivery dates are being announced.

 

[thatgui]

However, even if we can improve everything but can only get a Quad Core A9, it would still be a huge improvement over the current Pandora and would certainly sell a lot better than the Pandora will in the future.

With that, you have stable and growing customerbase, and with product that LOOKS like a premium product thanks to an improved case, it would also be easier approaching SoC manufacturers in the future and get the next generation of SoCs (i.e. the A57).


So regardless if we could get a big.LITTLE SoC or only an A9 Quad Core, it would be a path forward opening more possibilities.

Does the idea of having the SoC on a seperate module that can be exchanged spark any interest on your side ?

On a side note: mind having a look in the OP-Ticket system - its not much, but it would be appreciated

 

[EvilDragon]

However, even if we can improve everything but can only get a Quad Core A9, it would still be a huge improvement over the current Pandora and would certainly sell a lot better than the Pandora will in the future.

With that, you have stable and growing customerbase, and with product that LOOKS like a premium product thanks to an improved case, it would also be easier approaching SoC manufacturers in the future and get the next generation of SoCs (i.e. the A57).

So regardless if we could get a big.LITTLE SoC or only an A9 Quad Core, it would be a path forward opening more possibilities.

 Does the idea of having the SoC on a seperate module that can be exchanged spark any interest on your side ?

If we can get a module which is small enough (something like the OMAP4 module from Jorjin), then I'm happy using a module as well, sure.

But size and availability is the issue.

On a side note: mind having a look in the OP-Ticket system - its not much, but it would be appreciated

I usually check it once a week, but right now, GamesCom preparation is more important.

 

[rohezal]

ED please try to get a chip with open cl support, it would REALY help with parallel task like rendering an image (most work in the emulators in done here).

 

[_wb_]

I wouldn't mind if the SoC would be a recent one based on the "outdated" A9 instead of a cutting-edge big.LITTLE one. There are advantages in using something well-known, well-tested, and mature, even if it's a bit older. Cutting-edge SoCs may still have some problems (e.g. Samsung's Exynos 5410) and it is less likely to have Free and Open Source drivers for its GPU etc.

Also, I think it's important to distinguish between a recent A9-based SoC like the Rockchip RK3188 (28nm process, 1.8GHz quad core, Mali-400 GPU at 533MHz, available since 2013Q2) and an older one like TI's OMAP4430 (45nm process, 1.2GHz dual core, PowerVR GPU at 365MHz, available since 2011Q1). The OMAP4 wouldn't be a very big improvement compared to what we have now, especially because we would still be stuck with a relatively crappy proprietary GPU. Something like that Rockchip would be a much bigger improvement, not just because the GPU has FOSS drivers, but also because the lower node size means it will have better perf/Watt, and its ability to use DDR3 means better RAM speeds are possible.

 

[Mr_Loon]

Would it make sense to approach Odroid and see if you could use a slightly modified version of the Odroid XU as a 'motherboard' for P2?

They had a module for the Exynos4, but won't have one for the Exynos5 (already asked), so, no.

Basing P2 on a SoC that is outdated by at least one generation upon its release would make it a very hard sell as a premium priced device IMHO.

Yes and no.


The Pandora has multiple things that can improve:


One part of it is the overall hardware (processing power, WiFi, screen size and resolution, backlit keyboard, etc.), another part the case in terms of quality and usability (better keyboard, shoulder buttons, etc.)


Of course, improving EVERYTHING would give us a lot of press coverage and would be an awesome device, with probably quite a few sales.


This is what I'm aiming for.


However, even if we can improve everything but can only get a Quad Core A9, it would still be a huge improvement over the current Pandora and would certainly sell a lot better than the Pandora will in the future.


With that, you have stable and growing customerbase, and with product that LOOKS like a premium product thanks to an improved case, it would also be easier approaching SoC manufacturers in the future and get the next generation of SoCs (i.e. the A57).


So regardless if we could get a big.LITTLE SoC or only an A9 Quad Core, it would be a path forward opening more possibilities.

Just to be sure we're talking about the same thing, did you approach Odroid and enquire as to the possibility of buying XU boards which are identical to the current board except that in stead of being populated with USB / HDMI / Audio etc. connectors there are just headers on the board? Or did you ask if they planned to produce a mini module as per the Exynos 4?


The reason I ask this is that by my calculations at 9cm by 7cm the Ordoid XU board could just about be made to fit as a P2 motherboard especially if P2 had a 5" 4:3 screen and even more so if the PCB was included in the lid with all the advantages that brings  (what are the dimensoins of the current Pandora PCB?) It's fully appreciated that we are in no way comparing like with like here as the current PCB includes things like contacts for keys & game controls etc.

____________________________

My gut feel is still unchanged that it would be a very hard to make a P2 based on an A9 that was a financial success if it was sold at the current price of a 1Ghz unit i.e. approx $750 delivered. This would probably be 2 generations (A15 & A57) out of date upon release in 2014(?) and you would probably be able to buy a stick PC with an identical A9 processor for less than $50 delivered.

EDIT : NB it is fully understood that using an A9 is not the desired solution at that you're making every effort to find a suitable current gen SOC (A15) to use.

 

[Foxgod]

Maybe we should wait for Arm8 specs, before deciding if the P2 should have an Arm7.

 

[EvilDragon]

It's not possible putting an A15 behind the screen.

The biggest heat is being produced by the display and the SoC.

If you max the use of Pandoras CPU to 100% and wait a bit, you can even feel the warmth of the chip coming through the plastic.

The A15 produces more heat, a multicore will even be worse.

Put that behind a display where the backlight also produces heat and you're bound to have huge issues.

So yeah, I only asked about the module, because anything bigger won't help us.

 

[Mr_Loon]

It's not possible putting an A15 behind the screen.


The biggest heat is being produced by the display and the SoC.


If you max the use of Pandoras CPU to 100% and wait a bit, you can even feel the warmth of the chip coming through the plastic.


The A15 produces more heat, a multicore will even be worse.


Put that behind a display where the backlight also produces heat and you're bound to have huge issues.


So yeah, I only asked about the module, because anything bigger won't help us.

Interesting, seems like Samsung managed it with the Galaxy S4 but they might have had a little bigger R&D budget to find a  workable cooling solution.

 

[EvilDragon]

The S4 has OLED, which produces less heat, AFAIK.

Aditionally, they simply throttled the CPU when it's using too much power, so you can't use the full CPU power.

Also, I don't know if the S4 can run high performance applications without throttling the CPU even more over a longer time.

A smartphone doesn't really use a lot of the CPU for a longer period, it usually just peaks for a few seconds (i.e. while loading a website).

A Pandora with DOSBox emulation, for example, would use the CPU for extended periods of time with high load.

 

[ghostpatch]

Something like that Rockchip would be a much bigger improvement, not just because the GPU has FOSS drivers, but also because the lower node size means it will have better perf/Watt, and its ability to use DDR3 means better RAM speeds are possible.

you sure it has open source drivers, or is it only partial with big nasty blobs? that'd be so cool but somehow hard to belief

A Pandora with DOSBox emulation, for example, would use the CPU for extended periods of time with high load.

then the wait for a big.LITTLE with A12 cores instead of A15's might be worth it

 

[Mr_Loon]

The S4 has OLED, which produces less heat, AFAIK.


Aditionally, they simply throttled the CPU when it's using too much power, so you can't use the full CPU power.


Also, I don't know if the S4 can run high performance applications without throttling the CPU even more over a longer time.


A smartphone doesn't really use a lot of the CPU for a longer period, it usually just peaks for a few seconds (i.e. while loading a website).


A Pandora with DOSBox emulation, for example, would use the CPU for extended periods of time with high load.

There is plenty of stuff on Youtube showing the S4 used for emulation. Though I couldn't find anything about how much throttling is implemented / required to alleviate the necessity of active cooling.

From my perspective the question should not be does the A15 require throttling?, but, What sort of performance can you get from it even if throttled?

If performance is still great (when compared with A9) with throttling & passive cooling then it's still a sensible choice of SOC for P2 IMHO.

Don't think I'm saying that because Samsung have done it with a Galaxy S4 then OP GmbH should be able to do it with P2, it's fully appreciated that they have a lot more to spend on R&D / design etc. However if the R&D budget for P2 permits it then I'd suggest buying an S4 with the A15 5410 and giving it an extended stress test in terms of things like DosBox performance. This should at least give you some real idea, rather than mere speculation, if it is feasible to use the A15 5410 in a passively cooled device like P2.

 

[EvilDragon]

Well, what emulators did you see? If it's any emulator running fullspeed (or near fullspeed) on the Pandora, then the A15 wouldn't really need high CPU load. DOSBox would eat a lot, if you run Windows on it. Or Dreamcast, GameCube, etc. Emulation. Everything else probably doesn't really need A15 performance.

 

[_wb_]

Something like that Rockchip would be a much bigger improvement, not just because the GPU has FOSS drivers, but also because the lower node size means it will have better perf/Watt, and its ability to use DDR3 means better RAM speeds are possible.

you sure it has open source drivers, or is it only partial with big nasty blobs? that'd be so cool but somehow hard to belief

Well, there's Lima (http://limadriver.org/), which is currently still experimental, but it seems to be pretty good already judging from the demo I saw at FOSDEM. It is already slightly faster than the original binary blob, too.

Besides Mali, I think the only other ARM GPU that has a real FOSS driver is the Adreno (used in Qualcomm's Snapdragon SoCs).

This is an important consideration in my opinion; I wouldn't pick a SoC that uses a PowerVR again, because apparently it is very hard to RE and nobody seems to be even trying anymore at this point.

 

[Mr_Loon]

Well, what emulators did you see? If it's any emulator running fullspeed (or near fullspeed) on the Pandora, then the A15 wouldn't really need high CPU load. DOSBox would eat a lot, if you run Windows on it. Or Dreamcast, GameCube, etc. Emulation. Everything else probably doesn't really need A15 performance.

Good point, PPSSPP was probably the most demanding one. However IMHO the lack of any clear indication as to how the A15 5410 works under real stress for extended periods is a good reason to follow my suggestion of picking up an S4 or perhaps Odroid XU and giving them a good workout to check the feasibility of using the A15 5410 as a potential future SOC for P2. Not sure if the one running PPSSPP does in fact use the A15 5410 or the Snapdragon.

 

[_wb_]

If in some way ED manages to get a big.LITTLE SoC from Samsung, then please make sure it's the Exynos 5420 and not the 5410. It looks like the 5410 has severe issues that essentially mean you don't get any of the advantages of big.LITTLE, and supposedly these issues are fixed in the 5420. Also, the 5420 uses a Mali GPU instead of a PowerVR, which is better for performance and for FOSS driver support.

 

[WizardStan]

Personally I boubt the need of 8 cores. 4 cores (2x A7 and 2x A15) would be enough. You simply can't parallelise everything

You don't get all 8 cores at the same time, it switches. So a 2x2 would only be dual-core, automatically switching between efficiency and high performance as needed. Dual-core is a step up for the Pandora, and big.LITTLE light years ahead, but still not as big an improvement as we could get. If we could get a 2x2 where all four cores could be used at the same time that would be awesome, but as far as I know such a thing doesn't exist. There's recently been a 4x4 that can use all 8 at the same time, but ED's probably not going to get one of those any time soon