Idea to maintain community after P2


You can install Windows on the current Pandora via QEMU or Wine, that's more than enough.


The OpenBrick2 will have more CPU power so this can be used more efficiently for Win98 retrogaming, maybe eben WinXP.
I was personally hoping to have Win8 for all my school software. US Schoolz require Mac OS X or Windows sine they are for some reason afraid of anything Linux (eslecially ARM)
FTFY

Why in the hell would anyone want Win8 and why are schools in the US still so bad?
 
Last edited by a moderator:
So I guess you're saying that if it were an SoC that had both A15s and A7s on it that'd be no good for you? Or another CPU which offers comparable perf/W as A7s but higher peak perf, which is what Intel claims Silvermont accomplishes?
Yes. The reason being the suspicion that it will drive the price up and make the P2 less distinguishable from mainstream offerings. Like only the integrated keyboard remains as a difference when compared to the shield. I don't think we need this discussion about a larger community in that case at all, because the product will be dead from the beginning.


The second reason is that I can't really think of an application I regularly use that is too slow for me. That is probably due to the fact that I'm really quite old and recent things (i.e newer than an Amiga) don't interest me that much. But I was recently compiling some gcc versions with oe's bitbake on the pandora and it wasn't as annoying as I thought. A dual core A7 would even give some nice improvement here.


Entry level SoCs might be easier to source.

A7 sucks. A 1GHz A7 might perform similarly to Pandora's 1GHz A8 in "general purpose" tasks but in optimized NEON code it's nowhere close. It'd really suck if that's what we ended up with.
Yes. I was hoping a slightly higher clock speed would make up for this combined with some offloading to the second core while still beeing slightly more low power. I may be totally wrong here and even if I'm not this is definitely only a very small improvement to the current state (which is ok for me but seemingly not for the majority of forum members). Do you have some numbers from running drastic on A7?

Either way, you probably won't get something like a Pandora elsewhere.
Yes. You are right as always :)
 
For the sake of discussion, can both x86 and ARM SoCs coexist mutually exclusive on the same board with a hardware switch to switch between the two.
 
For the sake of discussion, can both x86 and ARM SoCs coexist mutually exclusive on the same board with a hardware switch to switch between the two.
Why do so many people want to have a much bigger Pandora?

Two CPUs, two times RAM, etc...

What more? Two GPUs?

It would be a thick tablet this way.
 
Last edited by a moderator:
For the sake of discussion, can both x86 and ARM SoCs coexist mutually exclusive on the same board with a hardware switch to switch between the two.
Why do so many people want to have a much bigger Pandora?

Two CPUs, two times RAM, etc...

What more? Two GPUs?

It would be a thick tablet this way.
Why two times RAM? Why two GPUs?  Couldn't they use the same RAM and GPU?  Like I said, I don't know about this stuff and just want to ask a question so please no spanking :)
 
Because someone wanted two seperate RAMs yesterday and GPU would be next in the list.

This stuff costs a lot of SPACE. There is a reason, why CC could not handle our boards - they are fully loaded already.
 
Why two times RAM? Why two GPUs?  Couldn't they use the same RAM and GPU?  Like I said, I don't know about this stuff and just want to ask a question so please no spanking :)
The CPU, RAM, and GPU all exist as a single package. You can't have one SoC with all that and then another CPU just floating off on the board somewhere able to make use of the first one's stuff.
 
I recall some time later it did become possible to boot Windows on the XBox natively by either hardware modding, or flashing the firmware (? might remember the later one wrong...).
This would be the first I'd ever heard of that. As far as I know, the only way to run Windows on an XBox is by installing Linux and then running it in a virtual machine.
 
Yes. The reason being the suspicion that it will drive the price up and make the P2 less distinguishable from mainstream offerings. Like only the integrated keyboard remains as a difference when compared to the shield. I don't think we need this discussion about a larger community in that case at all, because the product will be dead from the beginning.
The difference between a cheap as dirt A7 based SoC and a high end one is $20, maybe $30. Is this really where you want to cut costs? Where else would you like to? I don't think you'll find lots of people on board for making the price < 10% cheaper for much worse performance (GPU performance would probably also be down a lot)

These SoCs are meant for bottom of the barrel $100 devices, where you go cheap on everything. That's not what a Pandora-like device is going to be. If you think that the only way it'll be relevant is if it beats Shield on price then you're right, it'll be dead. But a lot of people who still value Pandora even vs Shield will disagree with you - "only" the integrated keyboard (although there's a bunch of other stuff really) is enough. If only Pandora gives what you want then that's what you'll go for, it's more or less always been like this. Time to worry will be when another player does something that's genuinely like Pandora.

The second reason is that I can't really think of an application I regularly use that is too slow for me. That is probably due to the fact that I'm really quite old and recent things (i.e newer than an Amiga) don't interest me that much. But I was recently compiling some gcc versions with oe's bitbake on the pandora and it wasn't as annoying as I thought. A dual core A7 would even give some nice improvement here.
Okay, so out of curiosity why do you want a new Pandora at all?

Entry level SoCs might be easier to source.
Not necessarily - MediaTek's won't be anyway.. Rockchip and Allwinner might be though.

Yes. I was hoping a slightly higher clock speed would make up for this combined with some offloading to the second core while still beeing slightly more low power. I may be totally wrong here and even if I'm not this is definitely only a very small improvement to the current state (which is ok for me but seemingly not for the majority of forum members). Do you have some numbers from running drastic on A7?
I don't have numbers myself, just going by somewhat vague reports from users, but yes the impression I get is that NEON isn't that strong (it's actually stronger on A8 than it is on A9)
 
Yes. I was hoping a slightly higher clock speed would make up for this combined with some offloading to the second core while still beeing slightly more low power. I may be totally wrong here and even if I'm not this is definitely only a very small improvement to the current state (which is ok for me but seemingly not for the majority of forum members). Do you have some numbers from running drastic on A7?
I don't have numbers myself, just going by somewhat vague reports from users, but yes the impression I get is that NEON isn't that strong (it's actually stronger on A8 than it is on A9)
You mean A7 instead of A9, right? I would expect the A9 to be a bit better than the A8 in all respects. If we can't get a big.LITTLE SoC, maybe we should use something like the RK3188  - but only if each of its cores are strictly better than what we have now.
 
You mean A7 instead of A9, right? I would expect the A9 to be a bit better than the A8 in all respects. If we can't get a big.LITTLE SoC, maybe we should use something like the RK3188  - but only if each of its cores are strictly better than what we have now.
No, I meant what I said. The NEON unit is stronger in A8 than A9. In A9 it's not as tightly coupled, and as a consequence it only has a 64-bit load/store path instead of a 128-bit, it can't dual-issue anything (on A8 it can dual-issue load/store/permute instructions with other instructions) and it probably doesn't have the same latency hiding for loads. The only advantage is you don't get nearly as big of a penalty transferring from NEON registers to ARM registers. Other than these things it's more or less the same, there's no out-of-order execution or register renaming or anything like that for NEON instructions on A9.

NEON is really difficult to optimize for on A8 (to get it working without a lot of stalls) but it's fairly capable if you manage it. That sort of translates to A9 but not entirely. But I'd still rather take an A9 because of how much faster per clock the generic scalar stuff tends to be and because they can usually clock a lot higher than A8s (and something like RK3188 is on a more advanced process and since it's such a tiny ~25mm^2 core it probably costs next to nothing). Even something like DraStic doesn't spend ALL of its time in NEON, recompiler output in particular benefits a lot from OoOE because no one schedules it for in-order processors.

I don't know what NEON is like on A7 but I wouldn't be surprised at all if it's substantially weaker - on A8 it couples two 64-bit ALUs (or at least that's how I see it - some ALU operations are only 64-bit wide but most are 128-bit) and it can do 4x16-bit or 8x8-bit multiplications. They could have cut these things in half on A7 for all I know. Would need someone to do some tests on it I guess, I don't have any A7 hardware and for some reason ARM stopped giving us any timing information in their TRMs. I did once read that NEON was basically like this on Cortex-A5. A7 is by the A5 team, you can see how they enhanced A5 to get to it - maybe they didn't change the NEON part that much. I do know that DraStic performs especially badly on Cortex-A5s.
 
Last edited by a moderator:
This thread has deviated quite a bit from the original OP topic, perhaps all this soc talk could be moved to the "possible soc update" thread.
 
Last edited by a moderator:
Thanks for the insightful answer, Exophase. I'm glad I asked, instead of just assuming it was a typo, which it was not :)

How big would the difference be, in terms of performance and perf/W, between a recent quad-core A9 like in RK3188 and a big.LITTLE 8-core A15/A7 like Exynos? Do you have an idea of how those would compare to the single-core A8 we have now?

And then the big question of course is: where does BayTrail fit in that comparison? I guess we'll have to wait for reliable data on that one...
 
The tricky thing about perf/W is that it's not a straight line but a curve. To really get any kind of idea you'd have to test power consumption at a bunch of different clock speeds. But reviewers rarely do this. The curve may not even be that smooth, if the OS or hardware power management has to bump up the voltage in unexpected ways. What we're seeing on some A15 implementations today is that the voltage is being kicked up a lot for the last few 100MHz you get out of it. So the power consumption sky rockets. And since the tests have focused on how much power it consumes with cores at full tilt it makes the power consumption look pretty bad.

Still, I'd expect that on the same process A9 will generally have better perf/W than A15 for much of the perf curve, and A7 will generally have better perf/W than both. This doesn't apply at all for perf regions one CPU can't even reach - that should go without saying, but this is something that has to be spelled out to people anyway..

So far BayTrail looks pretty impressive in both perf and perf/W. Even though it needs a higher clock speed to reach the same performance levels as A15 SoCs it still does so at much lower voltages, and the process + design lead further gives them lower baselines to start with. Intel has always shown very good voltage vs frequency scaling, probably due to the level of custom design they pursue and the level of control they have over their manufacturing.

Very rough data I've seen so far is about 1.5-2W for Cortex-A15 @ 1.7GHz on Exynos 5250. Less for Cortex-A15 @ 1.6GHz on Exynos 5410 (better process, lower clock). About 1W for a BayTrail core @ 2.4GHz - but I've also heard only 2.5W for 4 cores at 2.4GHz so the actual number may be quite lower (and the 1W would include some other stuff that doesn't scale with cores). The quad core number does seem too good to be true, but even 1W for one core full tilt is very good for this performance level. A9 cores on 40nm were already using > 1W while providing far less performance. Not sure about A9s on 28nm like RK3188, though...
 
I'm not that much interested in peak perf/W - the whole curve is important. A lot of stuff people will be running on their P2 will not be that demanding (e.g. all the stuff that already runs fullspeed on the current Pandora). Battery life is one of the areas where we can really make a difference.

Also, how good are the various cores at being idle with low power? Are there differences in power management possibilities?

How do the various GPUs (PowerVR vs Mali vs Intel) compare in terms of power consumption?

Sorry for asking all these questions, but it's just hard to make much sense out of all the marketing buzzwords and vague benchmark results you find when you Google such things.
 
The CPU, RAM, and GPU all exist as a single package. You can't have one SoC with all that and then another CPU just floating off on the board somewhere able to make use of the first one's stuff.
Meant to comment on this before...

The CPU and GPU being part of the same chip is very true, but the RAM isn't in the same package - you probably already know this but I figured I'd clarify just the same. Most phones (and Pandora) use Package on Package for their RAM, where a separate RAM module in a separate package is stacked on top of the SoC. But most of them also support non-PoP memory, and larger devices like tablets will often go this route. For something like Pandora non-PoP could be desirable because it's cheaper, but it's only an option if there's enough board space and the memory is available in non-PoP form (LPDDRx is harder to find in non-PoP)

But the idea of multiplexing the same RAM between multiple SoCs is still not attractive.

I'm not that much interested in peak perf/W - the whole curve is important. A lot of stuff people will be running on their P2 will not be that demanding (e.g. all the stuff that already runs fullspeed on the current Pandora). Battery life is one of the areas where we can really make a difference.
I strongly agree. Too bad this isn't what review sites have really been testing. But until recently they haven't really been testing power consumption under proper instrumentation at all, so maybe they'll move in this direction.

Oh, you do get companies presenting curves like this for their products. It's just hard to trust them when everyone says they're the best.

big.LITTLE makes the curve even weirder, it's got this big discontinuity where the A7 switches to the A15...

Also, how good are the various cores at being idle with low power? Are there differences in power management possibilities?
Very low idle power consumption is really, really important for phones (and Intel has been making it an emphasis even for their higher end products). Anything designed to be in phones will be made on a process that offers really low leakage transistors and will be carefully optimized for this. Right now it looks like everyone is getting good numbers with this with the possible exception of AMD, who isn't putting SoCs in phones anyway.

Anything current you pick will probably do a bit of better job than OMAP3530 or DM3730 - not to say those were doing a bad job.

How do the various GPUs (PowerVR vs Mali vs Intel) compare in terms of power consumption?
This is one of those thing that's even less well tested right now. The big problem here is that the tests will tend to run the GPU as fast as they can, unless it hits some cap like 60FPS or maybe 30FPS. Games should have an option to sacrifice frame rate, resolution, and/or image quality in the interest of preserving battery life. But they don't (as far as I know).

Bay Trail's GPU looks like it offers decent power consumption for middle-of-the-road performance. It's something like 1-1.5W full tilt, which is similar to what you get on Exynos 5410's SGX543MP3 which is a similar speed. But the measurements I'm basing this off of aren't really very great. I suspect that chips like S800, with the most powerful GPUs, actually also get the best perf/W too, because they achieve such high perf by being really wide. Being wide means they can get the same performance as their competitors while running at lower clock speeds, which is more power efficient.

Sorry for asking all these questions, but it's just hard to make much sense out of all the marketing buzzwords and vague benchmark results you find when you Google such things.
Yeah, who knows how many more years before the benchmarking situation with mobile stuff even approaches respectable :/
 
Thanks for the discussion! It now makes perfect sense that A7 neon is much weaker than I thought (I knew only about the missing dual issue) and therefore not a suitable successor.


My idea was that by going minimal on all components (SoC, RAM, screen, whatever) the P2 might a) not become more expensive to manufacture then the current Pandora and b ) get a marketable claim like 'Runs for 10+x hours at full speed' that I'm pretty sure would work (marketing is crazy) even if performance is capped to achieve that and other devices will get similar time with similar tasks.


But minimum should be better than the current Pandora and you made it clear there is no such SoC available (in a sense, because the current one isn't bad enough :)


Therefore I retract my motion that there could be a more low power P2.

Okay, so out of curiosity why do you want a new Pandora at all?
Mostly to repace the one im holding right now which fulfils most of my computing needs, but will most probably be significantly older when the P2 hits the market. But I'm also hoping for a (slightly) improved keyboard and a (slightly) higher screen resolution.
 
Last edited by a moderator:
Back
Top