SoC: Back and forth!

[Exophase]

That A80 seems some really interesting SoC and might be worth waiting some weeks.

Is there a real Datasheet GTA04 can work with?

Don't think the 43 Pages are comparable to the 6000 TI offers.

Also I can't find any information regarding clockspeed.

Not for A80 and nothing reliable for A31.

There's nothing you can get publicly for A80. What I gave before is a datasheet for A31, that's not the same as a user manual which is the 6000 page thing you referred to. You can find user manuals for older Allwinner SoCs like A10, A13, and A20. They're not 6000 pages, but they probably shouldn't be because they're not that complex.

All Allwinner SoCs (not including A80) seem to list the same 2D engine support in their datasheets. The user manuals describe how to use them, including register I/Os. It's not very complex, writing a driver for it shouldn't be very hard, if there isn't already one.

See the "Mixer" section here for more information: http://dl.linux-sunxi.org/A20/A20%20User%20Manual%202013-03-22.pdf

I still don't know if A80 will have this same engine but I bet it will.

 

[Askarus]

ou can find user manuals for older Allwinner SoCs like A10, A13, and A20. They're not 6000 pages, but they probably shouldn't be because they're not that complex.

Is it an advantage or a disadvantage?

Is OMAP way more complex than it needed to be or is Allwinner missing lots of features that makes it more complex?

 

[rohezal]

The A80 is a 4 core big 4 core little cpu. Means more power, but I am not sure if games could use it. The ps2 emulator can use up to 3 cores (but not sure how much the third one).

 

[Granitehead]

The A80 is a 4 core big 4 core little cpu. Means more power, but I am not sure if games could use it. The ps2 emulator can use up to 3 cores (but not sure how much the third one).

I guess mostly it wouldn't be faster than the same with only 2 cores. But it could possibly use less power than OMAP5 or Snapdragon, I think.

 

[fusion_power]

"Allwinner"? I don't know, sounds more like a unknown, cheap chinese company fo me.  h34r:

 

[Exophase]

"Allwinner"? I don't know, sounds more like a unknown, cheap chinese company fo me.  h34r:

Cheap, Chinese, yes. Unknown, no, they're in a lot of hardware now, have been for a couple years. There's hardware out there you can actually run Linux on. For example one of the first HDMI sticks used Allwinner A10 and quickly had Linux running on it. Because of that I got a Rockchip one later, only to find there was no such Linux availability.

The downside is they've been too low end. A80 is the first SoC they've announced that actually looks like it could actually be useful for a Pandora successor.

 

[WizardStan]

Is the A80 true octo core big.LITTLE? Or like the other servings it can only do 4 at a time?

edit: from the liliputing release "Allwinner tells us the chip design will allow devices to use up to all 8 processor cores at once."

I think I need to change my pants.

 

[God Ginrai]

lpddr2 is on 5430, omap 5432 has lpddr3.

Page 118 8.1.2  http://www.ti.com/lit/ds/symlink/omap5432.pdf

The OMAP5432 device has two 32-bit-wide DDR3 channels running at a clock frequency of 533 MHz. The

device supports both 1.5-V (DDR3) and 1.35-V (DDR3L) voltage operation. Because the bus width is 32

bits, it is not compatible with DDR3 UB DIMM. Primarily, the OMAP5432 device is intended to use the on-

board DDR3 memories. The DDR3 standard supports up to 1066 MHz, the operating frequency is limited

to 533 MHz in the OMAP5432 context.

No, the OMAP5432 supports DDR3, not LPDDR3.

Is the A80 true octo core big.LITTLE? Or like the other servings it can only do 4 at a time?

edit: from the liliputing release "Allwinner tells us the chip design will allow devices to use up to all 8 processor cores at once."

I think I need to change my pants.

That sounds awesome. At the same time, tho', I wasn't aware you were supposed to be able to use the big and LITTLE parts of big.LITTLE at the same time. I thought the point was to switch everything that was running on one set to the smaller set when it doesn't need all of that power.

-God Ginrai

 

[Hồng Thất Công]

^Ps2 emulation I hear

 

[WizardStan]

At the same time, tho', I wasn't aware you were supposed to be able to use the big and LITTLE parts of big.LITTLE at the same time

As I understand it, there's basically two ways of doing it: either you get two clusters but the OS only sees one and as CPU demand goes up the SoC seamlessly switches from the low power cluster to the high power; or multiprocessing mode where the kernel does actually see all 8 cores and can schedule tasks to them manually. This was how ARM envisioned it, that you could do one or the other depending on your requirements, but to the best of my knowledge only the first way has ever been implemented, until now.

 

[God Ginrai]

At the same time, tho', I wasn't aware you were supposed to be able to use the big and LITTLE parts of big.LITTLE at the same time

As I understand it, there's basically two ways of doing it: either you get two clusters but the OS only sees one and as CPU demand goes up the SoC seamlessly switches from the low power cluster to the high power; or multiprocessing mode where the kernel does actually see all 8 cores and can schedule tasks to them manually. This was how ARM envisioned it, that you could do one or the other depending on your requirements, but to the best of my knowledge only the first way has ever been implemented, until now.

Oh neat. Well, that's cool to hear. I feel like the former feature will be more useful for me, however, having the latter available will be nice for someone who wants to port or code an application that really takes advantage of it.

-God Ginrai

 

[Reyhn]

Well, there are possibilites we all do not yet know. Because the chip manufacturers have a planning phase of 2-3 years in advance (called Roadmap). Because they need to design and verify chips before they announce them to the public.

So I am sure there will be new possibilities by end of Feb and another set of new possibilities end of March etc. And even more by end of this year. Because they have been prepared behind the curtain...

This means it is impossible to ever become aware of all possibilities. And only then decide.

ED must decide without knowing all possibilities.

Of course, I understand that, I meant possibilities at a given time, knowing that in x weeks/months a choice has to be made.

 

[comradekingu]

Allwinner sounds very much like the GPL violator i know it as, have they changed their ways? Some googling suggests maybe  http://lists.gpl-violations.org/pipermail/legal/2013-August/004197.html  http://lists.phcomp.co.uk/pipermail/arm-netbook/2012-June/004136.html

Edit: i corrected my post above. LPDDR3 is not the same as DDR3U

 

[God Ginrai]

Allwinner sounds very much like the GPL violator i know it as, have they changed their ways? Some googling suggests maybe  http://lists.gpl-violations.org/pipermail/legal/2013-August/004197.html  http://lists.phcomp.co.uk/pipermail/arm-netbook/2012-June/004136.html

Actually, those links you included seem to point to Olimex as the GPL violator, not Allwinner. In fact, the mailing list response even suggests good things about Allwinner themselves:

you will therefore be aware that we put out a request to allwinner to

provide the NAND driver source code, and you will be aware that they

did in fact provide this (albeit not the exact same version because

they had to bypass the manager who is deliberately and constructively

interfering with the direct explicit orders of the Directors of

Allwinner to comply with the GPL).

-God Ginrai

 

[comradekingu]

I thought that was the same manager that was working at allwinner, hence why i put maybe. Then as now, i couldnt find much sourcecode on their site, other than:

   Supports Android 4.1/4.2 and above
    Compatible with Microsoft Windows 8

which is the proclamation for A31.  Does allwinner have any linux support offering of their own, or is it all community hacks? Lets say we get a new soc like the A80, being stuck with android and win 8 till maybe support comes along seems a far cry from what TI has on their similar omap5.

 

[ShleeDragon]

I'd love to see the allwinner stats and support... they don't have the "brand" and history of the TI OMAP.

The Pandora had a lot of problems due to random unexpected problems. Do we forsee similar unexpected problems if we used the allwinner over the omap?

Is the OMAP smooth sailing?

 

[xiongxioi]

There seems to be conflicting statements from a few news article I've been reading. 

http://semiaccurate.com/2014/02/07/allwinner-shows-8-core-a80-soc-mid-range/

I don't want to be miss leading, so take what I'll be saying with a grain of salt. 

Allwinner is rather popular amidst the low-end cheapo Chinese tablets. Their chips can be clocked to rather high frequency. 

However, the last time I checked them, they tend to have ludicrously short battery life (half an hour of dead trigger on a tablet), they tend to freeze a lot (whether this is caused by cheap parts is up to debate), and will cook their own guts within a few hours of usage (allwinner tablets usually will be half dead in a month or two). Even though allwinner has been trying to clear its name since they announced the a23 (costs 3 bucks each), the word allwinner still rings a terrible bell in my weak heart. 

There's rumor stating the dev board of allwinner a80 has an Antutu benchmark of over 40k, which is way higher than MTK and Samsung's octo-core. 

I don't know if we should give them a chance, even though the a80 is more powerful than the 805, its failure rate is likely very high, ED's going to deal with lots of returned units.  Ti on the other hand, has been trusty since the beginning of time. However, this is the first time Allwinner's stepping in the high end of mobile soc market, so the A80 may not be so bad. I hope ED can get his hand on a dev board and stress test it a bit to get a feeling. 

MTK and allwinner are serious competitors, so anything came out of the mouth of MTK's marketing guy may or may not reflect the truth, that guy has said on the A80 is basically: even samsung would not do true octo (global task scheduling) cortex A15 on their tablets (samsung were using clustered migrition which is switching between the big cores and the small cores), I don't think any Chinese companies are capable of doing true octo cortex a15, that's biting more than they can chew. (MTK's true octo is eight A7 cores. )

 

[SNESFAN]

The thing that seems to fail the most on the el cheepo tablets and game handhelds are the NAND and battery. If these are said to be gotten elsewhere and are of higher quality, then I personally don't see a problem. Some have freezes but that could very easily just be bad software port with their less than fabulous android ports they seem to slap on them. But could also very well be the quality of the RAM chips that will more than likely be of higher quality. If the processors were bad you'd think that the systems would start showing repeatable errors in doing a specific thing (computation error) or the thing would just flat out refuse to turn on (different than not booting) or have a noticeable dip in performance after a period of use and thermal buildup. That's just my thoughts, and I can't say I've heard many reports on those specific issues on all winner. I have however heard those types of problems on amlogic, rockchip, general plus, and mediatek processors. While from a reputation standpoint TI seems to be the most favorable. Allwinner I feel should be given a chance to prove its self in some burn in stress tests for a couple weeks to see how it holds up. Also use better quality on the volitile parts.

 

[Exophase]

Is the A80 true octo core big.LITTLE? Or like the other servings it can only do 4 at a time?

edit: from the liliputing release "Allwinner tells us the chip design will allow devices to use up to all 8 processor cores at once."

I think I need to change my pants.

Allwinner has stressed that it supports true HMP (heterogeneous multiprocessing), in other words can run any mix of the cores simultaneously. I would assume that means the software part of that is working but who knows.

It was never the point of big.LITTLE to run only one cluster at a time, that's just a simplistic usage model. But Samsung screwed up the coherency hardware somehow, then even after they fixed that still failed to implement the software. One intermediate step between full HMP and whole cluster migration is single core migration, where you have four virtual cores and each one can switch between A7 or A15 as needed.

 

[Askarus]

One thing that comes to my mind:

OMAP does not support POP.

Do Snapdragon, Allwinner and Bay Trail support it?

If so it might be a nice gain of board space I think.