possible soc updates.


My thought as far as SoC choice is that it's worthwhile to aim for being able to claim it's more powerful then the Vita, mainly from a marketing perspective.  The nature of 28nm parts having a full 1.5 node advantage over the Vita, and based on available products this is a relatively easy bar to hit.  That should also give a boost of around 10-20x over the GHz Pandora in terms of GPU.

OpenGL ES 3.0 and OpenCL 1.1/1.2 compliance would be preferable from the perspective of being able to use the SoC's GPU assets more effectively.  Something that also would help with being able to point to the Vita in terms of comparison in order to establish better then to ward off the mentality that "well my smart phone is more powerful."  On the other hand for the rest of the year these newer GPUs are going to tend to be treated as premium parts material and thus may have issues with being less available and at a higher cost then older design block equipped units.

ARM's indicated range for handsets with the Cortex A15 is 1GHz to 1.5GHz.  In terms of theoretical DMIP throughput that should put the max at 75% to 163% faster per core versus the 1 GHZ Pandora.  The RK3188 at the indicated 1.6-2GHz max should theoretically be 100-150% faster per core.  Real world numbers should tend to be higher for both over the Cortex A8, and benchmarks tend to indicate the Cortex A15 has a much larger performance advantage then theoretical DMIP numbers would imply over the Cortex A9. 

So whether it'd be better to go with something like the Rockchip 3188 or the inevitable Rockchip 28nm follow up along the lines of a Exynos 5210 Dual, given that design is pretty much exactly what ARMH is pushing, I'm not sure given EvilDragon isn't exactly getting on the horse to look into this yet.
 
Last edited by a moderator:
If I understand it right, many (most?) of these SoC designs use shared RAM between CPU and GPU.  We're likely going to want this to drive 1080p at a decent frame rate, whether on the built in screen or on an attached monitor plus have enough RAM to run the OS, an abstraction layer (QEMU/DosBox/etc) and the program itself (in some cases).

If our intentions are to run a full install of Slackware, Debian, Ubuntu, or other instead of something like the Pandora's minimalist Angstrom, additional RAM is always nice to have over swap.  The trade off being the power that it takes to run the RAM.

Since we're talking about a hypothetical device for release in 2015, I would lean toward the 4GB range to ensure there is plenty to go around between shared graphics RAM (512MB?), OS, abstraction layer and running programs * 2.

CPU wise, at least 2 cores.  ARM or X86?  I think we should give whatever SoC comes out in the next 6 months that runs inside of a 4W envelope a good look - regardless of architecture.

If X86 22nm Atom chips can run under the thermal limit for a passive device (even if we have to run a heat spreader and exposed fins out the front) AND have enough processing capability to run the original Half Life and derivatives via Steam on Linux with a decent frame rate, that would be one heckuva hat trick.  If they fall short of that kind of performance/W, then there is more reason to go ARM than there is to go X86.
 
If X86 22nm Atom chips can run under the thermal limit for a passive device (even if we have to run a heat spreader and exposed fins out the front) AND have enough processing capability to run the original Half Life and derivatives via Steam on Linux with a decent frame rate, that would be one heckuva hat trick.  If they fall short of that kind of performance/W, then there is more reason to go ARM than there is to go X86.
Ignoring your notions of what constitutes reasonable TDP for a handheld, the original Half Life required a Pentium 233MHz as per the Platinum Collection documentation.  If you're just talking about the Original Half Life it actually should have been around a 133MHz Pentium.  Qemu should be able to handle that with a RK3188 based on where things are landing with Qemu testing.  Fallout suggests a 300MHz Pentium 1 and that appears to be running well on existing GHz edition Pandoras via Qemu on Windows 95.  So ignoring your notions regarding 1080p, I'm not really seeing the issue provided the GPU block is updated to Vita competitive levels.

I looked into it and my research tended to indicate that there really weren't really a meaningful number of Pentium II or Pentium III focused titles.  It basically went from Pentium 1 to GHz+ Athlon, etc. with the ability to take advantage of the processors along the way.

If you want to talk about the modern Steam version the Linux system requirements according to Steamworks are:

  •  Minimum: Linux Ubuntu 12.04, Dual-core from Intel or AMD at 2.8 GHz, 1GB Memory, nVidia GeForce 8600/9600GT, ATI/AMD Radeaon HD2600/3600 (Graphic Drivers: nVidia 310, AMD 12.11), OpenGL 2.1, 4GB Hard Drive Space, OpenAL Compatible Sound Card

which I can say with almost absolute confidence a 22nm Atom processor within a power threshold reasonable for a handheld can't provide.  Now the nature of those requirements aren't directly related to the game itself but you're effectively including Steamworks and other infrastructure in the mix.

So I'm at something of a loss what you're trying to get at with this.
 
I noticed a lot of people overthinking the whole SOC discussion a bit.

In my personal opinion: X86 is not worth looking at as the Pandora's are designed as a open source model. That means Linux / Android / ... Wine "emulation" is again a problem. I get the feeling that people have not tried Wine before. It works, but half the time it requires fixes for problems, visual imperfections, mouse problems, etc ... this especially with games. And that is on x86 / Full OpenGL platform! The SOC we are talking about, are not full OpenGL ... have fun with rewriting Wine. Its nice pip dream from people, so they can play the massive amount of PC x86 games, but its just unrealistic. Even Nvidia with there Shield / Tegra4, was happy to just "stream" X86 games from computers in the room, turning the console in a glorified handheld controller.

Now for SOCs ... there are plenty of factors. Costs, power usage, integration, documentation, etc ... Going with the latest & fasted is all good and well, i think nobody will say no to a Tegra4 Pandora, the problem is... each has its problems ( Tegra 4 has a massive power drain when its A15 cores ). A7 is also a platform that is best avoided. While it saves power, because of its relative slowness, tick for tat, in a lot of cases, its actually worse, then a A9 core ( at the same manufacturing process. Do not compare older A9's vs new A7, with different manufacturing processes he ). Best example is a lot of Chinese Tablets, with those Quad A7, where you see a lot of people complaining, especially with high resolution screens.

I got some stupid Quad Core Samsung 5512 ( 32nm ) or whatever Chinese tablet around here with a 10" 1080p display and a relative small battery. Yet ... the thing runs for 6 to 8 hours ( what most people do not expect ). And that is just a "old" A9 design. It still runs circles around the Pandora's antiquated SOC.

One does not need the latest of the latest A15 ( what i personally advice against, unless you got a little.big, and then your just running a bunch of A7 cores anyway ). The more important things are:

* Cost of the Pandora2 itself

* Delivery capability

No offense, but a 400/500$ handheld toy is just that, a expensive toy. Even Nvidia needed to lower there price for the Shield, because any mass market appeal under pressure.

In my opinion, a Quad A9 ( 22nm ) or a Quad Krait ( 22nm ), are much better solutions, then A7/A15/X86 ...

And there is also the software angle. What MORE can you do, on a Quad A15, compared to a Quad A9/A7/Krait. Lets people give me a good answer on this one, and then the focus is much more clear. All the rest is just people dreaming of the impossible. But while people here are *kuch* arguing about what the Pandora2 can do, we see more & more "gaming" tablets appearing from some Chinese no-name ( and some slightly good Chinese manufactures ), who also have a large part of the same software available ( from the Android angle ), maybe not as optimized, but that is a different matter.

I am just saying, that the market is already going past the original goals provided by the Pandora 1. In its time, it was revolution. Today, its less then spectacular ( yes, i will be stepping on peoples feet, by saying that *lol* ). From this angle, the Pandora2 needs more specialized features, good price, and deliverable ( not in 2 months times 18 *lol* ), but also sensible solutions regarding SOCs. The RK3188 is not a bad ( a,d cheap / efficient ) solution, for a product in 2013. But beyond that... 
 
X86 is not worth looking at as the Pandora's are designed as a open source model. That means Linux / Android /
Linux runs perfectly well on x86 (aas does Android, mostly)

I'm not seeing your argument here.

the market is already going past the original goals provided by the Pandora 1
Hogwash. The pandora has all the GNU stuff ootb, a keyboard, and proper USB host.

Even the best Android gaming tablets are pitifully bad at doing real work.

What MORE can you do, on a Quad A15, compared to a Quad A9/A7/Krait.
Editing larger images, raytracing, complex realtime audio manipulation, smooth webbrowsing of badly-written sites, soft-decode high-res video, run CPU-intensive single-threaded games, improve the performance of QEMU.
 
Just stumbled upon on this ARM A9 quad core named TM7029 and manufactured by Action Semiconductor.

http://www.actions-semi.com/en/productview.aspx?cat=98&id=108

I wanna know from you, what you think about it.

Maybe it's not for Pandora 2 ...but what do you think about performance/specs ?
I've never heard of this company before, and they're not providing an awful lot of important information about their SoCs.. but from what they're saying they look pretty bottom-of-the-pile for Chinese solutions. Would easily pick RK3188 over this.

But a big risk in using any of these Chinese SoCs is that they may not be supported for years and years like TI's was, nor replaced with alternatives that are drop in compatible. Pandora 2's developers should prepare for this and be ready to replace CPU modules down the road..
 
Last edited by a moderator:
--snip--

What MORE can you do, on a Quad A15, compared to a Quad A9/A7/Krait.
Editing larger images, raytracing, complex realtime audio manipulation, smooth webbrowsing of badly-written sites, soft-decode high-res video, run CPU-intensive single-threaded games, improve the performance of QEMU.
Ahem. Our 8974s (The 800 line)  either match or win out of A15 and Apple cores in many CPU benchmarks and get the highest FPS by quite a longshot in all of Anandtech's benchmarks. See here: http://www.anandtech.com/show/7082/snapdragon-800-msm8974-performance-preview-qualcomm-mobile-development-tablet

Krait all the way  :wub:
 
Oh, I'd LOVE to get a Snapdragon 800... but would a company like Qualcomm even care about such a niche as us?


(Heh, I bough Eudora a looong time ago...)


Still, a device powered by a Snapdragon sold by EvilDragon in his DragonBox shop... that sounds awesome :)


I wouldn't even know who to ask at Qualcomm. Not much information about contacts on the website...
 
Don't know where this rumor that large chip makers won't sell in 1000's came from. Has there been any confirmed proof of this?


In this economy , I'd be surprised if they didn't sell you whatever they could. Especially if its for something that gets the chip into gaming, an area their competition ,Nvidia, wants to dominate in and which they have no real footing in at the moment
 
Last edited by a moderator:
When they sell a chip they need to do some kind of support, design help, prepare drivers, etc. All that costs them money, and for such tiny potential profits they might just not bother.
 
Last edited by a moderator:
Smaller quantities usually go through distributors and the overhead involved with this isn't worth bothering with. They'd much rather keep things to big contracts. They're probably not even really prepared to sell in retail channels instead of making tailored contracts each time. Least appealing of all is setting up their own distribution like TI and Freescale did.. But even those didn't do it for their high volume phone chips.

Small orders means splitting up groups of wafers, lots, etc into different sets. There could be a ton of overhead that makes it undesirable.

I've heard it from several industry insiders over the years, also you can often find disclaimers in product briefs that they're meant for high volume phone products. That and the complete lack of any kind of advertising, some of them don't even have sales contacts on their websites. I've also heard various people say they were denied, I'm sure ED can tell you more if he wants.
 
Last edited by a moderator:
So how did Ouya get Nvidia to commit 6000 T3 SOCs for their order. That's all they asked for at the beginning of thir KS.


I'm sure it had something to do with the strategy to be known as the soc for gaming. This is why I mentioned there might be a chance with Qualcomm, that's an area they need to get into to address Nvidia running away with it. So they may have an open ear.


Anyway, one of these next gen SOCs has to be used , so a way to get them has to be worked out some how.
 
Because nVidia wanted a pet project to sponsor. They wanted in-roads with a gaming device to push their game store and brand. This was apparent from the start. OUYA was never just another customer to them.

That and Tegra 3 was old even when the KS started and lost much of its marketability already.

The thing with something like OUYA vs something like Pandora is the former is a pretty low risk hardware-wise because it's just a box (and a controller, but a lot of people have done those). There are less complications than with a handheld. And given Pandora's track record I can't see a lot of companies wanting to make it a specially sponsored project.

It doesn't matter how much a next gen SoC has to be used, if they're not available they're not available, and right now there are painfully few options that even make sense IF available. Something like RK3188 is at least realistic since Chinese companies do sell through distributors. It's not a great option but it's better than nothing.
 
Last edited by a moderator:
^ I agree about Nvidia, thats basically what I said in my previous post re them getting involved with Ouya for the foot hold into gaming etc.  Although , one thing I don't agree with is the T3 age, when the Ouya KS occured , the T3 T33 (which the Ouya uses) was just hitting the market, even the initial low end T3 had only launched in products 5-7 months prior.

Anyway,  I think we should give it a crack instead of saying I don't think it's possible therefor I wont try.  At minimum , at least ED could  call  them to get a feel .

Things like this are not new to SOCS, trying to procure major software packages from the big vendors for example (not naming anyone) was also very difficult in my previous roles.  It's true, some don't want to deal with small orders , ie a  low profit to support ratio.  Sales staff would rather pursue bigger customers.  That's why it's imperative to be quite determined when it comes to procurement, believe me, I've been there many times. It means they never call you back, you have to pursue them every step of the way and some times even combine your order with someone else. 

At the end of the day , something has to happen, or there wont be a P2 (or Dragon Box), a RK3188 just isn't worth the expense in research and development in a P2 IMO.  You would be asking for $500 - $600 for a P2 with a SOC people associate with budget devices.  That's a very hard sell 

At least with an ED product we can distance it from the Pandroa, I'd suggest not only a new name , but new forum etc. Get as far away from the Pandora as possible. At least then there would be hope for adoption and sponsorship.
 
Last edited by a moderator:
I never said don't try. Just saying that people have to understand that there's a very good chance it won't happen.

Tegra 3 really was old for a new design with OUYA's projected release date. Maybe it seemed new enough at the time, because you don't normally see what components are selected when new devices are being developed. Do you think they'd start with something that first showed up in devices 6+ months ago? I mean, Pandora 2 probably would have to, that's true.

I don't think an RK3188 Pandora isn't worth it if it replaces the current one we have. Better that than nothing, and better selling that for $500 than selling the current one for $600. That SoC is still a big improvement over what's in the current device. It'd be an easier sell than the current one. Probably just as bit of a niche but you'd get a lot of current Pandora users upgrading. And it'd probably just be a stopgap. With a modular design upgrading a year or two down the gap wouldn't be terribly expensive.

Not that I like the option at all, I'm just saying if it's the best there is then I think they should do it.
 
Last edited by a moderator:
^ The thing is there are several versions of the T3 , the T3 T33 which the Ouya uses just hit the market ( 1-2 months tops) when they launched the Ouya on KS.  It was the slower T30 T3 that was 5-7 months old.

The RK3188 is a good SOC, no doubt, I cant say otherwise given it's me that has posted about it the most. And yes it would be better than the current SOC by a little bit IF Rockchip get  Linux support in order which they have only just taken first steps towards. HW Acceleration with the Mali GPu is still an issue for the time being though.

Assuming the above is sorted out, It would be a better device, and would sell more than the minuscule amount the P1 is selling .  But more is relative, how much more would be the question - not a great deal would be my opinion and probably not enough to warrant the expense in developing it.  I certainly wouldn't spend $500 - $600 on a RK3188 SOC based device, and I'm a fan of that SOC.

My opinion would be if you cant get at least a T4 at minimum or a chinese equiv that comes close, then don't bother.  I know that sounds negative, but it's  a realistic view IMO.

There are some interesting chinese SOCS on the horizon, I know there is a MTK Cortex A7 Octa core one launching soon with a decent Antutu (~ 30,000) score . See here:  http://www.phonearena.com/news/MediaTek-to-launch-2GHz-octa-core-SoC-the-MT6592_id44804

Will keep an eye out for them and others, but having said that, these guys sell as many SOCS as the companies in the west do, so you could apply the same arguments all over again about procurement
 
Last edited by a moderator:
A modestly better binning of the same SoC does not a new SoC make. That's the very least I would have expected them to do.

If it's even that. More likely just allowed to clock higher due to relaxed power limit.

Tegra 4 should have been realistic if it didn't end up getting delayed a fair bit. So that would have probably delayed OUYA with it. Not that it really matters, it'd have driven up the price substantially and that's the major benefit OUYA was going for. Let's face it, the budget angle is really big with it.

Is GPU support w/Mali-400 really going to be a big issue given how far the reverse engineered effort has come? Maybe I'm overly optimistic but I think that should work okay for a Pandora 2 release. But relying on anything from another RE effort could be pushing it..

Now, I don't have all the information I need, but Tegra 4 is not a runaway winner vs something like RK3188 for Pandora, given that the power consumption is going to be really high. I don't know what the perf/W will be like (I'm sure nVidia would have me believe that it'll be better, that we can clock their processor at 800MHz and it'll perform like an A9 at 1.6GHz all the time - won't happen) but I fully expect that even in the lower levels it'll be worse. Actually, even nVidia's claims are just something like 40% better on Tegra 4 than Tegra 3.. but that's with a process advantage that it wouldn't have over other A9s.

Besides, who's to say there'll be good Linux support on T4 either?

As far as I'm concerned MediaTek's 8 core A7 would be a much worse chip than RK3188, the single-threaded performance will be way behind and there will be close to nothing that any of us would be able to do that could make up the difference by using > 4 threads. Also do me a favor and don't mention AnTuTu scores to me in the future :p (btw MediaTek is Taiwanese, industry is somewhat removed from China, they do have the world's leading pure-play foundry.. doesn't mean MTK hasn't been essentially playing in the same market as Rockchip though)
 
Last edited by a moderator:
^ Yeah I know Antutu scores are not that accurate a way to measure a SOCs performance, but I think it was the only benchmark they released.  That MTK was just an example, I believe there others coming, I'm actually looking closely at what Rockchip come up with next.  Perhaps a dual or quad A15 , here's hoping.

But then again, as I mentioned before, Rockchip are a very large SOC manufacturer.  IF your worried about a big company selling small quantities, then your out of luck here also .  Looks like no matter which way you cut it, ED will need to try and get SOCs from one of the big guys.  
 
Last edited by a moderator:
You seriously think those HDMI sticks are high volume? Like, any one of them? With their shelf life before being replaced with a different one?

I'm saying that companies like Qualcomm don't sell low volume/to distributors not simply because they're large - TI and Freescale are both plenty large and both sold SoCs of interest in low volume through direct distribution no less. This is something that goes way back to their days of supplying components decades ago when there were no super high volume markets. Companies like Qualcomm don't do it because it's just not their business model, they're not setup for it and the overhead of doing it a few times one off isn't worth it. They're setup to secure contracts and provide support for very big orders, probably with some fixed overhead and limited resources.

I mean really, do you think that there are no low volume Qualcomm products out there just because no one tried hard enough to get through to them (please tell me if there's one I'm not aware of)? Pandora is hardly the only ones in this position. They'd most likely have to be appealing as a pet project that the company wants to invest one.. but the thing is Pandora doesn't offer anything that should interest Qualcomm specifically. Even if they were interested in pursuing a gaming platform to rival nVidia it's not really a great one for that - nVidia thinks they can sway top tier developers to target Shield specifically but I doubt Qualcomm would believe something like that in a Pandora device.

nVidia is somewhat of a maybe, at least more likely than Qualcomm.

The Chinese companies are totally different, they give stuff to Chinese distributors to fetter out to about a billion different manufacturers because that's the nature of their market. They don't try to win big contracts because huge players aren't that interested. Their situation is more like TI's in the 70s.

Re:AnTuTu, please just don't mention it at all :p "It's all that's out" shouldn't be an excuse. Don't let it pique your interest.
 
Last edited by a moderator:
Back
Top