iPad 2

[Lazy8s]

I guess I hadn't been paying attention: I started up Rage HD when connected to an HDTV today and got the nice surprise of native 1080p detail.

The visuals have a fair degree of complexity but nothing that begins to stress the 543MP2, and I really had not been able to appreciate the engine cooked up by Carmack and crew until seeing it this way.

 

[Pressure]

I guess I hadn't been paying attention: I started up Rage HD when connected to an HDTV today and got the nice surprise of native 1080p detail.

The visuals have a fair degree of complexity but nothing that begins to stress the 543MP2, and I really had not been able to appreciate the engine cooked up by Carmack and crew until seeing it this way.

I'm starting to wonder what kind of performance figures and power handling a 543MP16 really has.

Not to mention Series 6 'Rogue' ...

At some point we will have enough performance for 60 FPS at 1080p, which should satisfy the vast majority.

 

[wco81]

I'm starting to wonder what kind of performance figures and power handling a 543MP16 really has.

Not to mention Series 6 'Rogue' ...

At some point we will have enough performance for 60 FPS at 1080p, which should satisfy the vast majority.

Question is whether the App Store will support a pricing model for such games, meant to be connected to an HDTV.

Does Infinity Blade sell in high enough numbers to encourage other developers to make similar games?

Does Apple sell enough HDMI dongles for developers to target 1080p60 games? (Of course if the rumors are true, the next iPad will be able to display 1080p directly)

And finally, is Apple going to support pairing of some kind of physical controller to really push "serious" games on the platform? Maybe better if they came out with their own BT controller rather than rely on some 3rd-party peripheral to catch on.

 

[Pressure]

Question is whether the App Store will support a pricing model for such games, meant to be connected to an HDTV.

Does Infinity Blade sell in high enough numbers to encourage other developers to make similar games?

Does Apple sell enough HDMI dongles for developers to target 1080p60 games? (Of course if the rumors are true, the next iPad will be able to display 1080p directly)

And finally, is Apple going to support pairing of some kind of physical controller to really push "serious" games on the platform? Maybe better if they came out with their own BT controller rather than rely on some 3rd-party peripheral to catch on.

It sold 300,000 copies the first 5 days. It's only a question of time before the market is large enough.
Considering you already have way past 20 million iPads on the loose.

But 1080p60 gaming is probably still 2 years away with the speed we move at, although I am very interested in seeing what a dual-core ARM Cortex-A15 paired with PowerVR Series 6 is capable of in terms of visuals and performance (which I strongly believe the Apple A6 SoC will contain).

It will probably also be AirPlay-enabled, to get rid of the cable. Awesome times ahead for sure.

Not saying this will happen with the next iPad but I can clearly see it in the "future".

It would be wrong not to be afraid of Apple if you were in the console business.

 

[Exophase]

But 1080p60 gaming is probably still 2 years away with the speed we move at, although I am very interested in seeing what a dual-core ARM Cortex-A15 paired with PowerVR Series 6 is capable of in terms of visuals and performance (which I strongly believe the Apple A6 SoC will contain).

Is your strong belief inspired by anything concrete?

Considering the roughly one-year cadence of products I'd be very surprised if Apple got a Cortex-A15/Series 6 sporting iPad out in spring of 2012. A spring 2013 release for such a device sounds more reasonable, although it's possible that iPhone 6 would lead with such a device in late 2012. I expect A6 to be a 28nm A5 with higher clocks and lower power consumption, and possibly an increase in cores (either for the CPU or GPU). This would be in line with what we've seen going from iPhone 2G to 3G to 3GS to 4, for instance.

 

[Pressure]

Is your strong belief inspired by anything concrete?

Considering the roughly one-year cadence of products I'd be very surprised if Apple got a Cortex-A15/Series 6 sporting iPad out in spring of 2012. A spring 2013 release for such a device sounds more reasonable, although it's possible that iPhone 6 would lead with such a device in late 2012. I expect A6 to be a 28nm A5 with higher clocks and lower power consumption, and possibly an increase in cores (either for the CPU or GPU). This would be in line with what we've seen going from iPhone 2G to 3G to 3GS to 4, for instance.

Not in any solid facts, no.

Of course, another possibility is an A6 based on a quad-core Cortex-A9 and SGX543MP4 but you could achieve better performance with a dual-core Cortex-A15 and with a "Rogue" GPU. Here I am specifically thinking that their competitors will have access to NVIDIA's Kal-El (Tegra 3) shortly.

They have the engineering capabilities (since they acquired both Intrinsity and P.A. Semi) for designing both SoCs.

Apple created the modern, lucrative tablet market so they should have a desire to stay in the race.

 

[tangey]

Not in any solid facts, no.

Of course, another possibility is an A6 based on a quad-core Cortex-A9 and SGX543MP4 but you could achieve better performance with a dual-core Cortex-A15 and with a "Rogue" GPU. Here I am specifically thinking that their competitors will have access to NVIDIA's Kal-El (Tegra 3) shortly.

They have the engineering capabilities (since they acquired both Intrinsity and P.A. Semi) for designing both SoCs.

Apple created the modern, lucrative tablet market so they should have a desire to stay in the race.

Next year is too early to see rogue, for rogue think 2013.
A next gen apple chip might be an sgx543mp4, or indeed the current mp2 just doubled in clocked, or a combination.

 

[Ailuros]

Not in any solid facts, no.

I'd love to have tomorrow's hw today too.

Of course, another possibility is an A6 based on a quad-core Cortex-A9 and SGX543MP4 but you could achieve better performance with a dual-core Cortex-A15 and with a "Rogue" GPU.

And what about power consumption? Just because some funky go merry analysts speculate that Apple might begin mass production for a hypothetical 28nm A6 in Q2 2012 and even release it in the very same quarter (June) it shouldn't mean that it necessarily reflects reality. With the quantities Apple is dealing with, I believe that kind of horseshit when I see it.

Here I am specifically thinking that their competitors will have access to NVIDIA's Kal-El (Tegra 3) shortly.

Most likely a quad core A9@1.5GHz and a 2 Vec4 PS/1Vec4 VS ULP GF clocked higher (=/>400MHz?

I don't know if Apple could feel any pressure for the quad A9 CPU side of things, but for the GPU side of things they need to worry why exactly and from where? The SGX543MP2 is in Egypt offscreen 720p by 4x times faster than Tegra2 and by 2x compared to Mali400MP2 and that by most likely being clocked lower than both of them.

They have the engineering capabilities (since they acquired both Intrinsity and P.A. Semi) for designing both SoCs.

They obviously didn't care how big A5 would turn out as long as performance increased compared to iPad1 without increasing power consumption. That >120mm2 is quite expensive and for the lifetime of both the iPad2 and upcoming iPhone5 a die shrink would make a lot of sense and yes they could obviously also increase frequencies.

Apple created the modern, lucrative tablet market so they should have a desire to stay in the race.

They also have their very own software ecosystem and their own application store. At worse they might stay for the entire 2012 with a high clocked dual A9. Will that keep them selling as before? By the way I'd like to stand corrected but I think iPads sold somewhere around 40 and not 20M units all together.

 

[Pressure]

I'd love to have tomorrow's hw today too.
And what about power consumption? Just because some funky go merry analysts speculate that Apple might begin mass production for a hypothetical 28nm A6 in Q2 2012 and even release it in the very same quarter (June) it shouldn't mean that it necessarily reflects reality. With the quantities Apple is dealing with, I believe that kind of horseshit when I see it.

Cortex-A15 is (should be?) more power-efficient than Cortex-A9 but a shrink of the current architecture makes sense, like a tick-tock Intel strategy kind of way.

A shrink to 28 nm should take care of power consumption and die size.

I just figured that they would need more throughout in graphic power to effortlessly increase the pixel count 4 times (to 2048 x 1536) and maintain the performance of the current iPad 2.

Most likely a quad core A9@1.5GHz and a 2 Vec4 PS/1Vec4 VS ULP GF clocked higher (=/>400MHz?

I don't know if Apple could feel any pressure for the quad A9 CPU side of things, but for the GPU side of things they need to worry why exactly and from where? The SGX543MP2 is in Egypt offscreen 720p by 4x times faster than Tegra2 and by 2x compared to Mali400MP2 and that by most likely being clocked lower than both of them.

When I look back at what I wrote, this has more to do with the pixel count increase. The only comparable thing, performance-wise, is probably going to be the PS Vita.

They obviously didn't care how big A5 would turn out as long as performance increased compared to iPad1 without increasing power consumption. That >120mm2 is quite expensive and for the lifetime of both the iPad2 and upcoming iPhone5 a die shrink would make a lot of sense and yes they could obviously also increase frequencies.

You are correct on that account but they are still shoveling money in the bank at current prices but yeah, maximizing profits can only improve the bottom-line.

At some point though I feel like they should differentiate the SoCs from the iPad and iPhone / iPod Touch line.

They also have their very own software ecosystem and their own application store. At worse they might stay for the entire 2012 with a high clocked dual A9. Will that keep them selling as before? By the way I'd like to stand corrected but I think iPads sold somewhere around 40 and not 20M units all together.

Yeah, they did sell 16 million iPad 2's last quarter alone.

In general, I may just be too optimistic regarding the next SoC heh. But it sure is fun to speculate and thanks for calling me out on my over-the-top daydreaming.

 

[Ailuros]

Cortex-A15 is (should be?) more power-efficient than Cortex-A9 but a shrink of the current architecture makes sense, like a tick-tock Intel strategy kind of way.

A shrink to 28 nm should take care of power consumption and die size.

I don't have an idea if Apple is amongst the A15 lead partners; even if with Apple's typical secrecy we'll probably find out quite late.

I just figured that they would need more throughout in graphic power to effortlessly increase the pixel count 4 times (to 2048 x 1536) and maintain the performance of the current iPad 2.

Will the iPad2 successor use in such a hypothetical case 2048*1536 as a render target or a lower resolution for 3D? (honest question). Apart from that Mali400MP4 and the SGX543MP2/iPad2 are the only GPU blocks in the embedded space right now I'm aware of that count 4 TMUs in total. I assume (but guess that no one knows apart from Apple) that the MP2 is clocked at around 250MHz which gives a raw texel fill-rate of 1000MPixels and 2500MPixels with a 2.5 depth complexity. Each SGX543 has 16 z/stencil units, 32 all together for the MP2.

I don't know of how many units each Rogue core consists, but assuming from the Novathor A9600 performance claims it might be a MP2, which could mean 4 TMUs/core. Only if Apple goes for Rogue and at the same time MP will the resulting GPU block have higher texel and z/stencil fillrates per clock than the MP2.

I'm even willing to bet that ST Ericsson might not manage to get its Novathor A9600 out the door before 2013.

When I look back at what I wrote, this has more to do with the pixel count increase. The only comparable thing, performance-wise, is probably going to be the PS Vita.

Which is clocked at 200MHz under Samsung 45nm. Obviously under 32nm and even more at 28nm higher frequencies will be possible.

You are correct on that account but they are still shoveling money in the bank at current prices but yeah, maximizing profits can only improve the bottom-line.

At some point though I feel like they should differentiate the SoCs from the iPad and iPhone / iPod Touch line.

My simple reasoning just tells me that since the A5 is quite large and might have lowered their margins, they might want to shovel a "few" bucks back from that "loss". I might be completely wrong with that one, but I wouldn't consider something like a dual A9@=/>1.5GHz and a SGX543MP2@=/>400MHz to not be competitive. Given that Apple has its own sw ecosystem I'd claim exactly the opposite. I don't know how good 32nm could be but if they'd manage to come close to 500MHz it would make a MP4@250MHz redundant and probably at even less die area.

In general, I may just be too optimistic regarding the next SoC heh. But it sure is fun to speculate and thanks for calling me out on my over-the-top daydreaming.

Look at it that way: ST Ericsson is promising with its A9600 something that equals roughly PS3/XBox360 performance at least from the side of the GPU block. The difference compared to what we have today is HUGE. I'd rather say everything at the right time, even more so considering that manufacturing processes turn out more and more problematic at their kick start:

http://www.fudzilla.com/home/item/23909-28nm-not-in-great-shape

What are happy go merry analysts speculating again right now? Mass production and mass availability within Q2 2012? Yeahrightsureok....I could devour that Apple might start mass production that early, but given how tight things might be at TSMC both in terms of capacities and yields it sounds more like they'll be collecting inventory for whenever they've really planned a hard launch. Unless you'd expect the new Apple CEO to have a woodscrew moment and hold up an iPad3 to the press for it to be available several months later

 

[ltcommander.data]

I guess I hadn't been paying attention: I started up Rage HD when connected to an HDTV today and got the nice surprise of native 1080p detail.

The visuals have a fair degree of complexity but nothing that begins to stress the 543MP2, and I really had not been able to appreciate the engine cooked up by Carmack and crew until seeing it this way.

http://www.bethblog.com/index.php/2010/10/29/john-carmack-discusses-rage-on-iphoneipadipod-touch/

RAGE's quality is noteworthy because it still only uses OpenGL ES 1.1 being based on the Doom Resurrection engine. Things would really get interesting if Carmack had the time to write a new mobile OpenGL ES 2.0 engine and start using shaders.

Is your strong belief inspired by anything concrete?

Considering the roughly one-year cadence of products I'd be very surprised if Apple got a Cortex-A15/Series 6 sporting iPad out in spring of 2012. A spring 2013 release for such a device sounds more reasonable, although it's possible that iPhone 6 would lead with such a device in late 2012. I expect A6 to be a 28nm A5 with higher clocks and lower power consumption, and possibly an increase in cores (either for the CPU or GPU). This would be in line with what we've seen going from iPhone 2G to 3G to 3GS to 4, for instance.

http://news.cnet.com/8301-13924_3-20097766-64/a6-chip-to-reach-ipad-3-later-in-2012-says-analyst/

As analysts tend to do, one recently talked about the Apple A6 being quad core as if its a surety. If its true, I'm presuming it'll be quad core Cortex A9 rather than quad core Cortex A15 which would be very aggressive. As you say, given Apple's 2 year reuse of architectures, ARM11+PowerVR MBX and Cortex A8+PowerVR SGX535, continuing to use Cortex A9+ PowerVR SGX543MP makes the most sense, perhaps moving to quad core CPU and GPU if clock speed bumps can't give the required performance increase.

I guess I hadn't been paying attention: I started up Rage HD when connected to an HDTV today and got the nice surprise of native 1080p detail.

The visuals have a fair degree of complexity but nothing that begins to stress the 543MP2, and I really had not been able to appreciate the engine cooked up by Carmack and crew until seeing it this way.

Is your strong belief inspired by anything concrete?

Considering the roughly one-year cadence of products I'd be very surprised if Apple got a Cortex-A15/Series 6 sporting iPad out in spring of 2012. A spring 2013 release for such a device sounds more reasonable, although it's possible that iPhone 6 would lead with such a device in late 2012. I expect A6 to be a 28nm A5 with higher clocks and lower power consumption, and possibly an increase in cores (either for the CPU or GPU). This would be in line with what we've seen going from iPhone 2G to 3G to 3GS to 4, for instance.

At some point though I feel like they should differentiate the SoCs from the iPad and iPhone / iPod Touch line.

Well they might be starting a little of that if the rumours are to be believed that the 5th gen iPod Touch will be a minor refresh using the same SoC since iPod4,2 numbers have been identified in iOS 5 betas but not iPod5,1. Apple's device version numbers seem to consider clock speed increases a major change, hence the 533MHz 2nd gen iPod Touch being iPod2,1 while the 2nd gen iPhone 3G kept its 412MHz clock speed and was named iPhone1,2. An iPod 4,2 would presumably then remain an Apple A4 with the same clock speeds, but perhaps a transition to 512MB of RAM like the iPhone 4. Keeping the iPod Touch 1 SoC generation behind could be useful if they plan on keeping costs down to lower the price points into previous iPod Nano territory and perhaps directly challenge the 3DS's new price. There maybe battery life benefits to using a mature SoC as well assuming a mature stepping on a mature process drives down power consumption as practiced by Intel and AMD.

 

[RDGoodla]

Cortex-A15 is (should be?) more power-efficient than Cortex-A9 but a shrink of the current architecture makes sense, like a tick-tock Intel strategy kind of way.

A shrink to 28 nm should take care of power consumption and die size.

I just figured that they would need more throughout in graphic power to effortlessly increase the pixel count 4 times (to 2048 x 1536) and maintain the performance of the current iPad 2.

Even in iPad 3 developers may choose 1024 X 768 for optimal performance in game Apps.

In other words, four smaller pixels in iPad 3 become one larger pixel.

Three years later when high-end mobile devices have 2 times of PS3 graphical power (in 20 nm CMOS) then the 2048 X 1536 resolution will become a reasonable choice.

 

[Arun]

Don't have the time to reply to everything now, but I'd like to see 1024x768 4xMSAA *without* downsampling so the subpixels essentially become the pixels in the final 2048x1536 framebuffer. I suppose this could be done by rendering to a 4xMSAA off-screen buffer and then reading it as a texture *before* downsampling in a post-process pixel shader writing to the final framebuffer. While that would waste a fair bit of bandwidth and I'm not sure current extensions expose that functionality, it's not that bad and I'm sure they will eventually. Then again I doubt developers will bother given the relatively small benefit.

Ideally SGX could simply handle the entire process as normal MSAA then skip downsampling, but that would require an even more proprietary extension which is even less likely to be available (if the HW can even handle it which is far from certain). Sigh, enough daydreaming on pointless details!

 

[rpg.314]

I'd like to see rendering at 4x resolution. It's the best.

EDIT: Changed SSAA to 4x resolution.

 

[Exophase]

That's an interesting idea Arun. It doesn't seem like the ability to render to an off-screen texture and render to the framebuffer like that would represent two different levels of fidelity for this feature. I expect the hardware has no concept of a dedicated framebuffer and sees all render targets as the same.

Of course you could no longer call this AA. From some online polls it looks like there are differing opinions as to what's preferable, high res w/o AA or low res with AA. Of course with just multi-sample w/o down-sampling you'd only get sharper edges instead of more texture detail.. would like to see how people weigh the two against each other..

(I wonder what you'd call this.. instead of multi-sample anti-aliasing maybe sub-textured rendering )

I'd like to see 4xSSAA. It's the best.

Do SGX GPUs have hardware support for 4xSSAA? If not, I take it they're at least smart enough to let you render to texture then use that texture in the same tile render pass w/o a resolve, although I doubt the texture cache would be coherent with the tile SRAM.

 

[rpg.314]

Do SGX GPUs have hardware support for 4xSSAA? If not, I take it they're at least smart enough to let you render to texture then use that texture in the same tile render pass w/o a resolve, although I doubt the texture cache would be coherent with the tile SRAM.

I meant 4x resolution. I edited my last post.

 

[Ailuros]

I love it that you folks by mistake or on purpose mention supersampling, multisampling, whatever sampling but no one seems to care about a healthy portion of anisotropic filtering.

To me well filtered, as noiseless as possible with a fine balance of sharpness textures are far more important if we're going to talk about higher resolutions. Supersample or upsample as much as you want while wasting tons of fill-rate, bandwidth and memory footprint and stick to Voodoo style 1999 blurry textures.

Chance given I still don't understand why developers bothered only to enable higher resolution textures and MSAA for the iPad2 in some games, yet neglected to enable even a small portion of AF.

Exophase,

I don't see it being impossible in theory to combine on SGX 4x sparsed MSAA with 4x OGSS. I'd still ask why though and not just replace the latter with AF. Granted it won't cater for any possible shader aliasing, but you can't have it all either considering how much cheaper AF is in the end. What's the real benefit for the hypothetical OGSS part of the above combination apart from shader aliasing? A negative LOD offset for sharper textures without any noticable noise maybe? That's a huge waste of resources for such a tiny offset compared to how AF can "sharpen" textures.

 

[Exophase]

No one enables AF in iPad 2 games because SGX doesn't have AF; place the blame squarely on IMG. Adreno does have it, or at least it did and I assume they didn't drop it. I don't know if anyone else does.

Of course just because we're talking about AA and down-sampling and what have you doesn't mean we don't care about AF, but I have a hard time considering it a more important feature.. AF tends to only effect a small part of the screen and that depends largely on what kind of scene. AA also improves image quality. And MSAA doesn't waste any fill-rate or bandwidth on SGX and similar tilers, mostly just spends depth-fill. I don't know how it compares with AF in terms of die size but it seems like with AF you'd need more if you have it on every TMU (maybe you could separate it so there's just some shared AF unit sitting out there when it's needed - maybe this is how it's done?)

I'm not advocating using OGSS under any circumstances here, I was just responding to rpg.314's post. Which was just a typo to begin with.

 

[ltcommander.data]

No one enables AF in iPad 2 games because SGX doesn't have AF; place the blame squarely on IMG. Adreno does have it, or at least it did and I assume they didn't drop it. I don't know if anyone else does.

http://developer.apple.com/library/...html#//apple_ref/doc/uid/TP40008793-CH106-SW1

According to their developer guide, Apple supports GL_EXT_texture_filter_anisotropic across both OpenGL ES 1.1 and 2.0 and for both the MBX and SGX. If there is no hardware support is Apple handling this through emulation? Or is just supporting the GL_EXT_texture_filter_anisotropic extension not sufficient to implement an anisotropic filtering solution?

The guide hasn't been updated for the iPad 2 and the SGX543MP2, but it would seem strange that the SGX543MP2 wouldn't support AF if Apple supports it on the SGX535.

 

[Exophase]

I'm trying to get good information on it. According to a post from Xmas there's support for anisotropic filtering up to 2x on MBX:

http://forum.unity3d.com/threads/29055-Anisotropic-filtering

As this was something that was only later enabled on SGX iOS devices it may be emulated here with derivatives in shaders. At any rate, if it's with hardware or emulated we don't know the performance impact. TI actually removed the feature from an OMAP35xx manual listing. But I cautiously withdraw my claim that there's no AF at all, for now. 2x AF is still kind of weak, but better than nothing.