iPad 3 - SGX 543MP2 or 600 series?
Just wondering what the consensus on the rumor mill is for what will be in the iPad 3? I've heard the former more often, but I'm really hoping for the latter. Look at this.
http://www.itproportal.com/2011/02/1...gpx-fill-rate/
That's getting in the range of laptop cards like the Radeon 55** mobile. It depends on if developers take advantage of it of course, but the potential is there, the cards in the 360 and PS3 are only pushing around 200Gflops too, so maybe they weren't so crazy saying console quality by 2013. Depends on the processor too, but just from a GPU standpoint we're nearly there. The SGX 543MP2? 12 Gflops, so 20x the performance is about right for the 600 series.
So which one of these chips do you think will be in the iPad 3? Anand from Anandtech seems to think 543MP4 for various reasons, but it won't be the first time Apple gets new technology first, and they are a 10% shareholder in PowerVR.
Another thing is that if its the MP4 variant of the same core type, that's "only" double the performnace for what's basically confirmed as 4x the pixels, so it would be like the 3gs-4 where GPU performance often went down because of the new res. That would be unfortunate but bearable, but I still really hope its the 600 series.
Quote:
The Rogue GPU will deliver more than 210GFlops (no details as if single or double precision), deliver 350 million real polygons per second and more than five gigapixels per second visible fill rate, which translates into 13 gigapixels of effective fill rate.
This is because, unlike other rival technologies, PowerVR only processes pixels that will be displayed, leaving out hidden ones.
The Rogue GPU will deliver more than 210GFlops (no details as if single or double precision), deliver 350 million real polygons per second and more than five gigapixels per second visible fill rate, which translates into 13 gigapixels of effective fill rate.
This is because, unlike other rival technologies, PowerVR only processes pixels that will be displayed, leaving out hidden ones.
http://www.itproportal.com/2011/02/1...gpx-fill-rate/
That's getting in the range of laptop cards like the Radeon 55** mobile. It depends on if developers take advantage of it of course, but the potential is there, the cards in the 360 and PS3 are only pushing around 200Gflops too, so maybe they weren't so crazy saying console quality by 2013. Depends on the processor too, but just from a GPU standpoint we're nearly there. The SGX 543MP2? 12 Gflops, so 20x the performance is about right for the 600 series.
So which one of these chips do you think will be in the iPad 3? Anand from Anandtech seems to think 543MP4 for various reasons, but it won't be the first time Apple gets new technology first, and they are a 10% shareholder in PowerVR.
Another thing is that if its the MP4 variant of the same core type, that's "only" double the performnace for what's basically confirmed as 4x the pixels, so it would be like the 3gs-4 where GPU performance often went down because of the new res. That would be unfortunate but bearable, but I still really hope its the 600 series.
Comments
For me, I'd like to see Apple put the boot into Sony.
The Rogue sounds good. As good as a PS3..? That would be cool.
Lemon Bob Bon.
What they could or will contain is an open question. At this point I suspect Apple will focus on the retina displays performance and do so with a far more powerful GPU. With only two weeks to go I'm hopeful for a major update.
The Vita uses the SGX 543MP4, the quad core version of the dual core GPU in the iPad 2. It also has to push far less pixels than the iPad 3 would though so it would get better performance for the same power, plus we should know by now not to underestimate dedicated gaming hardware (look what the old x1900-like chip in the 360 or the 7800 in the PS3 can do), they no doubt have sophisticated API's and closer control of the hardware for developers.
From what I've seen of pictures of the PSP Vita, a quad 543MP4 is pretty badass. I think Apple will use a quad 543MP4, and save the more fancy PowerVR stuff for the next iteration.
I'm predicting the iPad 2X will need this quad GPU for Retina Display in 2D, but it will deliver better 3D graphics than iPad 2 because it will still be running at 1024x768 but be ~upscaled ala Xbox360 Lanczos~ to 2048x1536. Everybody wins.
I'm predicting the iPad 2X will need this quad GPU for Retina Display in 2D, but it will deliver better 3D graphics than iPad 2 because it will still be running at 1024x768 but be ~upscaled ala Xbox360 Lanczos~ to 2048x1536. Everybody wins.
Also a good point, most games on the 4 don't run at native since the GPU doesn't have enough bandwidth or fill rate. So games could benefit from the faster GPU even if there is 4x the pixels. Benchmarks would take a hit if they run at native, but actual games wouldn't since developers wouldn't be able to run at the full rez. I still really want the 600 series though
Also a good point, most games on the 4 don't run at native since the GPU doesn't have enough bandwidth or fill rate. So games could benefit from the faster GPU even if there is 4x the pixels. Benchmarks would take a hit if they run at native, but actual games wouldn't since developers wouldn't be able to run at the full rez. I still really want the 600 series though
If they can't drive the screen at full resolution for 3D I don't see many accepting the platform as a next generation device.
If they can't drive the screen at full resolution for 3D I don't see many accepting the platform as a next generation device.
Its already happened; like I said sophisticated 3D games don't run at native on the iPhone 4. If you look at benchmarks, if they ran at native the 3GS is faster than the 4.
Not running at native is hardly a dealbreaker for joe consumer. Think about it, even desktop graphics cards with over 100x the power consumption of this whole SoC would struggle at the rumored resolution.
If they can't drive the screen at full resolution for 3D I don't see many accepting the platform as a next generation device.
About a week-ish to find out?
We'll soon get to see.
Lemon Bon Bon.
Its already happened; like I said sophisticated 3D games don't run at native on the iPhone 4. If you look at benchmarks, if they ran at native the 3GS is faster than the 4.
While I'm no fan of anandtech I'm not surprised at the numbers there. I do wonder how much Apples drivers are impacting things in the tests. In the end though this is my whole point, if graphics performance regresses iPad 3 will not be seen as a new generation device. Instead it will be seen as a half step in the general direction.
So to state it another way, in order for iPad3 to be well accepted it needs to not regress performance wise at full resolution. It actuall needs to perform better. Is that a tall order? I don't really think so, it will require a state of the art processor though.
Not running at native is hardly a dealbreaker for joe consumer.
I really believe that is a mistake. It will depend upon how noticeable artifacts are, but I expect very noticeable differences between apps running at the old res vs apps running the new res.
Think about it, even desktop graphics cards with over 100x the power consumption of this whole SoC would struggle at the rumored resolution.
You are talking about an entirely different class of performance. I believe Apple can easily get 4x better graphics performance just via the current A5 layout. How, through process shrinks that would allow them to more than double clock rate and cores. Apple could potentially jump two or more process nodes, that would allow doubling of clock rate without significant thermal issues. As I've mentioned before Cortex A9 cores are already running much faster than Apples old tech on the sub 32nm processes. Like wise Intels new ATOMs have their GPUs running extremely fast, while the overall power profile has dropped.
Yes I'm optimistic. The question is am I out in left field or a little more rational. To put it simply I think Apple can maintain performance parity at native resolution if they are will to be aggressive. In any event only a few days left before we are all cheering or sobbing in our Wheaties. Maintaining parity in GPU performance across all those extra pixels would make iPad3 very much a respected upgrade. Going beyond that is a more interesting discussion, I'd be very surprised if they have performance that is significantly faster with native resolution.
About a week-ish to find out?
Yeah I'm excited, but sadly have no budget for such a device. I suspect I will be frustrated!
We'll soon get to see.
Lemon Bon Bon.
After the big roll out I have to wonder how bad the backlog will be. If they can deliver what I think they can, Apple may be out of stock most of the years.
I still doubt actual games will be at native though, I think they would be 1024x768 in HiDPI scaling mode, that way they could squeeze even more performance out of it. Artifacts? Pheh, I'll deal with them for more detailed worlds, and the apps that don't need that much performance can run at native, best of both worlds.
That's another possibility I hadn't mentioned, a 543MP4 but clocked higher than the one in the 2, so not only double the cores but a 50-100% clock increase.
This makes the most sense to me! It would also be the best explanation for the A5X designation. The only potential problem would be memory bandwidth.
So even with 4x the resolution it could perform a bit better than the 2 even at native. The PS Vitas MP4 is clocked up to 400MHz, I think. That would make me happy. But still not as much as the 600 series
My original understanding was that the 543 could go beyond four cores.
I still doubt actual games will be at native though, I think they would be 1024x768 in HiDPI scaling mode, that way they could squeeze even more performance out of it.
Why not native? If you get the same basic performance why not enjoy it.
Artifacts? Pheh, I'll deal with them for more detailed worlds, and the apps that don't need that much performance can run at native, best of both worlds.
Why not native? If you get the same basic performance why not enjoy it.
Same reason most console games are 720p rather than 1080, or why 3D games like Uncharted 3 on PS3 have to reduce the level of detail when the GPU has to render effectively two scenes. Developer ease, and they can do more complicated things on screen at a lower resolution because they have more power to work with, which would otherwise be spent on the extra pixels. If they can run at native, great, but sooner or later they'd hit the point where they could effectively up the power by reducing the resolution to 1024x768 upscaled mode.
Also, if they went with a dual-core version of this GPU, it would end consoles.
BTW: That SoC with this GPU is from ST-Ericsson: From wiki: http://en.wikipedia.org/wiki/List_of...oducts#Series6
This has been "released to licensees" just this Jan 2012. But I read elsewhere that it's not due in products til 2013.
tipoo, I've thinking about the 600 series (Rogue) in iPad 3, too. I'm currently inclined to agree with your reluctant conclusion that they'll settle for x2 cores (MP4) at x2 clock, to handle the x4 pixels. This is because only one license has a SoC for Rogue, but it's expected to be a year til it's available in products. However, Apple might be faster because (i) they can integrate SoC and product development because they are both in-house; (ii) Apple own 9-10% of Imagination Technologies, the GPU designer, so they might get early versions, work closely with them etc. Hope springs eternal...
Also, if they went with a dual-core version of this GPU, it would end consoles.
BTW: That SoC with this GPU is from ST-Ericsson: From wiki: http://en.wikipedia.org/wiki/List_of...oducts#Series6
This has been "released to licensees" just this Jan 2012. But I read elsewhere that it's not due in products til 2013.
[...] Work on the cores would have been started years ago to integrate it into a new SoC. However the GPU core isn't the big issue here as you still need to team it to a suitable CPU and system architecture to support that GPU.
Wouldn't "integration" into the SoC include system architecture to support it? Years in the pipeline seems enough time to do that...
On your first point, you're saying the current dual-core CPU isn't suitable to feed this GPU? Bus bandwidth to memory seems more a system architecture issue, esp. considering that licensees have great leeway to tweak ARM designs, maybe that SoC design firm that Apple bought could handle it...
OTOH... Of course, Apple could also go for a Cortex A15 upgrade (as the Nova above does), which would seem to address that issue. And, if we're thinking that they could get a Rogue into the iPad 3 one year ahead of competitors, why not get A15 a year ahead too? (Apple isn't as cosy with ARM as they are with IT, though they helped found it, so they mightn't get the needed early-access).
It seems highly unrealistic to expect both upgrades (esp if the A5X photo is real); yet it also seem neatest and simplest to have them go together. So... they both wait for iPad 4, and we get a half-hacked (aka tuned) version of the iPad 2 SoC in iPad 3.
Wouldn't "integration" into the SoC include system architecture to support it? Years in the pipeline seems enough time to do that...
Well this would likely be why Apple has a rumored two track development process. That is to have the time to engineer a major new design every two years. So yeah they have to design the SoC to cover the memory bandwidth requirements of the GPU. At the same time that means an all new SoC.
On your first point, you're saying the current dual-core CPU isn't suitable to feed this GPU? Bus bandwidth to memory seems more a system architecture issue,
That is system architecture and is exactly what I'm talking about. Frankly the CPU plays a small part here. The GPU on the other hand may require a new memory bus design, caches or other features. There are many ways for Apple to solve these issues but it does involve a bit of engineering to come up with the new architecture.
esp. considering that licensees have great leeway to tweak ARM designs, maybe that SoC design firm that Apple bought could handle it...
I have no doubt they can handle it. The question is could they have this new architecture ready for launch next week. That is really hard to say, if the debut what could be called A6 next week they would effectively be six to nine months ahead of the industry.
OTOH... Of course, Apple could also go for a Cortex A15 upgrade (as the Nova above does), which would seem to address that issue.
Cortex A15 doesn't have a lot to offer Apple, I believe the GPU would be the imperative. Let's face it with the right process shrink Apple could manage a 2-4X performance jump with the current Cortex A9's.
And, if we're thinking that they could get a Rogue into the iPad 3 one year ahead of competitors, why not get A15 a year ahead too? (Apple isn't as cosy with ARM as they are with IT, though they helped found it, so they mightn't get the needed early-access).
I suspect it will be more like six months or so. The problem is the new processor is a big initiative, with all of the associated risks, so why not focus on the absolute must haves. That would be the GPU and system architecture.
It seems highly unrealistic to expect both upgrades (esp if the A5X photo is real); yet it also seem neatest and simplest to have them go together. So... they both wait for iPad 4, and we get a half-hacked (aka tuned) version of the iPad 2 SoC in iPad 3.
Half-hacked? Wow I bet a few Apple engineers just lined up to tell you a thing or two. A SoC fast enough to really drive that high resolution screen well is a major engineering effort even if it derives from the old. In any event it will be most interesting to see what Apple delivers.
Also, if they went with a dual-core version of this GPU, it would end consoles.
Even todays consoles have different advantages though, bandwidth is a big problem on these smartphone SoCs as well as storage, not to mention controls. And todays desktop GPUs that would be in the next gen consoles have over 10x the raw Gflops of the current ones ~200. The new Radeon 7870 pushes over 3 terraflops.
I think we're still 10-12 months early for Cortex A15, so my best hope would be a faster clocked quad A9. As mentioned Apple isn't as cozy with ARM as with Imagination Technologies but both products are about a year out for most companies, so I hope for at least one or the other, the 600 series would put them incredibly far ahead of Android tablets. They already have a lead, but this would go from big to enormous.
Edit: of interest?
http://www.theverge.com/2012/3/1/283...mment-on-apple
if the debut what could be called A6 next week they would effectively be six to nine months ahead of the industry.
I think that sums it up: can Apple be so far ahead of the industry, not just in design, concept etc, but even in designing the chips themselves? I gave some factors above supporting this; but their past performance is a guide: they had a new SoC for iPad 1, and a year later, another new SoC for iPad 2.
Now, that's just two data points, and there could be other factors (e.g. maybe both chips were in the pipeline for years). Crucially, and the point, I haven't compared that with the industry - maybe everyone managed that too.
Cortex A15 doesn't have a lot to offer Apple
The A15 seems a significant upgrade, at least doubling the performance without doubling power consumption. ARM know what they are doing. Can you expand on why you think that?
I do agree that a CPU upgrade isn't crucial for the iPad (for this generation); but GPU absolutely is. This is true for the x4 pixels; but also in general.
The problem is the new processor is a big initiative, with all of the associated risks, so why not focus on the absolute must haves. That would be the GPU and system architecture.
Agreed. And the "A5X", and dual-track rumour etc all support this. Plus, just a retina display is enough to wow everyone. I was musing that it might be workable if the risks could be self-contained within the SoC group (and it has the massive resources of Apple behind it) - but risks like "delays" can't be self-contained.
Half-hacked? Wow I bet a few Apple engineers just lined up to tell you a thing or two. A SoC fast enough to really drive that high resolution screen well is a major engineering effort even if it derives from the old.
Hey, it wasn't meant as an insult. I was speaking in sympathy with them - to modify an existing design, when there's a new architecture available that solves all the problems you're facing... that was be frustrating to me. As a developer, I find that at some point, it's easier to start fresh than modify the existing design. Anyway, that's how I'd experience it - maybe they have a different attitude; and/or the technical issues are interesting in themselves and can inform and be used in the next architecture. Maybe I should keep my mouth shut instead of offending people
Wouldn't it be amazing if Apple was ahead of everybody in every aspect? I think it's possible but unlikely; but most of all, it's not needed for this generation of the iPad. And think that's one of the secrets of Apple: they *have* the technical chops, but they only use them in service of a user outcome, instead of an end in themselves. That's tricky. Most companies seem go one way or the other (i.e. all technical; or all user outcome).
Even todays consoles have different advantages though, bandwidth is a big problem on these smartphone SoCs as well as storage, not to mention controls.
Thanks, great points. HD game textures need massive textures. Could put them in the cloud but... bandwidth. I was thinking that mystery extra component (part number) could be a game controller.
And todays desktop GPUs that would be in the next gen consoles have over 10x the raw Gflops of the current ones ~200. The new Radeon 7870 pushes over 3 terraflops.
Yes, consoles are long overdue for an upgrade, but they've been making so much money they've grown complacent... which creates an opportunity for a new entrant. Also, there's a shift towards casual games, where iPhone/iPad are strong.
Alternatives: Apple could include console features in their TV, circumventing the storage and bandwidth issues.
I think we're still 10-12 months early for Cortex A15, so my best hope would be a faster clocked quad A9.
I think a quad-core GPU, but not a quad-core CPU. Because (1) Doubling CPU doesn't double performance, but doubling GPU does. So quad-core GPU is a more efficient use of silicon and power consumption. (2) CPU performance isn't needed as much as GPU performance on tablets in general, and especially for the iPad 3's retina display.
http://www.theverge.com/2012/3/1/283...mment-on-apple
Why be terse if there's nothing to say - it sort of screams that they're working closely, doesn't it?