As others have said, ARM SoCs can't improve 2x every year forever. ARM has run out of the low hanging fruit with the latest generation of CPUs where 3-issue out-of-order execution architectures were introduced with Krait, Swift and A15 cores. For a period of 4 or 5 years or so, ARM went from simple 1-issue, in-order, non-superscalar architectures to modern techniques (superscalar, wider issue widths, branch prediction, out-of-order, SMP) on a yearly cycle. They benefitted from all the hardwork in the PC, workstation, and server space and was able to introduce these techniques over a 4 or 5 year time frame instead of the 2 decades it originally took.
Arguably, this was because they needed to fit in a 1 to 4 W TDP envelope and they needed to wait 65 nm nodes or better to even do it at those power consumption levels.
1. The only metric that is really important to most everyone is single threaded (single core) performance and the whole memory train supporting it. A 30% performance improvement over a years time frame is actually quite normal, if better than it should. Maybe Apple can squeeze out another 100% improvement, but it's getting harder and harder to do.
2. 4-cores isn't important to most everyone. Most software can't actually take advantage of that many cores. Software that can take advantage of it, anything involving matrix math or easily split apart tasks, are few, and most aren't fit for mobile handhelds. Transcoding video is not something done on your phone. So, dual-core is typically the optimum number of cores needed for 90% of the folks. What's really really important is single-threaded performance. Maybe in a couple of years, video transcoding and editing will be common enough or maybe there be some killer app that can make use of it, but definitely not today.
3. When talking about 64-bit, what people are really talking about is the instruction set architecture and addressable memory. It does not refer to data widths, like in a 64-bit memory bus or 128-bit SIMD/vector instruction. It doesn't mean it'll be automatically faster. Typically, what it really means is greater than 4 GB of RAM support.
What does imagination have to do with anything? What makes you think I have none? Who here knows of what Apple will release next? Your comment makes no sense whatsoever...
You said:
I honestly see no reason why Apple shouldn't include a Quad Core A7 SoC inside the next iPhone/iPad.
If you can't see any possible reason why Apple shouldn't include a quad core A7, then you obviously don't have any imagination. I gave 4 possible reasons just off the top of my head.
If you had said "I think they'll do xyz", that would be a different matter. Instead, you implied that there couldn't be any plausible reasons for them not to - which means that you're incapable of considering options different than your own preferred outcome.
If you can't see any possible reason why Apple shouldn't include a quad core A7, then you obviously don't have any imagination.
All I said was that a quad core chip seems like the logical next step to me.
Quote:
I gave 4 possible reasons just off the top of my head.
Is it mandatory to agree with you? I respect your opinion but I simply think different.
Quote:
If you had said "I think they'll do xyz", that would be a different matter. Instead, you implied that there couldn't be any plausible reasons for them not to - which means that you're incapable of considering options different than your own preferred outcome.
It is evident that you completely misread my comment.
The is interesting part is the 64 bit iOS running on the A7.
This tells me that Apple has plans for much more capable iOS devices needing to address vast amount of RAM and process huge amount of data.
The existing iOS devices don't really need a 64 bit OS and we all know Apple did turn on the 64 bit switch for nothing.
Time will tell.
I have to disagree! IPads are currently starved for RAM, Saferi leaves a lot to be desired because of the lack of RAM. Admittedly the app could use better memory management but safari is at a huge disadvantage on iOS due to the constant page reloading.
I have to disagree! IPads are currently starved for RAM, Saferi leaves a lot to be desired because of the lack of RAM. Admittedly the app could use better memory management but safari is at a huge disadvantage on iOS due to the constant page reloading.
I agree. More RAM would do more good than greater clock speed or quad core. And it would certainly be more beneficial than 64 bit.
We won't see 5-9X GPU anymore - that was simply because the A4 had a terrible GPU compared to the A5. But another 2X is not unreasonable if they switch to the 600 Series PowerVR.
They cut the iPad 4 out of the graphics test in the video, likely because the iPad 4 already scores higher than the G6200 in the tests. There's a nice list here of desktop and mobile devices together:
That shows all the scores on one page for Egypt offscreen. The iPad 4 is still pretty high up as it uses the 128 ALU PowerVR SGX554 MP4. To top that, they'd probably have to go to the 96 ALU G6630 shown on the roadmap:
Looking at the gfx chart, this would put it above the Radeon 6630M in the Mac Mini. The 6630M is just above the PS3 and XBox 360 - games like LA Noire and Mafia 2 push the limit of the consoles but are smoother on the Mini. The graphics support in the PowerVR 6 goes up to OpenGL ES 3 too so more advanced geometry, shading and lighting effects:
[VIDEO]
Apple could double the CPU cores too, which helps things like media exports but it wouldn't be essential IMO. I think the graphics jump is the important one because combined with the controllers, this means that people can dock their iOS devices to their TVs and get last-gen console-quality graphics.
If they switch to IGZO displays, that would also help battery life.
Comments
As others have said, ARM SoCs can't improve 2x every year forever. ARM has run out of the low hanging fruit with the latest generation of CPUs where 3-issue out-of-order execution architectures were introduced with Krait, Swift and A15 cores. For a period of 4 or 5 years or so, ARM went from simple 1-issue, in-order, non-superscalar architectures to modern techniques (superscalar, wider issue widths, branch prediction, out-of-order, SMP) on a yearly cycle. They benefitted from all the hardwork in the PC, workstation, and server space and was able to introduce these techniques over a 4 or 5 year time frame instead of the 2 decades it originally took.
Arguably, this was because they needed to fit in a 1 to 4 W TDP envelope and they needed to wait 65 nm nodes or better to even do it at those power consumption levels.
1. The only metric that is really important to most everyone is single threaded (single core) performance and the whole memory train supporting it. A 30% performance improvement over a years time frame is actually quite normal, if better than it should. Maybe Apple can squeeze out another 100% improvement, but it's getting harder and harder to do.
2. 4-cores isn't important to most everyone. Most software can't actually take advantage of that many cores. Software that can take advantage of it, anything involving matrix math or easily split apart tasks, are few, and most aren't fit for mobile handhelds. Transcoding video is not something done on your phone. So, dual-core is typically the optimum number of cores needed for 90% of the folks. What's really really important is single-threaded performance. Maybe in a couple of years, video transcoding and editing will be common enough or maybe there be some killer app that can make use of it, but definitely not today.
3. When talking about 64-bit, what people are really talking about is the instruction set architecture and addressable memory. It does not refer to data widths, like in a 64-bit memory bus or 128-bit SIMD/vector instruction. It doesn't mean it'll be automatically faster. Typically, what it really means is greater than 4 GB of RAM support.
You said:
If you can't see any possible reason why Apple shouldn't include a quad core A7, then you obviously don't have any imagination. I gave 4 possible reasons just off the top of my head.
If you had said "I think they'll do xyz", that would be a different matter. Instead, you implied that there couldn't be any plausible reasons for them not to - which means that you're incapable of considering options different than your own preferred outcome.
Quote:
Originally Posted by jragosta
You said:
If you can't see any possible reason why Apple shouldn't include a quad core A7, then you obviously don't have any imagination.
All I said was that a quad core chip seems like the logical next step to me.
Quote:
I gave 4 possible reasons just off the top of my head.
Is it mandatory to agree with you? I respect your opinion but I simply think different.
Quote:
If you had said "I think they'll do xyz", that would be a different matter. Instead, you implied that there couldn't be any plausible reasons for them not to - which means that you're incapable of considering options different than your own preferred outcome.
It is evident that you completely misread my comment.
What you said was "I honestly see no reason why Apple shouldn't include a Quad Core A7 SoC inside the next iPhone/iPad."
I'm not the one who misread it.
Quote:
Originally Posted by jragosta
What you said was "I honestly see no reason why Apple shouldn't include a Quad Core A7 SoC inside the next iPhone/iPad."
I'm not the one who misread it.
Yep! I see no reason why they shouldn't. What's wrong with that?
I have to disagree! IPads are currently starved for RAM, Saferi leaves a lot to be desired because of the lack of RAM. Admittedly the app could use better memory management but safari is at a huge disadvantage on iOS due to the constant page reloading.
I agree. More RAM would do more good than greater clock speed or quad core. And it would certainly be more beneficial than 64 bit.
There's a test of the PowerVR G6200 here:
[VIDEO]
http://www.engadget.com/2013/07/29/mediatek-mt8135-biglittle-mp-powervr-series6-g6200/
They cut the iPad 4 out of the graphics test in the video, likely because the iPad 4 already scores higher than the G6200 in the tests. There's a nice list here of desktop and mobile devices together:
http://gfxbench.com/result.jsp?benchmark=gfx27&data-source=1&version=all&test=484&text-filter=&order=median&os-Android=true&os-iPhone_OS=true&os-Windows_8=true&os-Windows_Phone=true&os-Windows_RT=true&arch-ARM=true&arch-MIPS=true&arch-x86=true&base=device
That shows all the scores on one page for Egypt offscreen. The iPad 4 is still pretty high up as it uses the 128 ALU PowerVR SGX554 MP4. To top that, they'd probably have to go to the 96 ALU G6630 shown on the roadmap:
http://pc.watch.impress.co.jp/video/pcw/docs/578/851/p01.pdf
Given that the G6200 is 32 ALU, the G6630 could be 3x the speed of the current iPad 4:
http://withimagination.imgtec.com/index.php/powervr/powervr-g6630-go-fast-or-go-home
Looking at the gfx chart, this would put it above the Radeon 6630M in the Mac Mini. The 6630M is just above the PS3 and XBox 360 - games like LA Noire and Mafia 2 push the limit of the consoles but are smoother on the Mini. The graphics support in the PowerVR 6 goes up to OpenGL ES 3 too so more advanced geometry, shading and lighting effects:
[VIDEO]
Apple could double the CPU cores too, which helps things like media exports but it wouldn't be essential IMO. I think the graphics jump is the important one because combined with the controllers, this means that people can dock their iOS devices to their TVs and get last-gen console-quality graphics.
If they switch to IGZO displays, that would also help battery life.